html_url
stringlengths 48
51
| title
stringlengths 1
290
| comments
sequencelengths 0
30
| body
stringlengths 0
228k
β | number
int64 2
7.08k
|
|---|---|---|---|---|
https://github.com/huggingface/datasets/issues/798 | Cannot load TREC dataset: ConnectionError | [
"Hi ! Indeed there's an issue with those links.\r\nWe should probably use the target urls of the redirections instead",
"Hi, the same issue here, could you tell me how to download it through datasets? thanks ",
"Same issue. ",
"Actually it's already fixed on the master branch since #740 \r\nI'll do the 1.1.3 release soon",
"Hi\nthanks, but I did tried to install from the pip install git+... and it does\nnot work for me,. thanks for the help. I have the same issue with wmt16,\n\"ro-en\"\nthanks.\nBest\nRabeeh\n\nOn Mon, Nov 16, 2020 at 10:29 AM Quentin Lhoest <[email protected]>\nwrote:\n\n> Actually it's already fixed on the master branch since #740\n> <https://github.com/huggingface/datasets/pull/740>\n> I'll do the 1.1.3 release soon\n>\n> β\n> You are receiving this because you commented.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/datasets/issues/798#issuecomment-727854736>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/ABP4ZCEUBJKPOCLABXCKMPDSQDWH3ANCNFSM4TJBUKSA>\n> .\n>\n",
"I just tested on google colab using\r\n```python\r\n!pip install git+https://github.com/huggingface/datasets.git\r\nfrom datasets import load_dataset\r\nload_dataset(\"trec\")\r\n```\r\nand it works.\r\nCan you detail how you got the issue even when using the latest version on master ?\r\n\r\nAlso about wmt we'll look into it, thanks for reporting !",
"I think the new url with .edu is also broken:\r\n```\r\nConnectionError: Couldn't reach https://cogcomp.seas.upenn.edu/Data/QA/QC/train_5500.label\r\n```\r\nCant download the dataset anymore.",
"Hi ! The URL seems to work fine on my side, can you try again ?",
"Forgot to update, i wrote an email to the webmaster of seas.upenn.edu because i couldnt reach the url on any machine. This was the answer:\r\n```\r\nThank you for your report. The server was offline for maintenance and is now available again.\r\n```\r\nGuess all back to normal now π "
] | ## Problem
I cannot load "trec" dataset, it results with ConnectionError as shown below. I've tried on both Google Colab and locally.
* `requests.head('http://cogcomp.org/Data/QA/QC/train_5500.label')` returns <Response [302]>.
* `requests.head('http://cogcomp.org/Data/QA/QC/train_5500.label', allow_redirects=True)` raises `requests.exceptions.TooManyRedirects: Exceeded 30 redirects.`
* Opening `http://cogcomp.org/Data/QA/QC/train_5500.label' in a browser works, but opens a different address
* Increasing max_redirects to 100 doesn't help
Also, while debugging I've seen that requesting 'https://storage.googleapis.com/huggingface-nlp/cache/datasets/trec/default/1.1.0/dataset_info.json' returns <Response [404]> before, but it doesn't raise any errors. Not sure if that's relevant.
* datasets.__version__ == '1.1.2'
* requests.__version__ == '2.24.0'
## Error trace
```
>>> import datasets
>>> datasets.__version__
'1.1.2'
>>> dataset = load_dataset("trec", split="train")
Using custom data configuration default
Downloading and preparing dataset trec/default (download: 350.79 KiB, generated: 403.39 KiB, post-processed: Unknown size, total: 754.18 KiB) to /home/przemyslaw/.cache/huggingface/datasets/trec/default/1.1.0/ca4248481ad244f235f4cf277186cad2ee8769f975119a2bbfc41b8932b88bd7...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/load.py", line 611, in load_dataset
ignore_verifications=ignore_verifications,
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/builder.py", line 476, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/builder.py", line 531, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/home/przemyslaw/.cache/huggingface/modules/datasets_modules/datasets/trec/ca4248481ad244f235f4cf277186cad2ee8769f975119a2bbfc41b8932b88bd7/trec.py", line 140, in _split_generators
dl_files = dl_manager.download_and_extract(_URLs)
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/download_manager.py", line 254, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/download_manager.py", line 179, in download
num_proc=download_config.num_proc,
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/py_utils.py", line 225, in map_nested
_single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/py_utils.py", line 225, in <listcomp>
_single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/py_utils.py", line 163, in _single_map_nested
return function(data_struct)
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/file_utils.py", line 308, in cached_path
use_etag=download_config.use_etag,
File "/home/przemyslaw/.local/lib/python3.6/site-packages/datasets/utils/file_utils.py", line 475, in get_from_cache
raise ConnectionError("Couldn't reach {}".format(url))
ConnectionError: Couldn't reach http://cogcomp.org/Data/QA/QC/train_5500.label
```
I would appreciate some suggestions here. | 798 |
https://github.com/huggingface/datasets/issues/797 | Token classification labels are strings and we don't have the list of labels | [
"Indeed. Pinging @stefan-it here if he want to give an expert opinion :)",
"Related is https://github.com/huggingface/datasets/pull/636",
"Should definitely be a ClassLabel π ",
"Already done."
] | Not sure if this is an issue we want to fix or not, putting it here so it's not forgotten. Right now, in token classification datasets, the labels for NER, POS and the likes are typed as `Sequence` of `strings`, which is wrong in my opinion. These should be `Sequence` of `ClassLabel` or some types that gives easy access to the underlying labels.
The main problem for preprocessing those datasets is that the list of possible labels is not stored inside the `Dataset` object which makes converting the labels to IDs quite difficult (you either have to know the list of labels in advance or run a full pass through the dataset to get the list of labels, the `unique` method being useless with the type `Sequence[str]`). | 797 |
https://github.com/huggingface/datasets/issues/795 | Descriptions of raw and processed versions of wikitext are inverted | [
"Yes indeed ! Thanks for reporting",
"Fixed by:\r\n- #3241"
] | Nothing of importance, but it looks like the descriptions of wikitext-n-v1 and wikitext-n-raw-v1 are inverted for both n=2 and n=103. I just verified by loading them and the `<unk>` tokens are present in the non-raw versions, which confirms that it's a mere inversion of the descriptions and not of the datasets themselves.
Also it would be nice if those descriptions appeared in the dataset explorer.
https://github.com/huggingface/datasets/blob/87bd0864845ea0a1dd7167918dc5f341bf807bd3/datasets/wikitext/wikitext.py#L52 | 795 |
https://github.com/huggingface/datasets/issues/794 | self.options cannot be converted to a Python object for pickling | [
"Hi ! Thanks for reporting that's a bug on master indeed.\r\nWe'll fix that soon"
] | Hi,
Currently I am trying to load csv file with customized read_options. And the latest master seems broken if we pass the ReadOptions object.
Here is a code snippet
```python
from datasets import load_dataset
from pyarrow.csv import ReadOptions
load_dataset("csv", data_files=["out.csv"], read_options=ReadOptions(block_size=16*1024*1024))
```
error is `self.options cannot be converted to a Python object for pickling`
Would you mind to take a look? Thanks!
```
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-28-ab83fec2ded4> in <module>
----> 1 load_dataset("csv", data_files=["out.csv"], read_options=ReadOptions(block_size=16*1024*1024))
/tmp/datasets/src/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)
602 hash=hash,
603 features=features,
--> 604 **config_kwargs,
605 )
606
/tmp/datasets/src/datasets/builder.py in __init__(self, cache_dir, name, hash, features, **config_kwargs)
162 name,
163 custom_features=features,
--> 164 **config_kwargs,
165 )
166
/tmp/datasets/src/datasets/builder.py in _create_builder_config(self, name, custom_features, **config_kwargs)
281 )
282 else:
--> 283 suffix = Hasher.hash(config_kwargs_to_add_to_suffix)
284
285 if builder_config.data_files is not None:
/tmp/datasets/src/datasets/fingerprint.py in hash(cls, value)
51 return cls.dispatch[type(value)](cls, value)
52 else:
---> 53 return cls.hash_default(value)
54
55 def update(self, value):
/tmp/datasets/src/datasets/fingerprint.py in hash_default(cls, value)
44 @classmethod
45 def hash_default(cls, value):
---> 46 return cls.hash_bytes(dumps(value))
47
48 @classmethod
/tmp/datasets/src/datasets/utils/py_utils.py in dumps(obj)
365 file = StringIO()
366 with _no_cache_fields(obj):
--> 367 dump(obj, file)
368 return file.getvalue()
369
/tmp/datasets/src/datasets/utils/py_utils.py in dump(obj, file)
337 def dump(obj, file):
338 """pickle an object to a file"""
--> 339 Pickler(file, recurse=True).dump(obj)
340 return
341
~/.local/lib/python3.6/site-packages/dill/_dill.py in dump(self, obj)
444 raise PicklingError(msg)
445 else:
--> 446 StockPickler.dump(self, obj)
447 stack.clear() # clear record of 'recursion-sensitive' pickled objects
448 return
/usr/lib/python3.6/pickle.py in dump(self, obj)
407 if self.proto >= 4:
408 self.framer.start_framing()
--> 409 self.save(obj)
410 self.write(STOP)
411 self.framer.end_framing()
/usr/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)
474 f = self.dispatch.get(t)
475 if f is not None:
--> 476 f(self, obj) # Call unbound method with explicit self
477 return
478
~/.local/lib/python3.6/site-packages/dill/_dill.py in save_module_dict(pickler, obj)
931 # we only care about session the first pass thru
932 pickler._session = False
--> 933 StockPickler.save_dict(pickler, obj)
934 log.info("# D2")
935 return
/usr/lib/python3.6/pickle.py in save_dict(self, obj)
819
820 self.memoize(obj)
--> 821 self._batch_setitems(obj.items())
822
823 dispatch[dict] = save_dict
/usr/lib/python3.6/pickle.py in _batch_setitems(self, items)
850 k, v = tmp[0]
851 save(k)
--> 852 save(v)
853 write(SETITEM)
854 # else tmp is empty, and we're done
/usr/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)
494 reduce = getattr(obj, "__reduce_ex__", None)
495 if reduce is not None:
--> 496 rv = reduce(self.proto)
497 else:
498 reduce = getattr(obj, "__reduce__", None)
~/.local/lib/python3.6/site-packages/pyarrow/_csv.cpython-36m-x86_64-linux-gnu.so in pyarrow._csv.ReadOptions.__reduce_cython__()
TypeError: self.options cannot be converted to a Python object for pickling
``` | 794 |
https://github.com/huggingface/datasets/issues/792 | KILT dataset: empty string in triviaqa input field | [
"Just found out about https://github.com/huggingface/datasets/blob/master/datasets/kilt_tasks/README.md\r\n(Not very clear in https://huggingface.co/datasets/kilt_tasks links to http://github.com/huggingface/datasets/datasets/kilt_tasks/README.md which is dead, closing the issue though :))"
] | # What happened
Both train and test splits of the triviaqa dataset (part of the KILT benchmark) seem to have empty string in their input field (unlike the natural questions dataset, part of the same benchmark)
# Versions
KILT version is `1.0.0`
`datasets` version is `1.1.2`
[more here](https://gist.github.com/PaulLerner/3768c8d25f723edbac20d99b6a4056c1)
# How to reproduce
```py
In [1]: from datasets import load_dataset
In [4]: dataset = load_dataset("kilt_tasks")
# everything works fine, removed output for a better readibility
Dataset kilt_tasks downloaded and prepared to /people/lerner/.cache/huggingface/datasets/kilt_tasks/all_tasks/1.0.0/821c4295a2c35db2847585918d9c47d7f028f1a26b78825d8e77cd3aeb2621a1. Subsequent calls will reuse this data.
# empty string in triviaqa input field
In [36]: dataset['train_triviaqa'][0]
Out[36]:
{'id': 'dpql_5197',
'input': '',
'meta': {'left_context': '',
'mention': '',
'obj_surface': {'text': []},
'partial_evidence': {'end_paragraph_id': [],
'meta': [],
'section': [],
'start_paragraph_id': [],
'title': [],
'wikipedia_id': []},
'right_context': '',
'sub_surface': {'text': []},
'subj_aliases': {'text': []},
'template_questions': {'text': []}},
'output': {'answer': ['five Β£', '5 Β£', 'Β£5', 'five Β£'],
'meta': [],
'provenance': [{'bleu_score': [1.0],
'end_character': [248],
'end_paragraph_id': [30],
'meta': [],
'section': ['Section::::Question of legal tender.\n'],
'start_character': [246],
'start_paragraph_id': [30],
'title': ['Banknotes of the pound sterling'],
'wikipedia_id': ['270680']}]}}
In [35]: dataset['train_triviaqa']['input'][:10]
Out[35]: ['', '', '', '', '', '', '', '', '', '']
# same with test set
In [37]: dataset['test_triviaqa']['input'][:10]
Out[37]: ['', '', '', '', '', '', '', '', '', '']
# works fine with natural questions
In [34]: dataset['train_nq']['input'][:10]
Out[34]:
['how i.met your mother who is the mother',
'who had the most wins in the nfl',
'who played mantis guardians of the galaxy 2',
'what channel is the premier league on in france',
"god's not dead a light in the darkness release date",
'who is the current president of un general assembly',
'when do the eclipse supposed to take place',
'what is the name of the sea surrounding dubai',
'who holds the nba record for most points in a career',
'when did the new maze runner movie come out']
```
Stay safe :) | 792 |
https://github.com/huggingface/datasets/issues/790 | Error running pip install -e ".[dev]" on MacOS 10.13.6: faiss/python does not exist | [
"I saw that `faiss-cpu` 1.6.4.post2 was released recently to fix the installation on macos. It should work now",
"Closing this one.\r\nFeel free to re-open if you still have issues"
] | I was following along with https://huggingface.co/docs/datasets/share_dataset.html#adding-tests-and-metadata-to-the-dataset when I ran into this error.
```sh
git clone https://github.com/huggingface/datasets
cd datasets
virtualenv venv -p python3 --system-site-packages
source venv/bin/activate
pip install -e ".[dev]"
```
Python 3.7.7
| 790 |
https://github.com/huggingface/datasets/issues/788 | failed to reuse cache | [] | I packed the `load_dataset ` in a function of class, and cached data in a directory. But when I import the class and use the function, the data still have to be downloaded again. The information (Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.28 GiB, post-processed: Unknown size, total: 1.82 GiB) to ******) which logged to terminal shows the path is right to the cache directory, but the files still have to be downloaded again. | 788 |
https://github.com/huggingface/datasets/issues/786 | feat(dataset): multiprocessing _generate_examples | [
"I agree that would be cool :)\r\nRight now the only distributed dataset builder is based on Apache Beam so you can use distributed processing frameworks like Dataflow, Spark, Flink etc. to build your dataset but it's not really well suited for single-worker parallel processing afaik",
"`_generate_examples` can now be run in parallel thanks to https://github.com/huggingface/datasets/pull/5107. You can find more info [here](https://huggingface.co/docs/datasets/dataset_script#sharding)."
] | forking this out of #741, this issue is only regarding multiprocessing
I'd love if there was a dataset configuration parameter `workers`, where when it is `1` it behaves as it does right now, and when its `>1` maybe `_generate_examples` can also get the `pool` and return an iterable using the pool.
In my use case, I would instead of:
```python
for datum in data:
yield self.load_datum(datum)
```
do:
```python
return pool.map(self.load_datum, data)
```
As the dataset in question, as an example, has **only** 7000 rows, and takes 10 seconds to load each row on average, it takes almost 20 hours to load the entire dataset.
If this was a larger dataset (and many such datasets exist), it would take multiple days to complete.
Using multiprocessing, for example, 40 cores, could speed it up dramatically. For this dataset, hopefully to fully load in under an hour. | 786 |
https://github.com/huggingface/datasets/issues/784 | Issue with downloading Wikipedia data for low resource language | [
"Hello, maybe you could ty to use another date for the wikipedia dump (see the available [dates](https://dumps.wikimedia.org/jvwiki) here for `jv`) ?",
"@lhoestq\r\n\r\nI've tried `load_dataset('wikipedia', '20200501.zh', beam_runner='DirectRunner')` and got the same `FileNotFoundError` as @SamuelCahyawijaya.\r\n\r\nAlso, using another date (e.g. `load_dataset('wikipedia', '20201120.zh', beam_runner='DirectRunner')`) will give the following error message.\r\n\r\n```\r\nValueError: BuilderConfig 20201120.zh not found. Available: ['20200501.aa', '20200501.ab', '20200501.ace', '20200501.ady', '20200501.af', '20200501.ak', '20200501.als', '20200501.am', '20200501.an', '20200501.ang', '20200501.ar', '20200501.arc', '20200501.arz', '20200501.as', '20200501.ast', '20200501.atj', '20200501.av', '20200501.ay', '20200501.az', '20200501.azb', '20200501.ba', '20200501.bar', '20200501.bat-smg', '20200501.bcl', '20200501.be', '20200501.be-x-old', '20200501.bg', '20200501.bh', '20200501.bi', '20200501.bjn', '20200501.bm', '20200501.bn', '20200501.bo', '20200501.bpy', '20200501.br', '20200501.bs', '20200501.bug', '20200501.bxr', '20200501.ca', '20200501.cbk-zam', '20200501.cdo', '20200501.ce', '20200501.ceb', '20200501.ch', '20200501.cho', '20200501.chr', '20200501.chy', '20200501.ckb', '20200501.co', '20200501.cr', '20200501.crh', '20200501.cs', '20200501.csb', '20200501.cu', '20200501.cv', '20200501.cy', '20200501.da', '20200501.de', '20200501.din', '20200501.diq', '20200501.dsb', '20200501.dty', '20200501.dv', '20200501.dz', '20200501.ee', '20200501.el', '20200501.eml', '20200501.en', '20200501.eo', '20200501.es', '20200501.et', '20200501.eu', '20200501.ext', '20200501.fa', '20200501.ff', '20200501.fi', '20200501.fiu-vro', '20200501.fj', '20200501.fo', '20200501.fr', '20200501.frp', '20200501.frr', '20200501.fur', '20200501.fy', '20200501.ga', '20200501.gag', '20200501.gan', '20200501.gd', '20200501.gl', '20200501.glk', '20200501.gn', '20200501.gom', '20200501.gor', '20200501.got', '20200501.gu', '20200501.gv', '20200501.ha', '20200501.hak', '20200501.haw', '20200501.he', '20200501.hi', '20200501.hif', '20200501.ho', '20200501.hr', '20200501.hsb', '20200501.ht', '20200501.hu', '20200501.hy', '20200501.ia', '20200501.id', '20200501.ie', '20200501.ig', '20200501.ii', '20200501.ik', '20200501.ilo', '20200501.inh', '20200501.io', '20200501.is', '20200501.it', '20200501.iu', '20200501.ja', '20200501.jam', '20200501.jbo', '20200501.jv', '20200501.ka', '20200501.kaa', '20200501.kab', '20200501.kbd', '20200501.kbp', '20200501.kg', '20200501.ki', '20200501.kj', '20200501.kk', '20200501.kl', '20200501.km', '20200501.kn', '20200501.ko', '20200501.koi', '20200501.krc', '20200501.ks', '20200501.ksh', '20200501.ku', '20200501.kv', '20200501.kw', '20200501.ky', '20200501.la', '20200501.lad', '20200501.lb', '20200501.lbe', '20200501.lez', '20200501.lfn', '20200501.lg', '20200501.li', '20200501.lij', '20200501.lmo', '20200501.ln', '20200501.lo', '20200501.lrc', '20200501.lt', '20200501.ltg', '20200501.lv', '20200501.mai', '20200501.map-bms', '20200501.mdf', '20200501.mg', '20200501.mh', '20200501.mhr', '20200501.mi', '20200501.min', '20200501.mk', '20200501.ml', '20200501.mn', '20200501.mr', '20200501.mrj', '20200501.ms', '20200501.mt', '20200501.mus', '20200501.mwl', '20200501.my', '20200501.myv', '20200501.mzn', '20200501.na', '20200501.nah', '20200501.nap', '20200501.nds', '20200501.nds-nl', '20200501.ne', '20200501.new', '20200501.ng', '20200501.nl', '20200501.nn', '20200501.no', '20200501.nov', '20200501.nrm', '20200501.nso', '20200501.nv', '20200501.ny', '20200501.oc', '20200501.olo', '20200501.om', '20200501.or', '20200501.os', '20200501.pa', '20200501.pag', '20200501.pam', '20200501.pap', '20200501.pcd', '20200501.pdc', '20200501.pfl', '20200501.pi', '20200501.pih', '20200501.pl', '20200501.pms', '20200501.pnb', '20200501.pnt', '20200501.ps', '20200501.pt', '20200501.qu', '20200501.rm', '20200501.rmy', '20200501.rn', '20200501.ro', '20200501.roa-rup', '20200501.roa-tara', '20200501.ru', '20200501.rue', '20200501.rw', '20200501.sa', '20200501.sah', '20200501.sat', '20200501.sc', '20200501.scn', '20200501.sco', '20200501.sd', '20200501.se', '20200501.sg', '20200501.sh', '20200501.si', '20200501.simple', '20200501.sk', '20200501.sl', '20200501.sm', '20200501.sn', '20200501.so', '20200501.sq', '20200501.sr', '20200501.srn', '20200501.ss', '20200501.st', '20200501.stq', '20200501.su', '20200501.sv', '20200501.sw', '20200501.szl', '20200501.ta', '20200501.tcy', '20200501.te', '20200501.tet', '20200501.tg', '20200501.th', '20200501.ti', '20200501.tk', '20200501.tl', '20200501.tn', '20200501.to', '20200501.tpi', '20200501.tr', '20200501.ts', '20200501.tt', '20200501.tum', '20200501.tw', '20200501.ty', '20200501.tyv', '20200501.udm', '20200501.ug', '20200501.uk', '20200501.ur', '20200501.uz', '20200501.ve', '20200501.vec', '20200501.vep', '20200501.vi', '20200501.vls', '20200501.vo', '20200501.wa', '20200501.war', '20200501.wo', '20200501.wuu', '20200501.xal', '20200501.xh', '20200501.xmf', '20200501.yi', '20200501.yo', '20200501.za', '20200501.zea', '20200501.zh', '20200501.zh-classical', '20200501.zh-min-nan', '20200501.zh-yue', '20200501.zu']\r\n```\r\n\r\nI am pretty sure that `https://dumps.wikimedia.org/enwiki/20201120/dumpstatus.json` exists.",
"Thanks for reporting I created a PR to make the custom config work (language=\"zh\", date=\"20201120\").",
"@lhoestq Thanks!",
"For posterity, here's how I got the data I needed: I needed Bengali, so I had to check which dumps are available here: https://dumps.wikimedia.org/bnwiki/ , then I ran:\r\n```\r\nload_dataset(\"wikipedia\", language=\"bn\", date=\"20211101\",\r\n beam_runner=\"DirectRunner\")\r\n```"
] | Hi, I tried to download Sundanese and Javanese wikipedia data with the following snippet
```
jv_wiki = datasets.load_dataset('wikipedia', '20200501.jv', beam_runner='DirectRunner')
su_wiki = datasets.load_dataset('wikipedia', '20200501.su', beam_runner='DirectRunner')
```
And I get the following error for these two languages:
Javanese
```
FileNotFoundError: Couldn't find file at https://dumps.wikimedia.org/jvwiki/20200501/dumpstatus.json
```
Sundanese
```
FileNotFoundError: Couldn't find file at https://dumps.wikimedia.org/suwiki/20200501/dumpstatus.json
```
I found from https://github.com/huggingface/datasets/issues/577#issuecomment-688435085 that for small languages, they are directly downloaded and parsed from the Wikipedia dump site, but both of `https://dumps.wikimedia.org/jvwiki/20200501/dumpstatus.json` and `https://dumps.wikimedia.org/suwiki/20200501/dumpstatus.json` are no longer valid.
Any suggestions on how to handle this issue? Thanks! | 784 |
https://github.com/huggingface/datasets/issues/778 | Unexpected behavior when loading cached csv file? | [
"Hi ! Thanks for reporting.\r\nThe same issue was reported in #730 (but with the encodings instead of the delimiter). It was fixed by #770 .\r\nThe fix will be available in the next release :)",
"Thanks for the prompt reply and terribly sorry for the spam! \r\nLooking forward to the new release! "
] | I read a csv file from disk and forgot so specify the right delimiter. When i read the csv file again specifying the right delimiter it had no effect since it was using the cached dataset. I am not sure if this is unwanted behavior since i can always specify `download_mode="force_redownload"`. But i think it would be nice if the information what `delimiter` or what `column_names` were used would influence the identifier of the cached dataset.
Small snippet to reproduce the behavior:
```python
import datasets
with open("dummy_data.csv", "w") as file:
file.write("test,this;text\n")
print(datasets.load_dataset("csv", data_files="dummy_data.csv", split="train").column_names)
# ["test", "this;text"]
print(datasets.load_dataset("csv", data_files="dummy_data.csv", split="train", delimiter=";").column_names)
# still ["test", "this;text"]
```
By the way, thanks a lot for this amazing library! :) | 778 |
https://github.com/huggingface/datasets/issues/773 | Adding CC-100: Monolingual Datasets from Web Crawl Data | [
"cc @aconneau ;) ",
"These dataset files are no longer available. https://data.statmt.org/cc-100/ files provided in this link are no longer available. Can anybody fix that issue?\r\n@abhishekkrthakur @yjernite ",
"Hi ! Can you open an issue to report this problem ? This will help keep track of the fix :)",
"Ok"
] | ## Adding a Dataset
- **Name:** CC-100: Monolingual Datasets from Web Crawl Data
- **Description:** https://twitter.com/alex_conneau/status/1321507120848625665
- **Paper:** https://arxiv.org/abs/1911.02116
- **Data:** http://data.statmt.org/cc-100/
- **Motivation:** A large scale multi-lingual language modeling dataset. Text is de-duplicated and filtered by how "Wikipedia-like" it is, hopefully helping avoid some of the worst parts of the common crawl.
Instructions to add a new dataset can be found [here](https://huggingface.co/docs/datasets/share_dataset.html).
| 773 |
https://github.com/huggingface/datasets/issues/771 | Using `Dataset.map` with `n_proc>1` print multiple progress bars | [
"Yes it allows to monitor the speed of each process. Currently each process takes care of one shard of the dataset.\r\n\r\nAt one point we can consider using streaming batches to a pool of processes instead of sharding the dataset in `num_proc` parts. At that point it will be easy to use only one progress bar",
"Hi @lhoestq, I am facing a similar issue, it is annoying when lots of progress bars are printed. Is there a way to turn off this behavior? ",
"You can disable the progress bars with\r\n```python\r\nimport datasets\r\n\r\ndatasets.disable_progress_bar()\r\n```"
] | When using `Dataset.map` with `n_proc > 1`, only one of the processes should print a progress bar (to make the output readable). Right now, `n_proc` progress bars are printed. | 771 |
https://github.com/huggingface/datasets/issues/769 | How to choose proper download_mode in function load_dataset? | [
"`download_mode=datasets.GenerateMode.FORCE_REDOWNLOAD` should work.\r\nThis makes me think we we should rename this to DownloadMode.FORCE_REDOWNLOAD. Currently that's confusing",
"Can we just use `features=...` in `load_dataset` for this @lhoestq?",
"Indeed you should use `features` in this case. \r\n```python\r\nfeatures = Features({'text': Value('string'), 'label': Value('float32')})\r\ndataset = load_dataset('csv', data_files=['sst_test.csv'], features=features)\r\n```\r\nNote that because of an issue with the caching when you change the features (see #750 ) you still need to specify the `FORCE_REDOWNLOAD ` flag. I'm working on a fix for this one",
"https://github.com/huggingface/datasets/issues/769#issuecomment-717837832\r\n> This makes me think we we should rename this to DownloadMode.FORCE_REDOWNLOAD. Currently that's confusing\r\n\r\n@lhoestq do you still think we should rename it?\r\n",
"It's no big deal, but since it can be confusing to users I think it's worth renaming it, and deprecate `GenerateMode` until `datasets` 2.0 at least. IMO it's confusing to have `download_mode=GenerateMode.something`"
] | Hi, I am a beginner to datasets and I try to use datasets to load my csv file.
my csv file looks like this
```
text,label
"Effective but too-tepid biopic",3
"If you sometimes like to go to the movies to have fun , Wasabi is a good place to start .",4
"Emerges as something rare , an issue movie that 's so honest and keenly observed that it does n't feel like one .",5
```
First I try to use this command to load my csv file .
``` python
dataset=load_dataset('csv', data_files=['sst_test.csv'])
```
It seems good, but when i try to overwrite the convert_options to convert 'label' columns from int64 to float32 like this.
``` python
import pyarrow as pa
from pyarrow import csv
read_options = csv.ReadOptions(block_size=1024*1024)
parse_options = csv.ParseOptions()
convert_options = csv.ConvertOptions(column_types={'text': pa.string(), 'label': pa.float32()})
dataset = load_dataset('csv', data_files=['sst_test.csv'], read_options=read_options,
parse_options=parse_options, convert_options=convert_options)
```
It keeps the same:
```shell
Dataset(features: {'text': Value(dtype='string', id=None), 'label': Value(dtype='int64', id=None)}, num_rows: 2210)
```
I think this issue is caused by the parameter "download_mode" Default to REUSE_DATASET_IF_EXISTS because after I delete the cache_dir, it seems right.
Is it a bug? How to choose proper download_mode to avoid this issue?
| 769 |
https://github.com/huggingface/datasets/issues/768 | Add a `lazy_map` method to `Dataset` and `DatasetDict` | [
"This is cool! I think some aspects to think about and decide in terms of API are:\r\n- do we allow several methods (chained i guess)\r\n- how do we inspect the currently set method(s)\r\n- how do we control/reset them"
] | The library is great, but it would be even more awesome with a `lazy_map` method implemented on `Dataset` and `DatasetDict`. This would apply a function on a give item but when the item is requested. Two use cases:
1. load image on the fly
2. apply a random function and get different outputs at each epoch (like data augmentation or randomly masking a part of a sentence for BERT-like objectives). | 768 |
https://github.com/huggingface/datasets/issues/767 | Add option for named splits when using ds.train_test_split | [
"Yes definitely we should give more flexibility to control the name of the splits outputted by `train_test_split`.\r\n\r\nRelated is the very interesting feedback from @bramvanroy on how we should improve this method: /static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-split-main-dataset-into-train-dev-test-as-datasetdict%2F1090%2F5%5Cr%5Cn%5Cr%5CnAnd in particular that it should advantageously be able to split in 3 splits as well instead of just 2 like we copied from sklearn."
] | ### Feature Request π
Can we add a way to name your splits when using the `.train_test_split` function?
In almost every use case I've come across, I have a `train` and a `test` split in my `DatasetDict`, and I want to create a `validation` split. Therefore, its kinda useless to get a `test` split back from `train_test_split`, as it'll just overwrite my real `test` split that I intended to keep.
### Workaround
this is my hack for dealin with this, for now :slightly_smiling_face:
```python
from datasets import load_dataset
β
β
ds = load_dataset('imdb')
ds['train'], ds['validation'] = ds['train'].train_test_split(.1).values()
```
| 767 |
https://github.com/huggingface/datasets/issues/766 | [GEM] add DART data-to-text generation dataset | [
"Is this a duplicate of #924 ?",
"Yup, closing! Haven't been keeping track of the solved issues during the sprint."
] | ## Adding a Dataset
- **Name:** DART
- **Description:** DART consists of 82,191 examples across different domains with each input being a semantic RDF triple set derived from data records in tables and the tree ontology of the schema, annotated with sentence descriptions that cover all facts in the triple set.
- **Paper:** https://arxiv.org/abs/2007.02871v1
- **Data:** https://github.com/Yale-LILY/dart
- **Motivation:** the dataset will likely be included in the GEM benchmark
Instructions to add a new dataset can be found [here](https://huggingface.co/docs/datasets/share_dataset.html).
| 766 |
https://github.com/huggingface/datasets/issues/765 | [GEM] Add DART data-to-text generation dataset | [] | ## Adding a Dataset
- **Name:** DART
- **Description:** DART consists of 82,191 examples across different domains with each input being a semantic RDF triple set derived from data records in tables and the tree ontology of the schema, annotated with sentence descriptions that cover all facts in the triple set.
- **Paper:** https://arxiv.org/abs/2007.02871v1
- **Data:** https://github.com/Yale-LILY/dart
- **Motivation:** It will likely be included in the GEM generation evaluation benchmark
Instructions to add a new dataset can be found [here](https://huggingface.co/docs/datasets/share_dataset.html).
| 765 |
https://github.com/huggingface/datasets/issues/762 | [GEM] Add Czech Restaurant data-to-text generation dataset | [] | - Paper: https://www.aclweb.org/anthology/W19-8670.pdf
- Data: https://github.com/UFAL-DSG/cs_restaurant_dataset
- The dataset will likely be part of the GEM benchmark | 762 |
https://github.com/huggingface/datasets/issues/761 | Downloaded datasets are not usable offline | [
"Yes currently you need an internet connection because the lib tries to check for the etag of the dataset script online to see if you don't have it locally already.\r\n\r\nIf we add a way to store the etag/hash locally after the first download, it would allow users to first download the dataset with an internet connection, and still have it working without an internet connection.\r\n\r\nI'll let you know when we add this feature.",
"Already fixed by:\r\n- #1726"
] | I've been trying to use the IMDB dataset offline, but after downloading it and turning off the internet it still raises an error from the ```requests``` library trying to reach for the online dataset.
Is this the intended behavior ?
(Sorry, I wrote the the first version of this issue while still on nlp 0.3.0). | 761 |
https://github.com/huggingface/datasets/issues/760 | Add meta-data to the HANS dataset | [] | The current version of the [HANS dataset](https://github.com/huggingface/datasets/blob/master/datasets/hans/hans.py) is missing the additional information provided for each example, including the sentence parses, heuristic and subcase. | 760 |
https://github.com/huggingface/datasets/issues/759 | (Load dataset failure) ConnectionError: Couldnβt reach https://raw.githubusercontent.com/huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py | [
"Are you running the script on a machine with an internet connection ?",
"Yes , I can browse the url through Google Chrome.",
"Does this HEAD request return 200 on your machine ?\r\n```python\r\nimport requests \r\nrequests.head(\"https://raw.githubusercontent.com/huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py\")\r\n```\r\n\r\nIf it returns 200, could you try again to load the dataset ?",
"Thank you very much for your response.\r\nWhen I run \r\n``` \r\nimport requests \r\nrequests.head(\"https://raw.githubusercontent.com/huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py\")\r\n```\r\nIt returns 200.\r\n\r\nAnd I try again to load the dataset. I got the following errors again. \r\n\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\load.py\", line 608, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\builder.py\", line 475, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\builder.py\", line 531, in _download_and_prepare\r\n split_generators = self._split_generators(dl_manager, **split_generators_kwargs)\r\n File \"C:\\Users\\666666\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\cnn_dailymail\\0128610a44e10f25b4af6689441c72af86205282d26399642f7db38fa7535602\\cnn_dailymail.py\", line 253, in _split_generators\r\n dl_paths = dl_manager.download_and_extract(_DL_URLS)\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\download_manager.py\", line 254, in download_and_extract\r\n return self.extract(self.download(url_or_urls))\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\download_manager.py\", line 175, in download\r\n downloaded_path_or_paths = map_nested(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\py_utils.py\", line 224, in map_nested\r\n mapped = [\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\py_utils.py\", line 225, in <listcomp>\r\n _single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\py_utils.py\", line 163, in _single_map_nested\r\n return function(data_struct)\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\file_utils.py\", line 300, in cached_path\r\n output_path = get_from_cache(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\datasets\\utils\\file_utils.py\", line 475, in get_from_cache\r\n raise ConnectionError(\"Couldn't reach {}\".format(url))\r\nConnectionError: Couldn't reach https://drive.google.com/uc?export=download&id=0BwmD_VLjROrfTHk4NFg2SndKcjQ\r\n\r\nConnection error happened but the url was different.\r\n\r\nI add the following code.\r\n```\r\nrequests.head(\"https://drive.google.com/uc?export=download&id=0BwmD_VLjROrfTHk4NFg2SndKcjQ\")\r\n```\r\nThis didn't return 200\r\nIt returned like this:\r\n\r\nTraceback (most recent call last):\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connection.py\", line 159, in _new_conn\r\n conn = connection.create_connection(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\util\\connection.py\", line 84, in create_connection\r\n raise err\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\util\\connection.py\", line 74, in create_connection\r\n sock.connect(sa)\r\nTimeoutError: [WinError 10060] \r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connectionpool.py\", line 670, in urlopen\r\n httplib_response = self._make_request(\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connectionpool.py\", line 381, in _make_request\r\n self._validate_conn(conn)\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connectionpool.py\", line 978, in _validate_conn\r\n conn.connect()\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connection.py\", line 309, in connect\r\n conn = self._new_conn()\r\n File \"C:\\Users\\666666\\AppData\\Local\\Programs\\Python\\Python38\\lib\\site-packages\\urllib3\\connection.py\", line 171, in _new_conn\r\n raise NewConnectionError(\r\nurllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPSConnection object at 0x000001F6060618E0>: Failed to establish a new connection: [WinError 10060] ",
"Is google drive blocked on your network ?\r\nFor me \r\n```python\r\nrequests.head(\"https://drive.google.com/uc?export=download&id=0BwmD_VLjROrfTHk4NFg2SndKcjQ\")\r\n```\r\nreturns 200",
"I can browse the google drive through google chrome. It's weird. I can download the dataset through google drive manually.",
"Could you try to update `requests` maybe ?\r\nIt works with 2.23.0 on my side",
"My ```requests``` is 2.24.0 . It still can't return 200.",
"Is it possible I download the dataset manually from google drive and use it for further test ? How can I do this ? I want to reproduce the model in this link https://huggingface.co/patrickvonplaten/bert2bert-cnn_dailymail-fp16. But I can't download the dataset through load_dataset method . I have tried many times and the connection error always happens .\r\n",
"The head request should definitely work, not sure what's going on on your side.\r\nIf you find a way to make it work, please post it here since other users might encounter the same issue.\r\n\r\nIf you don't manage to fix it you can use `load_dataset` on google colab and then save it using `dataset.save_to_disk(\"path/to/dataset\")`.\r\nThen you can download the directory on your machine and do\r\n```python\r\nfrom datasets import load_from_disk\r\ndataset = load_from_disk(\"path/to/local/dataset\")\r\n```",
"Hi\r\nI want to know if this problem has been solved because I encountered a similar issue. Thanks.\r\n`train_data = datasets.load_dataset(\"xsum\", `split=\"train\")`\r\n`ConnectionError:` Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/1.1.3/datasets/xsum/xsum.py`",
"Hi @smile0925 ! Do you have an internet connection ? Are you using some kind of proxy that may block the access to this file ?\r\n\r\nOtherwise you can try to update `datasets` since we introduced retries for http requests in the 1.2.0 version\r\n```\r\npip install --upgrade datasets\r\n```\r\nLet me know if that helps.",
"Hi @lhoestq \r\nOh, may be you are right. I find that my server uses some kind of proxy that block the access to this file.\r\n\r\n\r\n",
"> Hi @lhoestq\r\n> Oh, may be you are right. I find that my server uses some kind of proxy that block the access to this file.\r\n> \r\n\r\nI have the same problem, have you solved it? Many thanks",
"Hi @ZhengxiangShi \r\nYou can first try whether your network can access these files. I need to use VPN to access these files, so I download the files that cannot be accessed to the local in advance, and then use them in the code. Like this,\r\n`train_data = datasets.load_dataset(\"xsum.py\", split=\"train\")`",
"For Ubuntu 20.04, there are the following feedback. \r\n\r\nGoogle Drive is ok, but raw.githubusercontent.com has a big problem. It seems that the raw github could not match the common urllib3 protocols. \r\n\r\n**1. Google Drive** \r\n\r\n```\r\nimport requests\r\n\r\nrequests.head(\"https://drive.google.com/uc?export=download&id=0BwmD_VLjROrfTHk4NFg2SndKcjQ\")\r\n<Response [200]>\r\n```\r\n\r\n**2. raw.githubusercontent.com**\r\n\r\n```\r\nimport requests\r\nrequests.head(\"https://raw.githubusercontent.com/huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py\")\r\n```\r\n........\r\n\r\nraise CertificateError(\r\nurllib3.util.ssl_match_hostname.CertificateError: hostname 'raw.githubusercontent.com' doesn't match either of 'default.ssl.fastly.net', 'fastly.com', '*.a.ssl.fastly.net', '*.hosts.fastly.net', '*.global.ssl.fastly.net', '*.fastly.com', 'a.ssl.fastly.net', 'purge.fastly.net', 'mirrors.fastly.net', 'control.fastly.net', 'tools.fastly.net'\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n........\r\nraise MaxRetryError(_pool, url, error or ResponseError(cause))\r\nurllib3.exceptions.MaxRetryError: HTTPSConnectionPool(host='raw.githubusercontent.com', port=443): Max retries exceeded with url: /huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py (Caused by SSLError(CertificateError(\"hostname 'raw.githubusercontent.com' doesn't match either of 'default.ssl.fastly.net', 'fastly.com', '*.a.ssl.fastly.net', '*.hosts.fastly.net', '*.global.ssl.fastly.net', '*.fastly.com', 'a.ssl.fastly.net', 'purge.fastly.net', 'mirrors.fastly.net', 'control.fastly.net', 'tools.fastly.net'\")))\r\n\r\n........\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n\r\n.......\r\n\r\nraise SSLError(e, request=request)\r\nrequests.exceptions.SSLError: HTTPSConnectionPool(host='raw.githubusercontent.com', port=443): Max retries exceeded with url: /huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py (Caused by SSLError(CertificateError(\"hostname 'raw.githubusercontent.com' doesn't match either of 'default.ssl.fastly.net', 'fastly.com', '*.a.ssl.fastly.net', '*.hosts.fastly.net', '*.global.ssl.fastly.net', '*.fastly.com', 'a.ssl.fastly.net', 'purge.fastly.net', 'mirrors.fastly.net', 'control.fastly.net', 'tools.fastly.net'\")))\r\n\r\n\r\n**3. XSUM**\r\n\r\n```\r\nfrom datasets import load_dataset\r\nraw_datasets = load_dataset(\"xsum\", split=\"train\")\r\n```\r\n\r\nConnectionError: Couldn't reach https://raw.githubusercontent.com/EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json (SSLError(MaxRetryError('HTTPSConnectionPool(host=\\'raw.githubusercontent.com\\', port=443): Max retries exceeded with url: /EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json (Caused by SSLError(CertificateError(\"hostname \\'raw.githubusercontent.com\\' doesn\\'t match either of \\'default.ssl.fastly.net\\', \\'fastly.com\\', \\'*.a.ssl.fastly.net\\', \\'*.hosts.fastly.net\\', \\'*.global.ssl.fastly.net\\', \\'*.fastly.com\\', \\'a.ssl.fastly.net\\', \\'purge.fastly.net\\', \\'mirrors.fastly.net\\', \\'control.fastly.net\\', \\'tools.fastly.net\\'\")))')))\r\n\r\n\r\n### The following snippet could not solve the implicit ssl error.\r\n\r\n```\r\nimport ssl\r\n\r\ntry:\r\n _create_unverified_https_context = ssl._create_unverified_context\r\nexcept AttributeError:\r\n pass\r\nelse:\r\n ssl._create_default_https_context = _create_unverified_https_context\r\n```\r\n\r\n",
"Only the oldest versions of `datasets` use raw.githubusercontent.com. Can you try updating `datasets` ?",
"Thank lhoestq fo the quick response. \r\n\r\nI solve the big issue with the command line as follows. \r\n\r\n**1. Open hosts (Ubuntu 20.04)**\r\n\r\n`$ sudo gedit /etc/hosts`\r\n\r\n**2. Add the command line into the hosts**\r\n\r\n`151.101.0.133 raw.githubusercontent.com`\r\n\r\n**3. Save hosts**\r\n\r\nAnd then the jupyter notebook can access to the datasets (module) and get the datasets of XSUM with raw.githubusercontent.com. \r\n\r\nSo it is not users' fault. But most of the suggestions in the web are wrong. Anyway, I solve the problem finally. \r\n\r\nBy the way, users need to add the other github commnads such as the following. \r\n\r\n`199.232.69.194 github.global.ssl.fastly.net`\r\n\r\nCheers!!!\r\n\r\n\r\n",
"I use the dataset 2.14.4 that published on Aug 8, 2023.εθͺζηβiPhoneε¨ 2023εΉ΄9ζ13ζ₯οΌ06:38οΌQuentin Lhoest ***@***.***> ειοΌο»Ώ\r\nOnly the oldest versions of datasets use raw.githubusercontent.com. Can you try updating datasets ?\r\n\r\nβReply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>"
] | Hey, I want to load the cnn-dailymail dataset for fine-tune.
I write the code like this
from datasets import load_dataset
test_dataset = load_dataset(βcnn_dailymailβ, β3.0.0β, split=βtrainβ)
And I got the following errors.
Traceback (most recent call last):
File βtest.pyβ, line 7, in
test_dataset = load_dataset(βcnn_dailymailβ, β3.0.0β, split=βtestβ)
File βC:\Users\666666\AppData\Local\Programs\Python\Python38\lib\site-packages\datasets\load.pyβ, line 589, in load_dataset
module_path, hash = prepare_module(
File βC:\Users\666666\AppData\Local\Programs\Python\Python38\lib\site-packages\datasets\load.pyβ, line 268, in prepare_module
local_path = cached_path(file_path, download_config=download_config)
File βC:\Users\666666\AppData\Local\Programs\Python\Python38\lib\site-packages\datasets\utils\file_utils.pyβ, line 300, in cached_path
output_path = get_from_cache(
File βC:\Users\666666\AppData\Local\Programs\Python\Python38\lib\site-packages\datasets\utils\file_utils.pyβ, line 475, in get_from_cache
raise ConnectionError(βCouldnβt reach {}β.format(url))
ConnectionError: Couldnβt reach https://raw.githubusercontent.com/huggingface/datasets/1.1.2/datasets/cnn_dailymail/cnn_dailymail.py
How can I fix this ? | 759 |
https://github.com/huggingface/datasets/issues/758 | Process 0 very slow when using num_procs with map to tokenizer | [
"Hi ! Thanks for reporting.\r\nIs the distribution of text length of your data evenly distributed across your dataset ? I mean, could it be because the examples in the first part of your dataset are slower to process ?\r\nAlso could how many CPUs can you use for multiprocessing ?\r\n```python\r\nimport multiprocessing\r\nprint(multiprocessing.cpu_count())\r\n```\r\nWhich tokenizer are you using ?",
"Using pre trained HF tokenizer. The result is the same with tokenizer multiprocessing off and on.\r\nI have (absolutely) no idea about the distribution, but since this issue occurs on all of my datasets(regardless of files), I don't think distribution is the problems.\r\n\r\nI can use up to 16 cores.",
"Ok weird, I don't manage to reproduce this issue on my side.\r\nDoes it happen even with `num_proc=2` for example ?\r\nAlso could you provide more details about your OS and the versions of tokenizers/datasets/multiprocess that you're using ?",
"Yes, I can confirm it also happens with ```num_proc=2```.\r\n```\r\ntokenizers 0.9.2\r\ndatasets 1.1.2\r\nmultiprocess 0.70.10\r\n```\r\n```\r\nLinux nipa2020-0629 4.4.0-178-generic #208-Ubuntu SMP Sun Apr 5 23:45:10 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux\r\n```",
"I can't reproduce on my side unfortunately with the same versions.\r\n\r\nDo you have issues when doing multiprocessing with python ?\r\n```python\r\nfrom tqdm.auto import tqdm\r\nfrom multiprocess import Pool, RLock\r\n\r\ndef process_data(shard):\r\n # implement\r\n\r\nnum_proc = 8\r\nshards = [] # implement, this must be a list of size num_proc\r\n\r\nwith Pool(num_proc, initargs=(RLock(),), initializer=tqdm.set_lock) as pool:\r\n results = [pool.apply_async(process_data, shard=shard) for shard in shards]\r\n transformed_shards = [r.get() for r in results]\r\n```",
"Nah, I'll just wait a few hours. Thank you for helping, though."
] | <img width="721" alt="image" src="https://user-images.githubusercontent.com/17930170/97066109-776d0d00-15ed-11eb-8bba-bb4d2e0fcc33.png">
The code I am using is
```
dataset = load_dataset("text", data_files=[file_path], split='train')
dataset = dataset.map(lambda ex: tokenizer(ex["text"], add_special_tokens=True,
truncation=True, max_length=args.block_size), num_proc=8)
dataset.set_format(type='torch', columns=['input_ids'])
dataset.save_to_disk(file_path+'.arrow')
```
| 758 |
https://github.com/huggingface/datasets/issues/757 | CUDA out of memory | [
"Could you provide more details ? What's the code you ran ?",
"```python\r\ntokenizer = FunnelTokenizer.from_pretrained('funnel-transformer/small')\r\n\r\ndef tokenize(batch):\r\n return tokenizer(batch['text'], padding='max_length', truncation=True,max_length=512)\r\n\r\ndataset = load_dataset(\"bookcorpus\",split='train[:1000]').shuffle()\r\ndataset = dataset.map(tokenize, batched=True, batch_size=512)\r\n\r\n# dataset = LineByLineTextDataset(\r\n# tokenizer=tokenizer,\r\n# file_path=\"./wiki1000.txt\",\r\n# block_size=128\r\n# )\r\n\r\ndata_collator = DataCollatorForLanguageModeling(\r\n tokenizer=tokenizer, mlm=True, mlm_probability=0.15\r\n)\r\n\r\nconfig=FunnelConfig(\r\n return_dict=True\r\n)\r\n\r\nmodel= FunnelForMaskedLM(config=config)\r\n\r\ntraining_args = TrainingArguments(\r\n output_dir=\"./checkpoints\",\r\n overwrite_output_dir=True,\r\n do_train=True,\r\n num_train_epochs=1,\r\n per_device_train_batch_size=16,\r\n per_device_eval_batch_size=16,\r\n save_steps=10000,\r\n logging_dir='./ptlogs'\r\n)\r\n\r\ntrainer = Trainer(\r\n model=model,\r\n args=training_args,\r\n data_collator=data_collator,\r\n train_dataset=dataset,\r\n)\r\ntrainer.train()\r\n```",
"`RuntimeError: CUDA out of memory. Tried to allocate 954.00 MiB (GPU 0; 15.90 GiB total capacity; 14.35 GiB already allocated; 753.75 MiB free; 14.39 GiB reserved in total by PyTorch)\r\nException raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):`\r\n\r\npart of error output",
"from funnel model to bert model : error still happened\r\n\r\nfrom your dataset to LineByLineTextDataset : error disapeared",
"notice i just loaded 1000 rows of data",
"the error happens when executing loss.backward()",
"Since you're using a data collator you don't need to tokenizer the dataset using `map`. Could you try not to use `map` and only the data collator instead ? The data collator is supposed to pad to the longest sequence in each batch afaik, instead of padding to 512.\r\n\r\nAlso cc @sgugger ",
"Closing this one.\r\nFeel free to re-open if you have other questions about this issue"
] | In your dataset ,cuda run out of memory as long as the trainer begins:
however, without changing any other element/parameter,just switch dataset to `LineByLineTextDataset`,everything becames OK.
| 757 |
https://github.com/huggingface/datasets/issues/752 | Clicking on a metric in the search page points to datasets page giving "Missing dataset" warning | [
"Thanks for the report, can reproduce. Will fix",
"Fixed now @ogabrielluiz "
] | Hi! Sorry if this isn't the right place to talk about the website, I just didn't exactly where to write this.
Searching a metric in https://huggingface.co/metrics gives the right results but clicking on a metric (E.g ROUGE) points to https://huggingface.co/datasets/rouge. Clicking on a metric without searching points to the right page.
Thanks for all the great work! | 752 |
https://github.com/huggingface/datasets/issues/751 | Error loading ms_marco v2.1 using load_dataset() | [
"There was a similar issue in #294 \r\nClearing the cache and download again the dataset did the job. Could you try to clear your cache and download the dataset again ?",
"I was able to load the dataset successfully, I'm pretty sure it's just a cache issue that you have.\r\nLet me know if clearing your cache fixes the problem",
"Yes, it indeed was a cache issue!\r\nThanks for reaching out!!"
] | Code:
`dataset = load_dataset('ms_marco', 'v2.1')`
Error:
```
`---------------------------------------------------------------------------
JSONDecodeError Traceback (most recent call last)
<ipython-input-16-34378c057212> in <module>()
9
10 # Downloading and loading a dataset
---> 11 dataset = load_dataset('ms_marco', 'v2.1')
10 frames
/usr/lib/python3.6/json/decoder.py in raw_decode(self, s, idx)
353 """
354 try:
--> 355 obj, end = self.scan_once(s, idx)
356 except StopIteration as err:
357 raise JSONDecodeError("Expecting value", s, err.value) from None
JSONDecodeError: Unterminated string starting at: line 1 column 388988661 (char 388988660)
`
``` | 751 |
https://github.com/huggingface/datasets/issues/750 | load_dataset doesn't include `features` in its hash | [] | It looks like the function `load_dataset` does not include what's passed in the `features` argument when creating a hash for a given dataset. As a result, if a user includes new features from an already downloaded dataset, those are ignored.
Example: some models on the hub have a different ordering for the labels than what `datasets` uses for MNLI so I'd like to do something along the lines of:
```
dataset = load_dataset("glue", "mnli")
features = dataset["train"].features
features["label"] = ClassLabel(names = ['entailment', 'contradiction', 'neutral']) # new label order
dataset = load_dataset("glue", "mnli", features=features)
``` | 750 |
https://github.com/huggingface/datasets/issues/749 | [XGLUE] Adding new dataset | [
"Amazing! ",
"Small poll @thomwolf @yjernite @lhoestq @JetRunner @qiweizhen .\r\n\r\nAs stated in the XGLUE paper: https://arxiv.org/pdf/2004.01401.pdf , for each of the 11 down-stream tasks training data is only available in English, whereas development and test data is available in multiple different language *cf.* here: \r\n\r\n\r\n\r\nSo, I'd suggest to have exactly 11 \"language-independent\" configs: \"ner\", \"pos\", ... and give the sample in each dataset in the config a \"language\" label being one of \"ar\", \"bg\", .... => To me this makes more sense than making languaga specific config, *e.g.* \"ner-de\", ...especially because training data is only available in English. Do you guys agree? ",
"In this case we should have named splits, so config `ner` has splits `train`, `validation`, `test-en`, `test-ar`, `test-bg`, etc...\r\n\r\nThis is more in the spirit of the task afaiu, and will avoid making users do the filtering step themselves when testing different models or different configurations of the same model.",
"I see your point! \r\n\r\nI think this would be quite feasible to do and makes sense to me as well! In the paper results are reported per language, so it seems more natural to do it this way. \r\n\r\nGood for me @yjernite ! What do the others think? @lhoestq \r\n",
"I agree with Yacine on this!",
"Okey actually not that easy to add things like `test-de` to `datasets` => this would be the first dataset to have this.\r\nSee: https://github.com/huggingface/datasets/pull/802",
"IMO we should have one config per language. That's what we're doing for xnli, xtreme etc.\r\nHaving split names that depend on the language seems wrong. We should try to avoid split names that are not train/val/test.\r\nSorry for late response on this one",
"@lhoestq agreed on having one config per language, but we also need to be able to have different split names and people are going to want to use hyphens, so we should at the very least warn them why it's failing :) E.g. for ANLI with different stages of data (currently using underscores) or https://www.tau-nlp.org/commonsenseqa with their train-sanity or dev-sanity splits",
"Yes sure ! Could you open a separate issue for that ?",
"Really cool dataset π btw. does Transformers support all 11 tasks π€ would be awesome to have a xglue script (like the \"normal\" glue one)",
"Just to make sure this is what we want here. If we add one config per language, \r\n\r\nthis means that this dataset ends up with well over 100 different configs most of which will have the same `train` split. The train split is always in English. Also, I'm not sure whether it's better for the user to be honest. \r\n\r\nI think it could be quite confusing for the user to have\r\n\r\n```python\r\ntrain_dataset = load_dataset(\"xglue\", \"ner-de\", split=\"train\")\r\n```\r\n\r\nin English even though it's `ner-de`.\r\n\r\nTo be honest, I'd prefer:\r\n\r\n```python\r\ntrain_dataset = load_dataset(\"xglue\", \"ner\", split=\"train\")\r\ntest_dataset_de = load_dataset(\"xglue\", \"ner\", split=\"test-de\")\r\ntest_dataset_fr = load_dataset(\"xglue\", \"ner\", split=\"test-fr\")\r\n```\r\n\r\nhere",
"Oh yes right I didn't notice the train set was always in english sorry.\r\nMoreover it seems that the way this dataset is used is to pick a pretrained multilingual model, fine-tune it on the english train set and then evaluate on each test set (one per language).\r\nSo to better fit the usual usage of this dataset, I agree that it's better to have one test split per language. \r\n\r\nSomething like your latest example patrick is fine imo :\r\n```python\r\ntrain_dataset = load_dataset(\"xglue\", \"ner\", split=\"train\")\r\ntest_dataset_de = load_dataset(\"xglue\", \"ner\", split=\"test.de\")\r\n```\r\n\r\nI just replace test-de with test.de since `-` is not allowed for split names (it has to follow the `\\w+` regex), and usually we specify the language after a point. ",
"Closing since XGLUE has been added in #802 , thanks patrick :) ",
"I need xglue Urdu summarization dataset so how can i get it?",
"According to the table in https://huggingface.co/datasets/xglue, Urdu only exists for POS and XNLI in XGLUE - not for summarization"
] | XGLUE is a multilingual GLUE like dataset propesed in this [paper](https://arxiv.org/pdf/2004.01401.pdf).
I'm planning on adding the dataset to the library myself in a couple of weeks.
Also tagging @JetRunner @qiweizhen in case I need some guidance | 749 |
https://github.com/huggingface/datasets/issues/744 | Dataset Explorer Doesn't Work for squad_es and squad_it | [
"Oups wrong click.\r\nThis one is for you @srush"
] | https://huggingface.co/nlp/viewer/?dataset=squad_es
https://huggingface.co/nlp/viewer/?dataset=squad_it
Both pages show "OSError: [Errno 28] No space left on device". | 744 |
https://github.com/huggingface/datasets/issues/743 | load_dataset for CSV files not working | [
"Thank you !\r\nCould you provide a csv file that reproduces the error ?\r\nIt doesn't have to be one of your dataset. As long as it reproduces the error\r\nThat would help a lot !",
"I think another good example is the following:\r\n`\r\nfrom datasets import load_dataset\r\n`\r\n`\r\ndataset = load_dataset(\"csv\", data_files=[\"./sts-dev.csv\"], delimiter=\"\\t\", column_names=[\"one\", \"two\", \"three\", \"four\", \"score\", \"sentence1\", \"sentence2\"], script_version=\"master\")`\r\n`\r\n\r\nDisplayed error `CSV parse error: Expected 7 columns, got 6` even tough I put 7 columns. First four columns from the csv don't have a name, so I've named them by default. The csv file is the .dev file from STSb benchmark dataset.\r\n\r\n",
"Hi, seems I also can't read csv file. I was trying with a dummy csv with only three rows.\r\n\r\n```\r\ntext,label\r\nI hate google,negative\r\nI love Microsoft,positive\r\nI don't like you,negative\r\n```\r\nI was using the HuggingFace image in Paperspace Gradient (datasets==1.1.3). The following code doesn't work:\r\n\r\n```\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', script_version=\"master\", data_files=['test_data.csv'], delimiter=\",\")\r\n```\r\nIt outputs the following:\r\n```\r\nUsing custom data configuration default\r\nDownloading and preparing dataset csv/default-3b6254ff4dd403e5 (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /root/.cache/huggingface/datasets/csv/default-3b6254ff4dd403e5/0.0.0/2960f95a26e85d40ca41a230ac88787f715ee3003edaacb8b1f0891e9f04dda2...\r\nDataset csv downloaded and prepared to /root/.cache/huggingface/datasets/csv/default-3b6254ff4dd403e5/0.0.0/2960f95a26e85d40ca41a230ac88787f715ee3003edaacb8b1f0891e9f04dda2. Subsequent calls will reuse this data.\r\n```\r\nBut `len(dataset)` gives `1` and I can't access rows with indexing `dataset[0]` (it gives `KeyError: 0`).\r\n\r\nHowever, loading from pandas dataframe is working.\r\n```\r\nfrom datasets import Dataset\r\nimport pandas as pd\r\ndf = pd.read_csv('test_data.csv')\r\ndataset = Dataset.from_pandas(df)\r\n```\r\n\r\n",
"This is because load_dataset without `split=` returns a dictionary of split names (train/validation/test) to dataset.\r\nYou can do\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', script_version=\"master\", data_files=['test_data.csv'], delimiter=\",\")\r\nprint(dataset[\"train\"][0])\r\n```\r\n\r\nOr if you want to directly get the train split:\r\n\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', script_version=\"master\", data_files=['test_data.csv'], delimiter=\",\", split=\"train\")\r\nprint(dataset[0])\r\n```\r\n",
"Good point\r\n\r\nDesign question for us, though: should `load_dataset` when no split is specified and only one split is present in the dataset (common use case with CSV/text/JSON datasets) return a `Dataset` instead of a `DatsetDict`? I feel like it's often what the user is expecting. I break a bit the paradigm of a unique return type but since this library is designed for widespread DS people more than CS people usage I would tend to think that UX should take precedence over CS reasons. What do you think?",
"In this case the user expects to get only one dataset object instead of the dictionary of datasets since only one csv file was specified without any split specifications.\r\nI'm ok with returning the dataset object if no split specifications are given for text/json/csv/pandas.\r\n\r\nFor the other datasets ton the other hand the user doesn't know in advance the splits so I would keep the dictionary by default. What do you think ?",
"Thanks for your quick response! I'm fine with specifying the split as @lhoestq suggested. My only concern is when I'm loading from python dict or pandas, the library returns a dataset instead of a dictionary of datasets when no split is specified. I know that they use a different function `Dataset.from_dict` or `Dataset.from_pandas` but the text/csv files use `load_dataset()`. However, to the user, they do the same task and we probably expect them to have the same behavior.",
"```\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', data_files='./amazon_data/Video_Games_5.csv', delimiter=\",\", split=['train', 'test'])\r\n```\r\nI was running the above line, but got this error.\r\n\r\n```ValueError: Unknown split \"test\". Should be one of ['train'].```\r\n\r\nThe data is amazon product data. I load the Video_Games_5.json.gz data into pandas and save it as csv file. and then load the csv file using the above code. I thought, ```split=['train', 'test']``` would split the data into train and test. did I misunderstood?\r\n\r\nThank you!\r\n\r\n",
"Hi ! the `split` argument in `load_dataset` is used to select the splits you want among the available splits.\r\nHowever when loading a csv with a single file as you did, only a `train` split is available by default.\r\n\r\nIndeed since `data_files='./amazon_data/Video_Games_5.csv'` is equivalent to `data_files={\"train\": './amazon_data/Video_Games_5.csv'}`, you can get a dataset with \r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', data_files='./amazon_data/Video_Games_5.csv', delimiter=\",\", split=\"train\")\r\n```\r\n\r\nAnd then to get both a train and test split you can do\r\n```python\r\ndataset = dataset.train_test_split()\r\nprint(dataset.keys())\r\n# ['train', 'test']\r\n```\r\n\r\n\r\nAlso note that a csv dataset may have several available splits if it is defined this way:\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', data_files={\r\n \"train\": './amazon_data/Video_Games_5_train.csv',\r\n \"test\": './amazon_data/Video_Games_5_test.csv'\r\n})\r\nprint(dataset.keys())\r\n# ['train', 'test']\r\n```\r\n",
"> In this case the user expects to get only one dataset object instead of the dictionary of datasets since only one csv file was specified without any split specifications.\r\n> I'm ok with returning the dataset object if no split specifications are given for text/json/csv/pandas.\r\n> \r\n> For the other datasets ton the other hand the user doesn't know in advance the splits so I would keep the dictionary by default. What do you think ?\r\n\r\nYes maybe this would be good. I think having to select 'train' from the resulting object why the user gave no split information is a confusing and not intuitive behavior.",
"> Similar to #622, I've noticed there is a problem when trying to load a CSV file with datasets.\r\n> \r\n> `from datasets import load_dataset`\r\n> `dataset = load_dataset(\"csv\", data_files=[\"./sample_data.csv\"], delimiter=\"\\t\", column_names=[\"title\", \"text\"], script_version=\"master\")`\r\n> \r\n> Displayed error:\r\n> `... ArrowInvalid: CSV parse error: Expected 2 columns, got 1`\r\n\r\nI'm also facing the same issue when trying to load from a csv file locally:\r\n\r\n```python\r\nfrom nlp import load_dataset\r\ndataset = load_dataset('csv', data_files='sample_data.csv')\r\n```\r\n\r\nError when executed from Google Colab:\r\n```python\r\nArrowInvalid Traceback (most recent call last)\r\n<ipython-input-34-79a8d4f65ed6> in <module>()\r\n 1 from nlp import load_dataset\r\n----> 2 dataset = load_dataset('csv', data_files='sample_data.csv')\r\n\r\n9 frames\r\n/usr/local/lib/python3.7/dist-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)\r\n 547 # Download and prepare data\r\n 548 builder_instance.download_and_prepare(\r\n--> 549 download_config=download_config, download_mode=download_mode, ignore_verifications=ignore_verifications,\r\n 550 )\r\n 551 \r\n\r\n/usr/local/lib/python3.7/dist-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)\r\n 461 if not downloaded_from_gcs:\r\n 462 self._download_and_prepare(\r\n--> 463 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n 464 )\r\n 465 # Sync info\r\n\r\n/usr/local/lib/python3.7/dist-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 535 try:\r\n 536 # Prepare split will record examples associated to the split\r\n--> 537 self._prepare_split(split_generator, **prepare_split_kwargs)\r\n 538 except OSError:\r\n 539 raise OSError(\"Cannot find data file. \" + (self.manual_download_instructions or \"\"))\r\n\r\n/usr/local/lib/python3.7/dist-packages/nlp/builder.py in _prepare_split(self, split_generator)\r\n 863 \r\n 864 generator = self._generate_tables(**split_generator.gen_kwargs)\r\n--> 865 for key, table in utils.tqdm(generator, unit=\" tables\", leave=False):\r\n 866 writer.write_table(table)\r\n 867 num_examples, num_bytes = writer.finalize()\r\n\r\n/usr/local/lib/python3.7/dist-packages/tqdm/notebook.py in __iter__(self, *args, **kwargs)\r\n 213 def __iter__(self, *args, **kwargs):\r\n 214 try:\r\n--> 215 for obj in super(tqdm_notebook, self).__iter__(*args, **kwargs):\r\n 216 # return super(tqdm...) will not catch exception\r\n 217 yield obj\r\n\r\n/usr/local/lib/python3.7/dist-packages/tqdm/std.py in __iter__(self)\r\n 1102 fp_write=getattr(self.fp, 'write', sys.stderr.write))\r\n 1103 \r\n-> 1104 for obj in iterable:\r\n 1105 yield obj\r\n 1106 # Update and possibly print the progressbar.\r\n\r\n/usr/local/lib/python3.7/dist-packages/nlp/datasets/csv/ede98314803c971fef04bcee45d660c62f3332e8a74491e0b876106f3d99bd9b/csv.py in _generate_tables(self, files)\r\n 78 read_options=self.config.pa_read_options,\r\n 79 parse_options=self.config.pa_parse_options,\r\n---> 80 convert_options=self.config.convert_options,\r\n 81 )\r\n 82 yield i, pa_table\r\n\r\n/usr/local/lib/python3.7/dist-packages/pyarrow/_csv.pyx in pyarrow._csv.read_csv()\r\n\r\n/usr/local/lib/python3.7/dist-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()\r\n\r\n/usr/local/lib/python3.7/dist-packages/pyarrow/error.pxi in pyarrow.lib.check_status()\r\n\r\nArrowInvalid: CSV parse error: Expected 1 columns, got 8\r\n```\r\n\r\nVersion:\r\n```\r\nnlp==0.4.0\r\n```",
"Hi @kauvinlucas\r\n\r\nYou can use the latest versions of `datasets` to do this.\r\nTo do so, just `pip install datasets` instead of `nlp` (the library was renamed) and then\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', data_files='sample_data.csv')",
"Hi \r\nI'm having a different problem with loading local csv. \r\n```Python\r\nfrom datasets import load_dataset \r\ndataset = load_dataset('csv', data_files='sample.csv') \r\n``` \r\n\r\ngives `ValueError: Specified named and prefix; you can only specify one.` error \r\n\r\nversions: \r\n- datasets: 1.1.3 \r\n- python: 3.9.6 \r\n- pyarrow: 2.0.0 ",
"Oh.. I figured it out. According to issue #[42387](https://github.com/pandas-dev/pandas/issues/42387) from pandas, this new version does not accept None for both parameters (which was being done by the repo I'm testing). Dowgrading Pandas==1.0.4 and Python==3.8 worked",
"Hi, \r\nI got an `OSError: Cannot find data file. ` when I tried to use load_dataset with tsv files. I have checked the paths, and they are correct. \r\n\r\nversions\r\n- python: 3.7.9\r\n- datasets: 1.1.3\r\n- pyarrow: 2.0.0\r\n- transformers: 4.2.2\r\n\r\n~~~\r\ndata_files = {\"train\": \"train.tsv\", \"test\",: \"test.tsv\"}\r\ndatasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\n~~~\r\n\r\nThe entire Error message is on below:\r\n\r\n```08/14/2021 16:55:44 - INFO - __main__ - load a local file for train: /project/media-framing/transformer4/data/0/val/label1.tsv\r\n08/14/2021 16:55:44 - INFO - __main__ - load a local file for test: /project/media-framing/transformer4/data/unlabel/test.tsv\r\nUsing custom data configuration default\r\nDownloading and preparing dataset csv/default-00a4200ae8507533 (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /usr4/cs542sp/hey1/.cache/huggingface/datasets/csv/default-00a4200ae8507533/0.0.0/2960f95a26e85d40ca41a230ac88787f715ee3003edaacb8b1f0891e9f04dda2...\r\nTraceback (most recent call last):\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 592, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 944, in _prepare_split\r\n num_examples, num_bytes = writer.finalize()\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/arrow_writer.py\", line 307, in finalize\r\n self.stream.close()\r\n File \"pyarrow/io.pxi\", line 132, in pyarrow.lib.NativeFile.close\r\n File \"pyarrow/error.pxi\", line 99, in pyarrow.lib.check_status\r\nOSError: error closing file\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"run_glue.py\", line 484, in <module>\r\n main()\r\n File \"run_glue.py\", line 243, in main\r\n datasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/load.py\", line 610, in load_dataset\r\n ignore_verifications=ignore_verifications,\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 515, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 594, in _download_and_prepare\r\n raise OSError(\"Cannot find data file. \" + (self.manual_download_instructions or \"\"))\r\nOSError: Cannot find data file. ```",
"Hi ! It looks like the error stacktrace doesn't match with your code snippet.\r\n\r\nWhat error do you get when running this ?\r\n```\r\ndata_files = {\"train\": \"train.tsv\", \"test\",: \"test.tsv\"}\r\ndatasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\n```\r\ncan you check that both tsv files are in the same folder as the current working directory of your shell ?",
"Hi @lhoestq, Below is the entire error message after I move both tsv files to the same directory. It's the same with I got before.\r\n```\r\n/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/torch/cuda/__init__.py:52: UserWarning: CUDA initialization: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx (Triggered internally at /pytorch/c10/cuda/CUDAFunctions.cpp:100.)\r\n return torch._C._cuda_getDeviceCount() > 0\r\n08/29/2021 22:56:43 - WARNING - __main__ - Process rank: -1, device: cpu, n_gpu: 0distributed training: False, 16-bits training: False\r\n08/29/2021 22:56:43 - INFO - __main__ - Training/evaluation parameters TrainingArguments(output_dir=/projectnb/media-framing/pred_result/label1/, overwrite_output_dir=True, do_train=True, do_eval=False, do_predict=True, evaluation_strategy=EvaluationStrategy.NO, prediction_loss_only=False, per_device_train_batch_size=8, per_device_eval_batch_size=8, gradient_accumulation_steps=1, eval_accumulation_steps=None, learning_rate=5e-05, weight_decay=0.0, adam_beta1=0.9, adam_beta2=0.999, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=8.0, max_steps=-1, lr_scheduler_type=SchedulerType.LINEAR, warmup_steps=0, logging_dir=runs/Aug29_22-56-43_scc1, logging_first_step=False, logging_steps=500, save_steps=3000, save_total_limit=None, no_cuda=False, seed=42, fp16=False, fp16_opt_level=O1, fp16_backend=auto, local_rank=-1, tpu_num_cores=None, tpu_metrics_debug=False, debug=False, dataloader_drop_last=False, eval_steps=500, dataloader_num_workers=0, past_index=-1, run_name=/projectnb/media-framing/pred_result/label1/, disable_tqdm=False, remove_unused_columns=True, label_names=None, load_best_model_at_end=False, metric_for_best_model=None, greater_is_better=None, ignore_data_skip=False, sharded_ddp=False, deepspeed=None, label_smoothing_factor=0.0, adafactor=False, _n_gpu=0)\r\n08/29/2021 22:56:43 - INFO - __main__ - load a local file for train: /project/media-framing/transformer4/temp_train.tsv\r\n08/29/2021 22:56:43 - INFO - __main__ - load a local file for test: /project/media-framing/transformer4/temp_test.tsv\r\n08/29/2021 22:56:43 - WARNING - datasets.builder - Using custom data configuration default-df627c23ac0e98ec\r\nDownloading and preparing dataset csv/default (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /usr4/cs542sp/hey1/.cache/huggingface/datasets/csv/default-df627c23ac0e98ec/0.0.0/9144e0a4e8435090117cea53e6c7537173ef2304525df4a077c435d8ee7828ff...\r\nTraceback (most recent call last):\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 693, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 1166, in _prepare_split\r\n num_examples, num_bytes = writer.finalize()\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/arrow_writer.py\", line 428, in finalize\r\n self.stream.close()\r\n File \"pyarrow/io.pxi\", line 132, in pyarrow.lib.NativeFile.close\r\n File \"pyarrow/error.pxi\", line 99, in pyarrow.lib.check_status\r\nOSError: error closing file\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"run_glue.py\", line 487, in <module>\r\n main()\r\n File \"run_glue.py\", line 244, in main\r\n datasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/load.py\", line 852, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 616, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 699, in _download_and_prepare\r\n + str(e)\r\nOSError: Cannot find data file. \r\nOriginal error:\r\nerror closing file\r\n```",
"Hi !\r\nCan you try running this into a python shell directly ?\r\n\r\n```python\r\nimport os\r\nfrom datasets import load_dataset\r\n\r\ndata_files = {\"train\": \"train.tsv\", \"test\": \"test.tsv\"}\r\nassert all(os.path.isfile(data_file) for data_file in data_files.values()), \"Couln't find files\"\r\n\r\ndatasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\nprint(\"success !\")\r\n```\r\n\r\nThis way all the code from `run_glue.py` doesn't interfere with our tests :)",
"Hi @lhoestq, \r\n\r\nBelow is what I got from terminal after I copied and run your code. I think the files themselves are good since there is no assertion error. \r\n\r\n```\r\nUsing custom data configuration default-df627c23ac0e98ec\r\nDownloading and preparing dataset csv/default (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /usr4/cs542sp/hey1/.cache/huggingface/datasets/csv/default-df627c23ac0e98ec/0.0.0/9144e0a4e8435090117cea53e6c7537173ef2304525df4a077c435d8ee7828ff...\r\nTraceback (most recent call last):\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 693, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 1166, in _prepare_split\r\n num_examples, num_bytes = writer.finalize()\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/arrow_writer.py\", line 428, in finalize\r\n self.stream.close()\r\n File \"pyarrow/io.pxi\", line 132, in pyarrow.lib.NativeFile.close\r\n File \"pyarrow/error.pxi\", line 99, in pyarrow.lib.check_status\r\nOSError: error closing file\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"test.py\", line 7, in <module>\r\n datasets = load_dataset(\"csv\", data_files=data_files, delimiter=\"\\t\")\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/load.py\", line 852, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 616, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/projectnb2/media-framing/env-trans4/lib/python3.7/site-packages/datasets/builder.py\", line 699, in _download_and_prepare\r\n + str(e)\r\nOSError: Cannot find data file. \r\nOriginal error:\r\nerror closing file\r\n```",
"Hi, could this be a permission error ? I think it fails to close the arrow file that contains the data from your CSVs in the cache.\r\n\r\nBy default datasets are cached in `~/.cache/huggingface/datasets`, could you check that you have the right permissions ?\r\nYou can also try to change the cache directory by passing `cache_dir=\"path/to/my/cache/dir\"` to `load_dataset`.",
"Thank you!! @lhoestq\r\n\r\nFor some reason, I don't have the default path for datasets to cache, maybe because I work from a remote system. The issue solved after I pass the `cache_dir` argument to the function. Thank you very much!!",
"> Hi, could this be a permission error ? I think it fails to close the arrow file that contains the data from your CSVs in the cache.\r\n> \r\n> By default datasets are cached in `~/.cache/huggingface/datasets`, could you check that you have the right permissions ? You can also try to change the cache directory by passing `cache_dir=\"path/to/my/cache/dir\"` to `load_dataset`.\r\n\r\nThis is the exact solution I have been finding for the whole afternoon. Thanks a lot!\r\nI tried to do a training on a cluster computing system. The user's home directory is shared between nodes.\r\nIt always gets **stuck** at dataset loading...\r\nThe reason might be, the node (with GPU) can't read/write data in the default cache folder (in my home directory).\r\nAfter using an intermediate cache folder, this issue is resolved for me."
] | Similar to #622, I've noticed there is a problem when trying to load a CSV file with datasets.
`
from datasets import load_dataset
`
`
dataset = load_dataset("csv", data_files=["./sample_data.csv"], delimiter="\t", column_names=["title", "text"], script_version="master")
`
Displayed error:
`
...
ArrowInvalid: CSV parse error: Expected 2 columns, got 1
`
I should mention that when I've tried to read data from `https://github.com/lhoestq/transformers/tree/custom-dataset-in-rag-retriever/examples/rag/test_data/my_knowledge_dataset.csv` it worked without a problem. I've read that there might be some problems with /r character, so I've removed them from the custom dataset, but the problem still remains.
I've added a colab reproducing the bug, but unfortunately I cannot provide the dataset.
https://colab.research.google.com/drive/1Qzu7sC-frZVeniiWOwzoCe_UHZsrlxu8?usp=sharing
Are there any work around for it ?
Thank you | 743 |
https://github.com/huggingface/datasets/issues/741 | Creating dataset consumes too much memory | [
"Thanks for reporting.\r\nIn theory since the dataset script is just made to yield examples to write them into an arrow file, it's not supposed to create memory issues.\r\n\r\nCould you please try to run this exact same loop in a separate script to see if it's not an issue with `PIL` ?\r\nYou can just copy paste what's inside `_generate_examples` and remove all the code for `datasets` (remove yield).\r\n\r\nIf the RAM usage stays low after 600 examples it means that it comes from some sort of memory leak in the library, or with pyarrow.",
"Here's an equivalent loading code:\r\n```python\r\nimages_path = \"PHOENIX-2014-T-release-v3/PHOENIX-2014-T/features/fullFrame-210x260px/train\"\r\n\r\nfor dir_path in tqdm(os.listdir(images_path)):\r\n frames_path = os.path.join(images_path, dir_path)\r\n np_frames = []\r\n for frame_name in os.listdir(frames_path):\r\n frame_path = os.path.join(frames_path, frame_name)\r\n im = Image.open(frame_path)\r\n np_frames.append(np.asarray(im))\r\n im.close()\r\n```\r\n\r\nThe process takes 0.3% of memory, even after 1000 examples on the small machine with 120GB RAM.\r\n\r\nI guess something in the datasets library doesn't release the reference to the objects I'm yielding, but no idea how to test for this",
"I've had similar issues with Arrow once. I'll investigate...\r\n\r\nFor now maybe we can simply use the images paths in the dataset you want to add. I don't expect to fix this memory issue until 1-2 weeks unfortunately. Then we can just update the dataset with the images. What do you think ?",
"If it's just 1-2 weeks, I think it's best if we wait. I don't think it is very urgent to add it, and it will be much more useful with the images loaded rather than not (the images are low resolution, and thus papers using this dataset actually fit the entire video into memory anyway)\r\n\r\nI'll keep working on other datasets in the meanwhile :) ",
"Ok found the issue. This is because the batch size used by the writer is set to 10 000 elements by default so it would load your full dataset in memory (the writer has a buffer that flushes only after each batch). Moreover to write in Apache Arrow we have to use python objects so what's stored inside the ArrowWriter's buffer is actually python integers (32 bits).\r\n\r\nLowering the batch size to 10 should do the job.\r\n\r\nI will add a flag to the DatasetBuilder class of dataset scripts, so that we can customize the batch size.",
"Thanks, that's awesome you managed to find the problem.\r\n\r\nAbout the 32 bits - really? there isn't a way to serialize the numpy array somehow? 32 bits would take 4 times the memory / disk space needed to store these videos.\r\n\r\nPlease let me know when the batch size is customizable and I'll try again!",
"The 32 bit integrers are only used in the writer's buffer because Arrow doesn't take numpy arrays correctly as input. On disk it's stored as uint8 in arrow format ;)",
"> I don't expect to fix this memory issue until 1-2 weeks unfortunately.\r\n\r\nHi @lhoestq \r\nnot to rush of course, but I was wondering if you have a new timeline so I know how to plan my work around this :) ",
"Hi ! Next week for sure :) ",
"Alright it should be good now.\r\nYou just have to specify `_writer_batch_size = 10` for example as a class attribute of the dataset builder class.",
"I added it, but still it consumes as much memory\r\n\r\nhttps://github.com/huggingface/datasets/pull/722/files#diff-2e0d865dd4a60dedd1861d6f8c5ed281ded71508467908e1e0b1dbe7d2d420b1R66\r\n\r\nDid I not do it correctly?",
"Yes you did it right.\r\nDid you rebase to include the changes of #828 ?\r\n\r\nEDIT: looks like you merged from master in the PR. Not sure why you still have an issue then, I will investigate",
"Hi @lhoestq, any update on this?\r\nPerhaps even a direction I could try myself?",
"Sorry for the delay, I was busy with the dataset sprint and the incredible amount of contributions to the library ^^'\r\n\r\nWhat you can try to do to find what's wrong is check at which frequency the arrow writer writes all the examples from its in-memory buffer on disk. This happens [here](https://github.com/huggingface/datasets/blob/master/src/datasets/arrow_writer.py#L257-L258) in the code.\r\n\r\nThe idea is that `write_on_file` writes the examples every `writer_batch_size` examples and clear the buffer `self. current_rows`. As soon as `writer_batch_size` is small enough you shouldn't have memory issues in theory.\r\n\r\nLet me know if you have questions or if I can help.\r\n\r\nSince the dataset sprint is over and I will also be done with all the PRs soon I will be able to go back at it and take a look.",
"Thanks. I gave it a try and no success. I'm not sure what's happening there",
"I had the same issue. It works for me by setting `DEFAULT_WRITER_BATCH_SIZE = 10` of my dataset builder class. (And not `_writer_batch_size` as previously mentioned). I guess this is because `_writer_batch_size` is overwritten in `__init__` (see [here](https://github.com/huggingface/datasets/blob/0e2563e5d5c2fc193ea27d7c24607bb35607f2d5/src/datasets/builder.py#L934))",
"Yes the class attribute you can change is `DEFAULT_WRITER_BATCH_SIZE`.\r\nOtherwise in `load_dataset` you can specify `writer_batch_size=`",
"Ok thanks for the tips. Maybe the documentation should be updated accordingly https://huggingface.co/docs/datasets/add_dataset.html.",
"Thanks for reporting this mistake in the docs.\r\nI just fixed it at https://github.com/huggingface/datasets/commit/85cf7ff920c90ca2e12bedca12b36d2a043c3da2",
"May I close this issue, @AmitMY?"
] | Moving this issue from https://github.com/huggingface/datasets/pull/722 here, because it seems like a general issue.
Given the following dataset example, where each example saves a sequence of 260x210x3 images (max length 400):
```python
def _generate_examples(self, base_path, split):
""" Yields examples. """
filepath = os.path.join(base_path, "annotations", "manual", "PHOENIX-2014-T." + split + ".corpus.csv")
images_path = os.path.join(base_path, "features", "fullFrame-210x260px", split)
with open(filepath, "r", encoding="utf-8") as f:
data = csv.DictReader(f, delimiter="|", quoting=csv.QUOTE_NONE)
for row in data:
frames_path = os.path.join(images_path, row["video"])[:-7]
np_frames = []
for frame_name in os.listdir(frames_path):
frame_path = os.path.join(frames_path, frame_name)
im = Image.open(frame_path)
np_frames.append(np.asarray(im))
im.close()
yield row["name"], {"video": np_frames}
```
The dataset creation process goes out of memory on a machine with 500GB RAM.
I was under the impression that the "generator" here is exactly for that, to avoid memory constraints.
However, even if you want the entire dataset in memory, it would be in the worst case
`260x210x3 x 400 max length x 7000 samples` in bytes (uint8) = 458.64 gigabytes
So I'm not sure why it's taking more than 500GB.
And the dataset creation fails after 170 examples on a machine with 120gb RAM, and after 672 examples on a machine with 500GB RAM.
---
## Info that might help:
Iterating over examples is extremely slow.
If I perform this iteration in my own, custom loop (Without saving to file), it runs at 8-9 examples/sec
And you can see at this state it is using 94% of the memory:
And it is only using one CPU core, which is probably why it's so slow:
| 741 |
https://github.com/huggingface/datasets/issues/737 | Trec Dataset Connection Error | [
"Thanks for reporting.\r\nThat's because the download url has changed. The old url now redirects to the new one but we don't support redirection for downloads.\r\n\r\nI'm opening a PR to update the url"
] | **Datasets Version:**
1.1.2
**Python Version:**
3.6/3.7
**Code:**
```python
from datasets import load_dataset
load_dataset("trec")
```
**Expected behavior:**
Download Trec dataset and load Dataset object
**Current Behavior:**
Get a connection error saying it couldn't reach http://cogcomp.org/Data/QA/QC/train_5500.label (but the link doesn't seem broken)
<details>
<summary>Error Logs</summary>
Using custom data configuration default
Downloading and preparing dataset trec/default (download: 350.79 KiB, generated: 403.39 KiB, post-processed: Unknown size, total: 754.18 KiB) to /root/.cache/huggingface/datasets/trec/default/1.1.0/ca4248481ad244f235f4cf277186cad2ee8769f975119a2bbfc41b8932b88bd7...
---------------------------------------------------------------------------
ConnectionError Traceback (most recent call last)
<ipython-input-8-66bf1242096e> in <module>()
----> 1 load_dataset("trec")
10 frames
/usr/local/lib/python3.6/dist-packages/datasets/utils/file_utils.py in get_from_cache(url, cache_dir, force_download, proxies, etag_timeout, resume_download, user_agent, local_files_only, use_etag)
473 elif response is not None and response.status_code == 404:
474 raise FileNotFoundError("Couldn't find file at {}".format(url))
--> 475 raise ConnectionError("Couldn't reach {}".format(url))
476
477 # Try a second time
ConnectionError: Couldn't reach http://cogcomp.org/Data/QA/QC/train_5500.label
</details> | 737 |
https://github.com/huggingface/datasets/issues/735 | Throw error when an unexpected key is used in data_files | [
"Thanks for reporting !\r\nWe'll add support for other keys"
] | I have found that only "train", "validation" and "test" are valid keys in the `data_files` argument. When you use any other ones, those attached files are silently ignored - leading to unexpected behaviour for the users.
So the following, unintuitively, returns only one key (namely `train`).
```python
datasets = load_dataset("text", data_files={"train": train_f, "valid": valid_f})
print(datasets.keys())
# dict_keys(['train'])
```
whereas using `validation` instead, does return the expected result:
```python
datasets = load_dataset("text", data_files={"train": train_f, "validation": valid_f})
print(datasets.keys())
# dict_keys(['train', 'validation'])
```
I would like to see more freedom in which keys one can use, but if that is not possible at least an error should be thrown when using an unexpected key. | 735 |
https://github.com/huggingface/datasets/issues/730 | Possible caching bug | [
"Thanks for reporting. That's a bug indeed.\r\nApparently only the `data_files` parameter is taken into account right now in `DatasetBuilder._create_builder_config` but it should also be the case for `config_kwargs` (or at least the instantiated `builder_config`)",
"Hi, does this bug be fixed? when I load JSON files, I get the same errors by the command \r\n`!python3 run.py --do_train --task qa --dataset squad-retrain-data/train-v2.0.json --output_dir ./re_trained_model/`\r\n\r\nchange the dateset to load json by refering to https://huggingface.co/docs/datasets/loading.html\r\n`dataset = datasets.load_dataset('json', data_files=args.dataset)`\r\n\r\nErrors:\r\n`Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /root/.cache/huggingface/datasets/json/default-c1e124ad488911b8/0.0.0/45636811569ec4a6630521c18235dfbbab83b7ab572e3393c5ba68ccabe98264...\r\n`",
"```ds = load_dataset(\"csv\", data_files={'train': 'train.csv', 'test': 'test.csv'})```\r\n\r\nGives the output\r\n```Using custom data configuration default-5c8ae7c208631aca```\r\n\r\nand the code hangs there.",
"> `ds = load_dataset(\"csv\", data_files={'train': 'train.csv', 'test': 'test.csv'})`\r\n> \r\n> Gives the output `Using custom data configuration default-5c8ae7c208631aca`\r\n> \r\n> and the code hangs there.\r\n\r\nHave you solved it? I met this problem too!",
"Can you Ctrl+C to kill the process and share the stacktrace here ? It should show at which location in the code it was hanging",
"I had the same issue and solved it by downgrading the datasets version from 2.7.0 -> 2.6.1\r\npip install -q datasets==2.6.1",
"> I had the same issue and solved it by downgrading the datasets version from 2.7.0 -> 2.6.1 pip install -q datasets==2.6.1\r\n\r\nThanks, it works for me"
] | The following code with `test1.txt` containing just "π€π€π€":
```
dataset = datasets.load_dataset('text', data_files=['test1.txt'], split="train", encoding="latin_1")
print(dataset[0])
dataset = datasets.load_dataset('text', data_files=['test1.txt'], split="train", encoding="utf-8")
print(dataset[0])
```
produces this output:
```
Downloading and preparing dataset text/default-15600e4d83254059 (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155...
Dataset text downloaded and prepared to /home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155. Subsequent calls will reuse this data.
{'text': 'Γ°\x9fΒ€\x97Γ°\x9fΒ€\x97Γ°\x9fΒ€\x97'}
Using custom data configuration default
Reusing dataset text (/home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155)
{'text': 'Γ°\x9fΒ€\x97Γ°\x9fΒ€\x97Γ°\x9fΒ€\x97'}
```
Just changing the order (and deleting the temp files):
```
dataset = datasets.load_dataset('text', data_files=['test1.txt'], split="train", encoding="utf-8")
print(dataset[0])
dataset = datasets.load_dataset('text', data_files=['test1.txt'], split="train", encoding="latin_1")
print(dataset[0])
```
produces this:
```
Using custom data configuration default
Downloading and preparing dataset text/default-15600e4d83254059 (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155...
Dataset text downloaded and prepared to /home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155. Subsequent calls will reuse this data.
{'text': 'π€π€π€'}
Using custom data configuration default
Reusing dataset text (/home/arne/.cache/huggingface/datasets/text/default-15600e4d83254059/0.0.0/52cefbb2b82b015d4253f1aeb1e6ee5591124a6491e834acfe1751f765925155)
{'text': 'π€π€π€'}
```
Is it intended that the cache path does not depend on the config entries?
tested with datasets==1.1.2 and python==3.8.5 | 730 |
https://github.com/huggingface/datasets/issues/729 | Better error message when one forgets to call `add_batch` before `compute` | [] | When using metrics, if for some reason a user forgets to call `add_batch` to a metric before `compute` (with no arguments), the error message is a bit cryptic and could probably be made clearer.
## Reproducer
```python
import datasets
import torch
from datasets import Metric
class GatherMetric(Metric):
def _info(self):
return datasets.MetricInfo(
description="description",
citation="citation",
inputs_description="kwargs",
features=datasets.Features({
'predictions': datasets.Value('int64'),
'references': datasets.Value('int64'),
}),
codebase_urls=[],
reference_urls=[],
format='numpy'
)
def _compute(self, predictions, references):
return {"predictions": predictions, "labels": references}
metric = GatherMetric(cache_dir="test-metric")
inputs = torch.randint(0, 2, (1024,))
targets = torch.randint(0, 2, (1024,))
batch_size = 8
for i in range(0, 1024, batch_size):
pass # User forgets to call `add_batch`
result = metric.compute()
```
## Stack trace:
```
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-13-267729d187fa> in <module>
3 pass
4 # metric.add_batch(predictions=inputs[i:i+batch_size], references=targets[i:i+batch_size])
----> 5 result = metric.compute()
~/git/datasets/src/datasets/metric.py in compute(self, *args, **kwargs)
380 if predictions is not None:
381 self.add_batch(predictions=predictions, references=references)
--> 382 self._finalize()
383
384 self.cache_file_name = None
~/git/datasets/src/datasets/metric.py in _finalize(self)
343 elif self.process_id == 0:
344 # Let's acquire a lock on each node files to be sure they are finished writing
--> 345 file_paths, filelocks = self._get_all_cache_files()
346
347 # Read the predictions and references
~/git/datasets/src/datasets/metric.py in _get_all_cache_files(self)
280 filelocks = []
281 for process_id, file_path in enumerate(file_paths):
--> 282 filelock = FileLock(file_path + ".lock")
283 try:
284 filelock.acquire(timeout=self.timeout)
TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'
```
| 729 |
https://github.com/huggingface/datasets/issues/728 | Passing `cache_dir` to a metric does not work | [] | When passing `cache_dir` to a custom metric, the folder is concatenated to itself at some point and this results in a FileNotFoundError:
## Reproducer
```python
import datasets
import torch
from datasets import Metric
class GatherMetric(Metric):
def _info(self):
return datasets.MetricInfo(
description="description",
citation="citation",
inputs_description="kwargs",
features=datasets.Features({
'predictions': datasets.Value('int64'),
'references': datasets.Value('int64'),
}),
codebase_urls=[],
reference_urls=[],
format='numpy'
)
def _compute(self, predictions, references):
return {"predictions": predictions, "labels": references}
metric = GatherMetric(cache_dir="test-metric")
inputs = torch.randint(0, 2, (1024,))
targets = torch.randint(0, 2, (1024,))
batch_size = 8
for i in range(0, 1024, batch_size):
metric.add_batch(predictions=inputs[i:i+batch_size], references=targets[i:i+batch_size])
result = metric.compute()
```
## Stack trace:
```
---------------------------------------------------------------------------
FileNotFoundError Traceback (most recent call last)
~/git/datasets/src/datasets/metric.py in _finalize(self)
349 reader = ArrowReader(path=self.data_dir, info=DatasetInfo(features=self.features))
--> 350 self.data = Dataset(**reader.read_files([{"filename": f} for f in file_paths]))
351 except FileNotFoundError:
~/git/datasets/src/datasets/arrow_reader.py in read_files(self, files, original_instructions)
227 # Prepend path to filename
--> 228 pa_table = self._read_files(files)
229 files = copy.deepcopy(files)
~/git/datasets/src/datasets/arrow_reader.py in _read_files(self, files)
166 for f_dict in files:
--> 167 pa_table: pa.Table = self._get_dataset_from_filename(f_dict)
168 pa_tables.append(pa_table)
~/git/datasets/src/datasets/arrow_reader.py in _get_dataset_from_filename(self, filename_skip_take)
291 )
--> 292 mmap = pa.memory_map(filename)
293 f = pa.ipc.open_stream(mmap)
~/.pyenv/versions/3.7.9/envs/base/lib/python3.7/site-packages/pyarrow/io.pxi in pyarrow.lib.memory_map()
~/.pyenv/versions/3.7.9/envs/base/lib/python3.7/site-packages/pyarrow/io.pxi in pyarrow.lib.MemoryMappedFile._open()
~/.pyenv/versions/3.7.9/envs/base/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()
~/.pyenv/versions/3.7.9/envs/base/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
FileNotFoundError: [Errno 2] Failed to open local file 'test-metric/gather_metric/default/test-metric/gather_metric/default/default_experiment-1-0.arrow'. Detail: [errno 2] No such file or directory
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
<ipython-input-17-e42d43cc981f> in <module>
2 for i in range(0, 1024, batch_size):
3 metric.add_batch(predictions=inputs[i:i+batch_size], references=targets[i:i+batch_size])
----> 4 result = metric.compute()
~/git/datasets/src/datasets/metric.py in compute(self, *args, **kwargs)
380 if predictions is not None:
381 self.add_batch(predictions=predictions, references=references)
--> 382 self._finalize()
383
384 self.cache_file_name = None
~/git/datasets/src/datasets/metric.py in _finalize(self)
351 except FileNotFoundError:
352 raise ValueError(
--> 353 "Error in finalize: another metric instance is already using the local cache file. "
354 "Please specify an experiment_id to avoid colision between distributed metric instances."
355 )
ValueError: Error in finalize: another metric instance is already using the local cache file. Please specify an experiment_id to avoid colision between distributed metric instances.
```
The code works when we remove the `cache_dir=...` from the metric. | 728 |
https://github.com/huggingface/datasets/issues/727 | Parallel downloads progress bar flickers | [] | When there are parallel downloads using the download manager, the tqdm progress bar flickers since all the progress bars are on the same line.
To fix that we could simply specify `position=i` for i=0 to n the number of files to download when instantiating the tqdm progress bar.
Another way would be to have one "master" progress bar that tracks the number of finished downloads, and then one progress bar per process that show the current downloads. | 727 |
https://github.com/huggingface/datasets/issues/726 | "Checksums didn't match for dataset source files" error while loading openwebtext dataset | [
"Hi try, to provide more information please.\r\n\r\nExample code in a colab to reproduce the error, details on what you are trying to do and what you were expected and details on your environment (OS, PyPi packages version).",
"> Hi try, to provide more information please.\r\n> \r\n> Example code in a colab to reproduce the error, details on what you are trying to do and what you were expected and details on your environment (OS, PyPi packages version).\r\n\r\nI have update the description, sorry for the incomplete issue by mistake.",
"Hi, I have manually downloaded the compressed dataset `openwebtext.tar.xz' and use the following command to preprocess the examples:\r\n```\r\n>>> dataset = load_dataset('/home/admin/workspace/datasets/datasets-master/datasets-master/datasets/openwebtext', data_dir='/home/admin/workspace/datasets')\r\nUsing custom data configuration default\r\nDownloading and preparing dataset openwebtext/default (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/admin/.cache/huggingface/datasets/openwebtext/default/0.0.0/5c636399c7155da97c982d0d70ecdce30fbca66a4eb4fc768ad91f8331edac02...\r\nDataset openwebtext downloaded and prepared to /home/admin/.cache/huggingface/datasets/openwebtext/default/0.0.0/5c636399c7155da97c982d0d70ecdce30fbca66a4eb4fc768ad91f8331edac02. Subsequent calls will reuse this data.\r\n>>> len(dataset['train'])\r\n74571\r\n>>>\r\n```\r\nThe size of the pre-processed example file is only 354MB, however the processed bookcorpus dataset is 4.6g. Are there any problems?",
"NonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n\r\ni got this issue when i try to work on my own datasets kindly tell me, from where i can get checksums of train and dev file in my github repo",
"Hi, I got the similar issue for xnli dataset while working on colab with python3.7. \r\n\r\n`nlp.load_dataset(path = 'xnli')`\r\n\r\nThe above command resulted in following issue : \r\n```\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip']\r\n```\r\n\r\nAny idea how to fix this ?",
"Did anyone figure out how to fix this error?",
"Fixed by:\r\n- #2857",
"Says fixed but I'm still getting it. \r\n\r\ncommand:\r\n\r\n dataset = load_dataset(\"ted_talks_iwslt\", language_pair=(\"en\", \"es\"), year=\"2014\",download_mode=\"force_redownload\")\r\n\r\ngot:\r\n\r\nUsing custom data configuration en_es_2014-35a2d3350a0f9823\r\nDownloading and preparing dataset ted_talks_iwslt/en_es_2014 (download: 2.15 KiB, generated: Unknown size, post-processed: Unknown size, total: 2.15 KiB) to /home/ken/.cache/huggingface/datasets/ted_talks_iwslt/en_es_2014-35a2d3350a0f9823/1.1.0/43935b3fe470c753a023642e1f54b068c590847f9928bd3f2ec99f15702ad6a6...\r\nDownloading:\r\n2.21k/? [00:00<00:00, 141kB/s]\r\n\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://drive.google.com/u/0/uc?id=1Cz1Un9p8Xn9IpEMMrg2kXSDt0dnjxc4z&export=download']"
] | Hi,
I have encountered this problem during loading the openwebtext dataset:
```
>>> dataset = load_dataset('openwebtext')
Downloading and preparing dataset openwebtext/plain_text (download: 12.00 GiB, generated: 37.04 GiB, post-processed: Unknown size, total: 49.03 GiB) to /home/admin/.cache/huggingface/datasets/openwebtext/plain_text/1.0.0/5c636399c7155da97c982d0d70ecdce30fbca66a4eb4fc768ad91f8331edac02...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/admin/workspace/anaconda3/envs/torch1.6-py3.7/lib/python3.7/site-packages/datasets/load.py", line 611, in load_dataset
ignore_verifications=ignore_verifications,
File "/home/admin/workspace/anaconda3/envs/torch1.6-py3.7/lib/python3.7/site-packages/datasets/builder.py", line 476, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/admin/workspace/anaconda3/envs/torch1.6-py3.7/lib/python3.7/site-packages/datasets/builder.py", line 536, in _download_and_prepare
self.info.download_checksums, dl_manager.get_recorded_sizes_checksums(), "dataset source files"
File "/home/admin/workspace/anaconda3/envs/torch1.6-py3.7/lib/python3.7/site-packages/datasets/utils/info_utils.py", line 39, in verify_checksums
raise NonMatchingChecksumError(error_msg + str(bad_urls))
datasets.utils.info_utils.NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://zenodo.org/record/3834942/files/openwebtext.tar.xz']
```
I think this problem is caused because the released dataset has changed. Or I should download the dataset manually?
Sorry for release the unfinised issue by mistake. | 726 |
https://github.com/huggingface/datasets/issues/724 | need to redirect /nlp to /datasets and remove outdated info | [
"Should be fixed now: \r\n\r\n\r\n\r\nNot sure I understand what you mean by the second part?\r\n",
"Thank you!\r\n\r\n> Not sure I understand what you mean by the second part?\r\n\r\nCompare the 2:\r\n* https://huggingface.co/datasets/wikihow\r\n* https://huggingface.co/nlp/viewer/?dataset=wikihow&config=all\r\nCan you see the difference? 2nd has formatting, 1st doesn't.\r\n",
"For context, those are two different pages (not an old vs new one), one is from the dataset viewer (you can browse data inside the datasets) while the other is just a basic reference page displayed some metadata about the dataset.\r\n\r\nFor the second one, we'll move to markdown parsing soon, so it'll be formatted better.",
"I understand. I was just flagging the lack of markup issue."
] | It looks like the website still has all the `nlp` data, e.g.: https://huggingface.co/nlp/viewer/?dataset=wikihow&config=all
should probably redirect to: https://huggingface.co/datasets/wikihow
also for some reason the new information is slightly borked. If you look at the old one it was nicely formatted and had the links marked up, the new one is just a jumble of text in one chunk and no markup for links (i.e. not clickable). | 724 |
https://github.com/huggingface/datasets/issues/723 | Adding pseudo-labels to datasets | [
"Nice ! :)\r\nIt's indeed the first time we have such contributions so we'll have to figure out the appropriate way to integrate them.\r\nCould you add details on what they could be used for ?\r\n",
"They can be used as training data for a smaller model.",
"Sounds just like a regular dataset to me then, no?",
"A new configuration for those datasets should do the job then.\r\nNote that until now datasets like xsum only had one configuration. It means that users didn't have to specify the configuration name when loading the dataset. If we add new configs, users that update the lib will have to update their code to specify the default/standard configuration name (not the one with pseudo labels).",
"Could also be a `user-namespace` dataset maybe?",
"Oh yes why not. I'm more in favor of this actually since pseudo labels are things that users (not dataset authors in general) can compute by themselves and share with the community",
"\r\n\r\nI assume I should (for example) rename the xsum dir, change the URL, and put the modified dir somewhere in S3?",
"You can use the `datasets-cli` to upload the folder with your version of xsum with the pseudo labels.\r\n\r\n```\r\ndatasets-cli upload_dataset path/to/xsum\r\n```"
] | I recently [uploaded pseudo-labels](https://github.com/huggingface/transformers/blob/master/examples/seq2seq/precomputed_pseudo_labels.md) for CNN/DM, XSUM and WMT16-en-ro to s3, and thom mentioned I should add them to this repo.
Since pseudo-labels are just a large model's generations on an existing dataset, what is the right way to structure this contribution.
I read https://huggingface.co/docs/datasets/add_dataset.html, but it doesn't really cover this type of contribution.
I could, for example, make a new directory, `xsum_bart_pseudolabels` for each set of pseudolabels or add some sort of parametrization to `xsum.py`: https://github.com/huggingface/datasets/blob/5f4c6e830f603830117877b8990a0e65a2386aa6/datasets/xsum/xsum.py
What do you think @lhoestq ?
| 723 |
https://github.com/huggingface/datasets/issues/721 | feat(dl_manager): add support for ftp downloads | [
"We only support http by default for downloading.\r\nIf you really need to use ftp, then feel free to use a library that allows to download through ftp in your dataset script (I see that you've started working on #722 , that's awesome !). The users will get a message to install the extra library when they load the dataset.\r\n\r\nTo make the download_manager work with a custom downloader, you can call `download_manager.download_custom` instead of `download_manager.download_and_extract`. The expected arguments are the following:\r\n```\r\nurl_or_urls: url or `list`/`dict` of urls to download and extract. Each\r\n url is a `str`.\r\ncustom_download: Callable with signature (src_url: str, dst_path: str) -> Any\r\n as for example `tf.io.gfile.copy`, that lets you download from google storage\r\n```\r\n",
"Also maybe it coud be interesting to have a direct support of ftp inside the `datasets` library. Do you know any good libraries that we might consider adding as a (optional ?) dependency ?",
"Downloading an `ftp` file is as simple as:\r\n```python\r\nimport urllib \r\nurllib.urlretrieve('ftp://server/path/to/file', 'file')\r\n```\r\n\r\nI believe this should be supported by the library, as its not using any dependency and is trivial amount of code.",
"I know its unorthodox, but I added `ftp` download support to `file_utils` in the same PR https://github.com/huggingface/datasets/pull/722\r\nSo its possible to understand the interaction of the download component with the ftp download ability",
"Awesome ! I'll take a look :)",
"@AmitMY Can you now download the Phoenix2014 Dataset?",
"@hoanganhpham1006 yes.\r\nSee pull request https://github.com/huggingface/datasets/pull/722 , it has a loader for this dataset, mostly ready.\r\nThere's one issue that delays it being merged - https://github.com/huggingface/datasets/issues/741 - regarding memory consumption.",
"The problem which I have now is that this dataset seems does not allow to download? Can you share it with me pls",
"The dataset loader is not yet ready, because of that issue.\r\nIf you want to just download the dataset the old-fashioned way, just go to: https://www-i6.informatik.rwth-aachen.de/ftp/pub/rwth-phoenix/2016/phoenix-2014-T.v3.tar.gz (the ftp link is now broken, and its available over https)",
"Got it, thank you so much!",
"FTP downloads are supported."
] | I am working on a new dataset (#302) and encounter a problem downloading it.
```python
# This is the official download link from https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/
_URL = "ftp://wasserstoff.informatik.rwth-aachen.de/pub/rwth-phoenix/2016/phoenix-2014-T.v3.tar.gz"
dl_manager.download_and_extract(_URL)
```
I get an error:
> ValueError: unable to parse ftp://wasserstoff.informatik.rwth-aachen.de/pub/rwth-phoenix/2016/phoenix-2014-T.v3.tar.gz as a URL or as a local path
I checked, and indeed you don't consider `ftp` as a remote file.
https://github.com/huggingface/datasets/blob/4c2af707a6955cf4b45f83ac67990395327c5725/src/datasets/utils/file_utils.py#L188
Adding `ftp` to that list does not immediately solve the issue, so there probably needs to be some extra work.
| 721 |
https://github.com/huggingface/datasets/issues/720 | OSError: Cannot find data file when not using the dummy dataset in RAG | [
"Same issue here. I will be digging further, but it looks like the [script](https://github.com/huggingface/datasets/blob/master/datasets/wiki_dpr/wiki_dpr.py#L132) is attempting to open a file that is not downloaded yet. \r\n\r\n```\r\n99dcbca09109e58502e6b9271d4d3f3791b43f61f3161a76b25d2775ab1a4498.lock\r\n```\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nUnpicklingError Traceback (most recent call last)\r\n~/anaconda3/envs/eqa/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)\r\n 446 try:\r\n--> 447 return pickle.load(fid, **pickle_kwargs)\r\n 448 except Exception:\r\n\r\nUnpicklingError: pickle data was truncated\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nOSError Traceback (most recent call last)\r\n~/src/datasets/src/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 559 \r\n--> 560 if verify_infos:\r\n 561 verify_splits(self.info.splits, split_dict)\r\n\r\n~/src/datasets/src/datasets/builder.py in _prepare_split(self, split_generator)\r\n 847 writer.write(example)\r\n--> 848 finally:\r\n 849 num_examples, num_bytes = writer.finalize()\r\n\r\n~/anaconda3/envs/eqa/lib/python3.7/site-packages/tqdm/notebook.py in __iter__(self, *args, **kwargs)\r\n 227 try:\r\n--> 228 for obj in super(tqdm_notebook, self).__iter__(*args, **kwargs):\r\n 229 # return super(tqdm...) will not catch exception\r\n\r\n~/anaconda3/envs/eqa/lib/python3.7/site-packages/tqdm/std.py in __iter__(self)\r\n 1132 try:\r\n-> 1133 for obj in iterable:\r\n 1134 yield obj\r\n\r\n/hdd/rag/cache/huggingface/modules/datasets_modules/datasets/wiki_dpr/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2/wiki_dpr.py in _generate_examples(self, data_file, vectors_files)\r\n 131 break\r\n--> 132 vecs = np.load(open(vectors_files.pop(0), \"rb\"), allow_pickle=True)\r\n 133 vec_idx = 0\r\n\r\n~/anaconda3/envs/eqa/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)\r\n 449 raise IOError(\r\n--> 450 \"Failed to interpret file %s as a pickle\" % repr(file))\r\n 451 \r\n\r\nOSError: Failed to interpret file <_io.BufferedReader name='/hdd/rag/downloads/99dcbca09109e58502e6b9271d4d3f3791b43f61f3161a76b25d2775ab1a4498'> as a pickle\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nOSError Traceback (most recent call last)\r\n<ipython-input-8-24351ff8ce44> in <module>\r\n 4 retriever = RagRetriever.from_pretrained(\"facebook/rag-sequence-nq\", \r\n 5 index_name=\"exact\",\r\n----> 6 use_dummy_dataset=False)\r\n\r\n~/src/transformers/src/transformers/retrieval_rag.py in from_pretrained(cls, retriever_name_or_path, **kwargs)\r\n 321 generator_tokenizer = rag_tokenizer.generator\r\n 322 return cls(\r\n--> 323 config, question_encoder_tokenizer=question_encoder_tokenizer, generator_tokenizer=generator_tokenizer\r\n 324 )\r\n 325 \r\n\r\n~/src/transformers/src/transformers/retrieval_rag.py in __init__(self, config, question_encoder_tokenizer, generator_tokenizer)\r\n 310 self.config = config\r\n 311 if self._init_retrieval:\r\n--> 312 self.init_retrieval()\r\n 313 \r\n 314 @classmethod\r\n\r\n~/src/transformers/src/transformers/retrieval_rag.py in init_retrieval(self)\r\n 338 \r\n 339 logger.info(\"initializing retrieval\")\r\n--> 340 self.index.init_index()\r\n 341 \r\n 342 def postprocess_docs(self, docs, input_strings, prefix, n_docs, return_tensors=None):\r\n\r\n~/src/transformers/src/transformers/retrieval_rag.py in init_index(self)\r\n 248 split=self.dataset_split,\r\n 249 index_name=self.index_name,\r\n--> 250 dummy=self.use_dummy_dataset,\r\n 251 )\r\n 252 self.dataset.set_format(\"numpy\", columns=[\"embeddings\"], output_all_columns=True)\r\n\r\n~/src/datasets/src/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)\r\n 615 builder_instance.download_and_prepare(\r\n 616 download_config=download_config,\r\n--> 617 download_mode=download_mode,\r\n 618 ignore_verifications=ignore_verifications,\r\n 619 )\r\n\r\n~/src/datasets/src/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)\r\n 481 # Sync info\r\n 482 self.info.dataset_size = sum(split.num_bytes for split in self.info.splits.values())\r\n--> 483 self.info.download_checksums = dl_manager.get_recorded_sizes_checksums()\r\n 484 self.info.size_in_bytes = self.info.dataset_size + self.info.download_size\r\n 485 # Save info\r\n\r\n~/src/datasets/src/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 560 if verify_infos:\r\n 561 verify_splits(self.info.splits, split_dict)\r\n--> 562 \r\n 563 # Update the info object with the splits.\r\n 564 self.info.splits = split_dict\r\n\r\nOSError: Cannot find data file.\r\n```\r\n\r\nThank you.",
"An update on my end. This seems like a transient issue. Reran the script from scratch overnight with no errors. ",
"Closing this one. Feel free to re-open if you have other questions about this issue"
] | ## Environment info
transformers version: 3.3.1
Platform: Linux-4.19
Python version: 3.7.7
PyTorch version (GPU?): 1.6.0
Tensorflow version (GPU?): No
Using GPU in script?: Yes
Using distributed or parallel set-up in script?: No
## To reproduce
Steps to reproduce the behaviour:
```
import os
os.environ['HF_DATASETS_CACHE'] = '/workspace/notebooks/POCs/cache'
from transformers import RagTokenizer, RagRetriever, RagTokenForGeneration
tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-nq")
retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=False)
```
Plese note that I'm using the whole dataset: **use_dummy_dataset=False**
After around 4 hours (downloading and some other things) this is returned:
```
Downloading and preparing dataset wiki_dpr/psgs_w100.nq.exact (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /workspace/notebooks/POCs/cache/wiki_dpr/psgs_w100.nq.exact/0.0.0/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2...
---------------------------------------------------------------------------
UnpicklingError Traceback (most recent call last)
/opt/conda/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)
459 try:
--> 460 return pickle.load(fid, **pickle_kwargs)
461 except Exception:
UnpicklingError: pickle data was truncated
During handling of the above exception, another exception occurred:
OSError Traceback (most recent call last)
/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
552 # Prepare split will record examples associated to the split
--> 553 self._prepare_split(split_generator, **prepare_split_kwargs)
554 except OSError:
/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _prepare_split(self, split_generator)
840 for key, record in utils.tqdm(
--> 841 generator, unit=" examples", total=split_info.num_examples, leave=False, disable=not_verbose
842 ):
/opt/conda/lib/python3.7/site-packages/tqdm/notebook.py in __iter__(self, *args, **kwargs)
217 try:
--> 218 for obj in super(tqdm_notebook, self).__iter__(*args, **kwargs):
219 # return super(tqdm...) will not catch exception
/opt/conda/lib/python3.7/site-packages/tqdm/std.py in __iter__(self)
1128 try:
-> 1129 for obj in iterable:
1130 yield obj
~/.cache/huggingface/modules/datasets_modules/datasets/wiki_dpr/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2/wiki_dpr.py in _generate_examples(self, data_file, vectors_files)
131 break
--> 132 vecs = np.load(open(vectors_files.pop(0), "rb"), allow_pickle=True)
133 vec_idx = 0
/opt/conda/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)
462 raise IOError(
--> 463 "Failed to interpret file %s as a pickle" % repr(file))
464 finally:
OSError: Failed to interpret file <_io.BufferedReader name='/workspace/notebooks/POCs/cache/downloads/f34d5f091294259b4ca90e813631e69a6ded660d71b6cbedf89ddba50df94448'> as a pickle
During handling of the above exception, another exception occurred:
OSError Traceback (most recent call last)
<ipython-input-10-f28df370ac47> in <module>
1 # ln -s /workspace/notebooks/POCs/cache /root/.cache/huggingface/datasets
----> 2 retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=False)
/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in from_pretrained(cls, retriever_name_or_path, **kwargs)
307 generator_tokenizer = rag_tokenizer.generator
308 return cls(
--> 309 config, question_encoder_tokenizer=question_encoder_tokenizer, generator_tokenizer=generator_tokenizer
310 )
311
/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in __init__(self, config, question_encoder_tokenizer, generator_tokenizer)
298 self.config = config
299 if self._init_retrieval:
--> 300 self.init_retrieval()
301
302 @classmethod
/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in init_retrieval(self)
324
325 logger.info("initializing retrieval")
--> 326 self.index.init_index()
327
328 def postprocess_docs(self, docs, input_strings, prefix, n_docs, return_tensors=None):
/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in init_index(self)
238 split=self.dataset_split,
239 index_name=self.index_name,
--> 240 dummy=self.use_dummy_dataset,
241 )
242 self.dataset.set_format("numpy", columns=["embeddings"], output_all_columns=True)
/opt/conda/lib/python3.7/site-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)
609 download_config=download_config,
610 download_mode=download_mode,
--> 611 ignore_verifications=ignore_verifications,
612 )
613
/opt/conda/lib/python3.7/site-packages/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
474 if not downloaded_from_gcs:
475 self._download_and_prepare(
--> 476 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
477 )
478 # Sync info
/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
553 self._prepare_split(split_generator, **prepare_split_kwargs)
554 except OSError:
--> 555 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
556
557 if verify_infos:
OSError: Cannot find data file.
```
Thanks
| 720 |
https://github.com/huggingface/datasets/issues/712 | Error in the notebooks/Overview.ipynb notebook | [
"Do this:\r\n``` python\r\nsquad_dataset = list_datasets(with_details=True)[datasets.index('squad')]\r\npprint(squad_dataset.__dict__) # It's a simple python dataclass\r\n```",
"Thanks! This worked. I have created a PR to fix this in the notebook. "
] | Hi,
I got the following error in **cell number 3** while exploring the **Overview.ipynb** notebook in google colab. I used the [link ](https://colab.research.google.com/github/huggingface/datasets/blob/master/notebooks/Overview.ipynb) provided in the main README file to open it in colab.
```python
# You can access various attributes of the datasets before downloading them
squad_dataset = list_datasets()[datasets.index('squad')]
pprint(squad_dataset.__dict__) # It's a simple python dataclass
```
Error message
```
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-5-8dc805c4949c> in <module>()
2 squad_dataset = list_datasets()[datasets.index('squad')]
3
----> 4 pprint(squad_dataset.__dict__) # It's a simple python dataclass
AttributeError: 'str' object has no attribute '__dict__'
```
The object `squad_dataset` is a `str` not a `dataclass` . | 712 |
https://github.com/huggingface/datasets/issues/709 | How to use similarity settings other then "BM25" in Elasticsearch index ? | [
"Datasets does not use elasticsearch API to define custom similarity. If you want to use a custom similarity, the best would be to run a curl request directly to your elasticsearch instance (see sample hereafter, directly from ES documentation), then you should be able to use `my_similarity` in your configuration passed to datasets\r\n\r\n```\r\ncurl -X PUT \"localhost:9200/index?pretty\" -H 'Content-Type: application/json' -d'\r\n{\r\n \"settings\": {\r\n \"index\": {\r\n \"similarity\": {\r\n \"my_similarity\": {\r\n \"type\": \"DFR\",\r\n \"basic_model\": \"g\",\r\n \"after_effect\": \"l\",\r\n \"normalization\": \"h2\",\r\n \"normalization.h2.c\": \"3.0\"\r\n }\r\n }\r\n }\r\n }\r\n}\r\n'\r\n\r\n```"
] | **QUESTION : How should we use other similarity algorithms supported by Elasticsearch other than "BM25" ?**
**ES Reference**
https://www.elastic.co/guide/en/elasticsearch/reference/current/index-modules-similarity.html
**HF doc reference:**
https://huggingface.co/docs/datasets/faiss_and_ea.html
**context :**
========
I used the latest Elasticsearch server version 7.9.2
When I set DFR which is one of the other similarity algorithms supported by elasticsearch in the mapping, I get an error
For example DFR that I had tried in the first instance in mappings as below.,
`"mappings": {"properties": {"text": {"type": "text", "analyzer": "standard", "similarity": "DFR"}}},`
I get the following error
RequestError: RequestError(400, 'mapper_parsing_exception', 'Unknown Similarity type [DFR] for field [text]')
The other thing as another option I had tried was to declare "similarity": "my_similarity" within settings and then assigning "my_similarity" inside the mappings as below
`es_config = {
"settings": {
"number_of_shards": 1,
**"similarity": "my_similarity"**: {
"type": "DFR",
"basic_model": "g",
"after_effect": "l",
"normalization": "h2",
"normalization.h2.c": "3.0"
} ,
"analysis": {"analyzer": {"stop_standard": {"type": "standard", " stopwords": "_english_"}}},
},
"mappings": {"properties": {"text": {"type": "text", "analyzer": "standard", "similarity": "my_similarity"}}},
}`
For this , I got the following error
RequestError: RequestError(400, 'illegal_argument_exception', 'unknown setting [index.similarity] please check that any required plugins are installed, or check the breaking changes documentation for removed settings')
| 709 |
https://github.com/huggingface/datasets/issues/708 | Datasets performance slow? - 6.4x slower than in memory dataset | [
"Facing a similar issue here. My model using SQuAD dataset takes about 1h to process with in memory data and more than 2h with datasets directly.",
"And if you use in-memory-data with datasets with `load_dataset(..., keep_in_memory=True)`?",
"Thanks for the tip @thomwolf ! I did not see that flag in the docs. I'll try with that.",
"We should add it indeed and also maybe a specific section with all the tips for maximal speed. What do you think @lhoestq @SBrandeis @yjernite ?",
"By default the datasets loaded with `load_dataset` live on disk.\r\nIt's possible to load them in memory by using some transforms like `.map(..., keep_in_memory=True)`.\r\n\r\nSmall correction to @thomwolf 's comment above: currently we don't have the `keep_in_memory` parameter for `load_dataset` AFAIK but it would be nice to add it indeed :)",
"Yes indeed we should add it!",
"Great! Thanks a lot.\r\n\r\nI did a test using `map(..., keep_in_memory=True)` and also a test using in-memory only data.\r\n\r\n```python\r\nfeatures = dataset.map(tokenize, batched=True, remove_columns=dataset['train'].column_names)\r\nfeatures.set_format(type='torch', columns=['input_ids', 'token_type_ids', 'attention_mask'])\r\n\r\nfeatures_in_memory = dataset.map(tokenize, batched=True, keep_in_memory=True, remove_columns=dataset['train'].column_names)\r\nfeatures_in_memory.set_format(type='torch', columns=['input_ids', 'token_type_ids', 'attention_mask'])\r\n\r\nin_memory = [features['train'][i] for i in range(len(features['train']))]\r\n```\r\n\r\nFor using the features without any tweak, I got **1min17s** for copying the entire DataLoader to CUDA:\r\n\r\n```\r\n%%time\r\n\r\nfor i, batch in enumerate(DataLoader(features['train'], batch_size=16, num_workers=4)):\r\n batch['input_ids'].to(device)\r\n```\r\n\r\nFor using the features mapped with `keep_in_memory=True`, I also got **1min17s** for copying the entire DataLoader to CUDA:\r\n\r\n```\r\n%%time\r\n\r\nfor i, batch in enumerate(DataLoader(features_in_memory['train'], batch_size=16, num_workers=4)):\r\n batch['input_ids'].to(device)\r\n```\r\n\r\nAnd for the case using every element in memory, converted from the original dataset, I got **12.5s**:\r\n\r\n```\r\n%%time\r\n\r\nfor i, batch in enumerate(DataLoader(in_memory, batch_size=16, num_workers=4)):\r\n batch['input_ids'].to(device)\r\n```\r\n\r\nTaking a closer look in my SQuAD code, using a profiler, I see a lot of calls to `posix read` api. It seems that it is really reliying on disk, which results in a very high train time.",
"I am having the same issue here. When loading from memory I can get the GPU up to 70% util but when loading after mapping I can only get 40%.\r\n\r\nIn disk:\r\n```\r\nbook_corpus = load_dataset('bookcorpus', 'plain_text', cache_dir='/home/ad/Desktop/bookcorpus', split='train[:20%]')\r\nbook_corpus = book_corpus.map(encode, batched=True, num_proc=20, load_from_cache_file=True, batch_size=2500)\r\nbook_corpus.set_format(type='torch', columns=['text', \"input_ids\", \"attention_mask\", \"token_type_ids\"])\r\n\r\ntraining_args = TrainingArguments(\r\n output_dir=\"./mobile_bert_big\",\r\n overwrite_output_dir=True,\r\n num_train_epochs=1,\r\n per_device_train_batch_size=32,\r\n per_device_eval_batch_size=16,\r\n save_steps=50,\r\n save_total_limit=2,\r\n logging_first_step=True,\r\n warmup_steps=100,\r\n logging_steps=50,\r\n eval_steps=100,\r\n no_cuda=False,\r\n gradient_accumulation_steps=16,\r\n fp16=True)\r\n\r\ntrainer = Trainer(\r\n model=model,\r\n args=training_args,\r\n data_collator=data_collator,\r\n train_dataset=book_corpus,\r\n tokenizer=tokenizer)\r\n```\r\n\r\nIn disk I can only get 0,17 it/s:\r\n`[ 13/28907 01:03 < 46:03:27, 0.17 it/s, Epoch 0.00/1] `\r\n\r\nIf I load it with torch.utils.data.Dataset()\r\n```\r\nclass BCorpusDataset(torch.utils.data.Dataset):\r\n def __init__(self, encodings):\r\n self.encodings = encodings\r\n\r\n def __getitem__(self, idx):\r\n item = [torch.tensor(val[idx]) for key, val in self.encodings.items()][0]\r\n return item\r\n\r\n def __len__(self):\r\n length = [len(val) for key, val in self.encodings.items()][0]\r\n return length\r\n\r\n**book_corpus = book_corpus.select([i for i in range(16*2000)])** # filtering to not have 20% of BC in memory...\r\nbook_corpus = book_corpus(book_corpus)\r\n```\r\nI can get:\r\n` [ 5/62 00:09 < 03:03, 0.31 it/s, Epoch 0.06/1]`\r\n\r\nBut obviously I can not get BookCorpus in memory xD\r\n\r\nEDIT: it is something weird. If i load in disk 1% of bookcorpus:\r\n```\r\nbook_corpus = load_dataset('bookcorpus', 'plain_text', cache_dir='/home/ad/Desktop/bookcorpus', split='train[:1%]')\r\n```\r\n\r\nI can get 0.28 it/s, (the same that in memory) but if I load 20% of bookcorpus:\r\n```\r\nbook_corpus = load_dataset('bookcorpus', 'plain_text', cache_dir='/home/ad/Desktop/bookcorpus', split='train[:20%]')\r\n```\r\nI get again 0.17 it/s. \r\n\r\nI am missing something? I think it is something related to size, and not disk or in-memory.",
"There is a way to increase the batches read from memory? or multiprocessed it? I think that one of two or it is reading with just 1 core o it is reading very small chunks from disk and left my GPU at 0 between batches",
"My fault! I had not seen the `dataloader_num_workers` in `TrainingArguments` ! Now I can parallelize and go fast! Sorry, and thanks."
] | I've been very excited about this amazing datasets project. However, I've noticed that the performance can be substantially slower than using an in-memory dataset.
Now, this is expected I guess, due to memory mapping data using arrow files, and you don't get anything for free. But I was surprised at how much slower.
For example, in the `yelp_polarity` dataset (560000 datapoints, or 17500 batches of 32), it was taking me 3:31 to just get process the data and get it on the GPU (no model involved). Whereas, the equivalent in-memory dataset would finish in just 0:33.
Is this expected? Given that one of the goals of this project is also accelerate dataset processing, this seems a bit slower than I would expect. I understand the advantages of being able to work on datasets that exceed memory, and that's very exciting to me, but thought I'd open this issue to discuss.
For reference I'm running a AMD Ryzen Threadripper 1900X 8-Core Processor CPU, with 128 GB of RAM and an NVME SSD Samsung 960 EVO. I'm running with an RTX Titan 24GB GPU.
I can see with `iotop` that the dataset gets quickly loaded into the system read buffers, and thus doesn't incur any additional IO reads. Thus in theory, all the data *should* be in RAM, but in my benchmark code below it's still 6.4 times slower.
What am I doing wrong? And is there a way to force the datasets to completely load into memory instead of being memory mapped in cases where you want maximum performance?
At 3:31 for 17500 batches, that's 12ms per batch. Does this 12ms just become insignificant as a proportion of forward and backward passes in practice, and thus it's not worth worrying about this in practice?
In any case, here's my code `benchmark.py`. If you run it with an argument of `memory` it will copy the data into memory before executing the same test.
``` py
import sys
from datasets import load_dataset
from transformers import DataCollatorWithPadding, BertTokenizerFast
from torch.utils.data import DataLoader
from tqdm import tqdm
if __name__ == '__main__':
tokenizer = BertTokenizerFast.from_pretrained('bert-base-cased')
collate_fn = DataCollatorWithPadding(tokenizer, padding=True)
ds = load_dataset('yelp_polarity')
def do_tokenize(x):
return tokenizer(x['text'], truncation=True)
ds = ds.map(do_tokenize, batched=True)
ds.set_format('torch', ['input_ids', 'token_type_ids', 'attention_mask'])
if len(sys.argv) == 2 and sys.argv[1] == 'memory':
# copy to memory - probably a faster way to do this - but demonstrates the point
# approximately 530 batches per second - 17500 batches in 0:33
print('using memory')
_ds = [data for data in tqdm(ds['train'])]
else:
# approximately 83 batches per second - 17500 batches in 3:31
print('using datasets')
_ds = ds['train']
dl = DataLoader(_ds, shuffle=True, collate_fn=collate_fn, batch_size=32, num_workers=4)
for data in tqdm(dl):
for k, v in data.items():
data[k] = v.to('cuda')
```
For reference, my conda environment is [here](https://gist.github.com/05b6101518ff70ed42a858b302a0405d)
Once again, I'm very excited about this library, and how easy it is to load datasets, and to do so without worrying about system memory constraints.
Thanks for all your great work.
| 708 |
https://github.com/huggingface/datasets/issues/707 | Requirements should specify pyarrow<1 | [
"Hello @mathcass I would want to work on this issue. May I do the same? ",
"@punitaojha, certainly. Feel free to work on this. Let me know if you need any help or clarity.",
"Hello @mathcass \r\n1. I did fork the repository and clone the same on my local system. \r\n\r\n2. Then learnt about how we can publish our package on pypi.org. Also, found some instructions on same in setup.py documentation.\r\n\r\n3. Then I Perplexity document link that you shared above. I created a colab link from there keep both tensorflow and pytorch means a mixed option and tried to run it in colab but I encountered no errors at a point where you mentioned. Can you help me to figure out the issue. \r\n\r\n4.Here is the link of the colab file with my saved responses. \r\nhttps://colab.research.google.com/drive/1hfYz8Ira39FnREbxgwa_goZWpOojp2NH?usp=sharing",
"Also, please share some links which made you conclude that pyarrow < 1 would help. ",
"Access granted for the colab link. ",
"Thanks for looking at this @punitaojha and thanks for sharing the notebook. \r\n\r\nI just tried to reproduce this on my own (based on the environment where I had this issue) and I can't reproduce it somehow. If I run into this again, I'll include some steps to reproduce it. I'll close this as invalid. \r\n\r\nThanks again. ",
"I am sorry for hijacking this closed issue, but I believe I was able to reproduce this very issue. Strangely enough, it also turned out that running `pip install \"pyarrow<1\" --upgrade` did indeed fix the issue (PyArrow was installed in version `0.14.1` in my case).\r\n\r\nPlease see the Colab below:\r\n\r\nhttps://colab.research.google.com/drive/15QQS3xWjlKW2aK0J74eEcRFuhXUddUST\r\n\r\nThanks!"
] | I was looking at the docs on [Perplexity](https://huggingface.co/transformers/perplexity.html) via GPT2. When you load datasets and try to load Wikitext, you get the error,
```
module 'pyarrow' has no attribute 'PyExtensionType'
```
I traced it back to datasets having installed PyArrow 1.0.1 but there's not pinning in the setup file.
https://github.com/huggingface/datasets/blob/e86a2a8f869b91654e782c9133d810bb82783200/setup.py#L68
Downgrading by installing `pip install "pyarrow<1"` resolved the issue. | 707 |
https://github.com/huggingface/datasets/issues/705 | TypeError: '<' not supported between instances of 'NamedSplit' and 'NamedSplit' | [
"Hi !\r\nThanks for reporting :) \r\nIndeed this is an issue on the `datasets` side.\r\nI'm creating a PR",
"Thanks @lhoestq !"
] | ## Environment info
<!-- You can run the command `transformers-cli env` and copy-and-paste its output below.
Don't forget to fill out the missing fields in that output! -->
- `transformers` version: 3.3.1 (installed from master)
- `datasets` version: 1.0.2 (installed as a dependency from transformers)
- Platform: Linux-4.15.0-118-generic-x86_64-with-debian-stretch-sid
- Python version: 3.7.9
I'm testing my own text classification dataset using [this example](https://github.com/huggingface/transformers/tree/master/examples/text-classification#run-generic-text-classification-script-in-tensorflow) from transformers. The dataset is split into train / dev / test, and in csv format, containing just a text and a label columns, using comma as sep. Here's a sample:
```
text,label
"Registra-se a presenΓ§a do acadΓͺmico <name> . <REL_SEP> Ao me deparar com a descrição de dois autores no polo ativo da ação junto ao PJe , margem esquerda foi informado pela procuradora do reclamante que se trata de uma reclamação trabalhista individual . <REL_SEP> Diante disso , face a ausΓͺncia injustificada do autor <name> , determina-se o ARQUIVAMENTO do presente processo , com relação a este , nos termos do [[ art . 844 da CLT ]] . <REL_SEP> CUSTAS AUTOR - DISPENSADO <REL_SEP> Custas pelo autor no importe de R $326,82 , calculadas sobre R $16.341,03 , dispensadas na forma da lei , em virtude da concessΓ£o dos benefΓcios da JustiΓ§a Gratuita , ora deferida . <REL_SEP> Cientes os presentes . <REL_SEP> AudiΓͺncia encerrada Γ s 8h42min . <REL_SEP> <name> <REL_SEP> JuΓza do Trabalho <REL_SEP> Ata redigida por << <name> >> , SecretΓ‘rio de AudiΓͺncia .",NO_RELATION
```
However, @Santosh-Gupta reported in #7351 that he had the exact same problem using the ChemProt dataset. His colab notebook is referenced in the following section.
## To reproduce
Steps to reproduce the behavior:
1. Created a new conda environment using conda env -n transformers python=3.7
2. Cloned transformers master, `cd` into it and installed using pip install --editable . -r examples/requirements.txt
3. Installed tensorflow with `pip install tensorflow`
3. Ran `run_tf_text_classification.py` with the following parameters:
```
--train_file <DATASET_PATH>/train.csv \
--dev_file <DATASET_PATH>/dev.csv \
--test_file <DATASET_PATH>/test.csv \
--label_column_id 1 \
--model_name_or_path neuralmind/bert-base-portuguese-cased \
--output_dir <OUTPUT_PATH> \
--num_train_epochs 4 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 4 \
--do_train \
--do_eval \
--do_predict \
--logging_steps 1000 \
--evaluate_during_training \
--save_steps 1000 \
--overwrite_output_dir \
--overwrite_cache
```
I have also copied [@Santosh-Gupta 's colab notebook](https://colab.research.google.com/drive/11APei6GjphCZbH5wD9yVlfGvpIkh8pwr?usp=sharing) as a reference.
<!-- If you have code snippets, error messages, stack traces please provide them here as well.
Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.-->
Here is the stack trace:
```
2020-10-02 07:33:41.622011: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1
/media/discoD/repositorios/transformers_pedro/src/transformers/training_args.py:333: FutureWarning: The `evaluate_during_training` argument is deprecated in favor of `evaluation_strategy` (which has more options)
FutureWarning,
2020-10-02 07:33:43.471648: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcuda.so.1
2020-10-02 07:33:43.471791: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.472664: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1070 computeCapability: 6.1
coreClock: 1.7085GHz coreCount: 15 deviceMemorySize: 7.92GiB deviceMemoryBandwidth: 238.66GiB/s
2020-10-02 07:33:43.472684: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1
2020-10-02 07:33:43.472765: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcublas.so.10
2020-10-02 07:33:43.472809: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcufft.so.10
2020-10-02 07:33:43.472848: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcurand.so.10
2020-10-02 07:33:43.474209: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusolver.so.10
2020-10-02 07:33:43.474276: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusparse.so.10
2020-10-02 07:33:43.561219: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7
2020-10-02 07:33:43.561397: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.562345: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.563219: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1858] Adding visible gpu devices: 0
2020-10-02 07:33:43.563595: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2020-10-02 07:33:43.570091: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 3591830000 Hz
2020-10-02 07:33:43.570494: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x560842432400 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-10-02 07:33:43.570511: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-10-02 07:33:43.570702: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.571599: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1070 computeCapability: 6.1
coreClock: 1.7085GHz coreCount: 15 deviceMemorySize: 7.92GiB deviceMemoryBandwidth: 238.66GiB/s
2020-10-02 07:33:43.571633: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1
2020-10-02 07:33:43.571645: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcublas.so.10
2020-10-02 07:33:43.571654: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcufft.so.10
2020-10-02 07:33:43.571664: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcurand.so.10
2020-10-02 07:33:43.571691: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusolver.so.10
2020-10-02 07:33:43.571704: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusparse.so.10
2020-10-02 07:33:43.571718: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7
2020-10-02 07:33:43.571770: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.572641: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:43.573475: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1858] Adding visible gpu devices: 0
2020-10-02 07:33:47.139227: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1257] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-10-02 07:33:47.139265: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1263] 0
2020-10-02 07:33:47.139272: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1276] 0: N
2020-10-02 07:33:47.140323: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:47.141248: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:47.142085: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-02 07:33:47.142854: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1402] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 5371 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0, compute capability: 6.1)
2020-10-02 07:33:47.146317: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x5608b95dc5c0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-10-02 07:33:47.146336: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce GTX 1070, Compute Capability 6.1
10/02/2020 07:33:47 - INFO - __main__ - n_replicas: 1, distributed training: False, 16-bits training: False
10/02/2020 07:33:47 - INFO - __main__ - Training/evaluation parameters TFTrainingArguments(output_dir='/media/discoD/models/datalawyer/pedidos/transformers_tf', overwrite_output_dir=True, do_train=True, do_eval=True, do_predict=True, evaluate_during_training=True, evaluation_strategy=<EvaluationStrategy.STEPS: 'steps'>, prediction_loss_only=False, per_device_train_batch_size=4, per_device_eval_batch_size=4, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=1, learning_rate=5e-05, weight_decay=0.0, adam_beta1=0.9, adam_beta2=0.999, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=4.0, max_steps=-1, warmup_steps=0, logging_dir='runs/Oct02_07-33-43_user-XPS-8700', logging_first_step=False, logging_steps=1000, save_steps=1000, save_total_limit=None, no_cuda=False, seed=42, fp16=False, fp16_opt_level='O1', local_rank=-1, tpu_num_cores=None, tpu_metrics_debug=False, debug=False, dataloader_drop_last=False, eval_steps=1000, dataloader_num_workers=0, past_index=-1, run_name='/media/discoD/models/datalawyer/pedidos/transformers_tf', disable_tqdm=False, remove_unused_columns=True, label_names=None, load_best_model_at_end=False, metric_for_best_model=None, greater_is_better=False, tpu_name=None, xla=False)
10/02/2020 07:33:53 - INFO - filelock - Lock 140407857405776 acquired on /home/user/.cache/huggingface/datasets/e0f1e9ed46db1e2429189f06b479cbd4075c0976104c1aacf8f77d9a53d2ad87.03756fef6da334f50a7ff73608e21b5018229944ca250416ce7352e25d84a552.py.lock
10/02/2020 07:33:53 - INFO - filelock - Lock 140407857405776 released on /home/user/.cache/huggingface/datasets/e0f1e9ed46db1e2429189f06b479cbd4075c0976104c1aacf8f77d9a53d2ad87.03756fef6da334f50a7ff73608e21b5018229944ca250416ce7352e25d84a552.py.lock
Using custom data configuration default
Traceback (most recent call last):
File "run_tf_text_classification.py", line 283, in <module>
main()
File "run_tf_text_classification.py", line 222, in main
max_seq_length=data_args.max_seq_length,
File "run_tf_text_classification.py", line 43, in get_tfds
ds = datasets.load_dataset("csv", data_files=files)
File "/media/discoD/anaconda3/envs/transformers/lib/python3.7/site-packages/datasets/load.py", line 604, in load_dataset
**config_kwargs,
File "/media/discoD/anaconda3/envs/transformers/lib/python3.7/site-packages/datasets/builder.py", line 158, in __init__
**config_kwargs,
File "/media/discoD/anaconda3/envs/transformers/lib/python3.7/site-packages/datasets/builder.py", line 269, in _create_builder_config
for key in sorted(data_files.keys()):
TypeError: '<' not supported between instances of 'NamedSplit' and 'NamedSplit'
```
## Expected behavior
Should be able to run the text-classification example as described in [https://github.com/huggingface/transformers/tree/master/examples/text-classification#run-generic-text-classification-script-in-tensorflow](https://github.com/huggingface/transformers/tree/master/examples/text-classification#run-generic-text-classification-script-in-tensorflow)
Originally opened this issue at transformers' repository: [https://github.com/huggingface/transformers/issues/7535](https://github.com/huggingface/transformers/issues/7535). @jplu instructed me to open here, since according to [this](https://github.com/huggingface/transformers/issues/7535#issuecomment-702778885) evidence, the problem is from datasets.
Thanks! | 705 |
https://github.com/huggingface/datasets/issues/699 | XNLI dataset is not loading | [
"also i tried below code to solve checksum error \r\n`datasets-cli test ./datasets/xnli --save_infos --all_configs`\r\n\r\nand it shows \r\n\r\n```\r\n2020-10-02 07:06:16.588760: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1\r\nTraceback (most recent call last):\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/load.py\", line 268, in prepare_module\r\n local_path = cached_path(file_path, download_config=download_config)\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/utils/file_utils.py\", line 308, in cached_path\r\n use_etag=download_config.use_etag,\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/utils/file_utils.py\", line 474, in get_from_cache\r\n raise FileNotFoundError(\"Couldn't find file at {}\".format(url))\r\nFileNotFoundError: Couldn't find file at https://raw.githubusercontent.com/huggingface/datasets/1.0.2/datasets/./datasets/xnli/xnli.py\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/load.py\", line 279, in prepare_module\r\n local_path = cached_path(file_path, download_config=download_config)\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/utils/file_utils.py\", line 308, in cached_path\r\n use_etag=download_config.use_etag,\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/utils/file_utils.py\", line 474, in get_from_cache\r\n raise FileNotFoundError(\"Couldn't find file at {}\".format(url))\r\nFileNotFoundError: Couldn't find file at https://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/./datasets/xnli/xnli.py\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"/opt/conda/bin/datasets-cli\", line 36, in <module>\r\n service.run()\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/commands/test.py\", line 76, in run\r\n module_path, hash = prepare_module(path)\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/load.py\", line 283, in prepare_module\r\n combined_path, github_file_path, file_path\r\nFileNotFoundError: Couldn't find file locally at ./datasets/xnli/xnli.py, or remotely at https://raw.githubusercontent.com/huggingface/datasets/1.0.2/datasets/./datasets/xnli/xnli.py or https://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/./datasets/xnli/xnli.py\r\n```\r\n\r\n",
"Hi !\r\nYes the download url changed.\r\nIt's updated on the master branch. I'm doing a release today to fix that :)",
"the issue is fixed with latest release \r\n\r\n"
] | `dataset = datasets.load_dataset(path='xnli')`
showing below error
```
/opt/conda/lib/python3.7/site-packages/nlp/utils/info_utils.py in verify_checksums(expected_checksums, recorded_checksums, verification_name)
36 if len(bad_urls) > 0:
37 error_msg = "Checksums didn't match" + for_verification_name + ":\n"
---> 38 raise NonMatchingChecksumError(error_msg + str(bad_urls))
39 logger.info("All the checksums matched successfully" + for_verification_name)
40
NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip']
```
I think URL is now changed to "https://cims.nyu.edu/~sbowman/xnli/XNLI-MT-1.0.zip" | 699 |
https://github.com/huggingface/datasets/issues/691 | Add UI filter to filter datasets based on task | [
"Already supported."
] | This is great work, so huge shoutout to contributors and huggingface.
The [/nlp/viewer](https://huggingface.co/nlp/viewer/) is great and the [/datasets](https://huggingface.co/datasets) page is great. I was wondering if in both or either places we can have a filter that selects if a dataset is good for the following tasks (non exhaustive list)
- Classification
- Multi label
- Multi class
- Q&A
- Summarization
- Translation
I believe this feature might have some value, for folks trying to find datasets for a particular task, and then testing their model capabilities.
Thank you :) | 691 |
https://github.com/huggingface/datasets/issues/690 | XNLI dataset: NonMatchingChecksumError | [
"Thanks for reporting.\r\nThe data file must have been updated by the host.\r\nI'll update the checksum with the new one.",
"Well actually it looks like the link isn't working anymore :(",
"The new link is https://cims.nyu.edu/~sbowman/xnli/XNLI-1.0.zip\r\nI'll update the dataset script",
"I'll do a release in the next few days to make the fix available for everyone.\r\nIn the meantime you can load `xnli` with\r\n```\r\nxnli = load_dataset('xnli', script_version=\"master\")\r\n```\r\nThis will use the latest version of the xnli script (available on master branch), instead of the old one.",
"That's awesome! Thanks a lot!"
] | Hi,
I tried to download "xnli" dataset in colab using
`xnli = load_dataset(path='xnli')`
but got 'NonMatchingChecksumError' error
`NonMatchingChecksumError Traceback (most recent call last)
<ipython-input-27-a87bedc82eeb> in <module>()
----> 1 xnli = load_dataset(path='xnli')
3 frames
/usr/local/lib/python3.6/dist-packages/datasets/utils/info_utils.py in verify_checksums(expected_checksums, recorded_checksums, verification_name)
37 if len(bad_urls) > 0:
38 error_msg = "Checksums didn't match" + for_verification_name + ":\n"
---> 39 raise NonMatchingChecksumError(error_msg + str(bad_urls))
40 logger.info("All the checksums matched successfully" + for_verification_name)
41
NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip']`
The same code worked well several days ago in colab but stopped working now. Thanks! | 690 |
https://github.com/huggingface/datasets/issues/687 | `ArrowInvalid` occurs while running `Dataset.map()` function | [
"Hi !\r\n\r\nThis is because `encode` expects one single text as input (str), or one tokenized text (List[str]).\r\nI believe that you actually wanted to use `encode_batch` which expects a batch of texts.\r\nHowever this method is only available for our \"fast\" tokenizers (ex: BertTokenizerFast).\r\nBertJapanese is not one of them unfortunately and I don't think it will be added for now (see https://github.com/huggingface/transformers/pull/7141)...\r\ncc @thomwolf for confirmation.\r\n\r\nTherefore what I'd suggest for now is disable batching and process one text at a time using `encode`.\r\nNote that you can make it faster by using multiprocessing:\r\n\r\n```python\r\nnum_proc = None # Specify here the number of processes if you want to use multiprocessing. ex: num_proc = 4\r\nencoded = train_ds.map(\r\n lambda example: {'tokens': t.encode(example['title'], max_length=1000)}, num_proc=num_proc\r\n)\r\n```\r\n",
"Thank you very much for the kind and precise suggestion!\r\nI'm looking forward to seeing BertJapaneseTokenizer built into the \"fast\" tokenizers.\r\n\r\nI tried `map` with multiprocessing as follows, and it worked!\r\n\r\n```python\r\n# There was a Pickle problem if I use `lambda` for multiprocessing\r\ndef encode(examples):\r\n return {'tokens': t.encode(examples['title'], max_length=1000)}\r\n\r\nnum_proc = 8\r\nencoded = train_ds.map(encode, num_proc=num_proc)\r\n```\r\n\r\nThank you again!"
] | It seems to fail to process the final batch. This [colab](https://colab.research.google.com/drive/1_byLZRHwGP13PHMkJWo62Wp50S_Z2HMD?usp=sharing) can reproduce the error.
Code:
```python
# train_ds = Dataset(features: {
# 'title': Value(dtype='string', id=None),
# 'score': Value(dtype='float64', id=None)
# }, num_rows: 99999)
# suggested in #665
class PicklableTokenizer(BertJapaneseTokenizer):
def __getstate__(self):
state = dict(self.__dict__)
state['do_lower_case'] = self.word_tokenizer.do_lower_case
state['never_split'] = self.word_tokenizer.never_split
del state['word_tokenizer']
return state
def __setstate(self):
do_lower_case = state.pop('do_lower_case')
never_split = state.pop('never_split')
self.__dict__ = state
self.word_tokenizer = MecabTokenizer(
do_lower_case=do_lower_case, never_split=never_split
)
t = PicklableTokenizer.from_pretrained('bert-base-japanese-whole-word-masking')
encoded = train_ds.map(
lambda examples: {'tokens': t.encode(examples['title'], max_length=1000)}, batched=True, batch_size=1000
)
```
Error Message:
```
99% 99/100 [00:22<00:00, 39.07ba/s]
---------------------------------------------------------------------------
ArrowInvalid Traceback (most recent call last)
<timed exec> in <module>
/usr/local/lib/python3.6/site-packages/datasets/arrow_dataset.py in map(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint)
1242 fn_kwargs=fn_kwargs,
1243 new_fingerprint=new_fingerprint,
-> 1244 update_data=update_data,
1245 )
1246 else:
/usr/local/lib/python3.6/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
151 "output_all_columns": self._output_all_columns,
152 }
--> 153 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
154 if new_format["columns"] is not None:
155 new_format["columns"] = list(set(new_format["columns"]) & set(out.column_names))
/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
161 # Call actual function
162
--> 163 out = func(self, *args, **kwargs)
164
165 # Update fingerprint of in-place transforms + update in-place history of transforms
/usr/local/lib/python3.6/site-packages/datasets/arrow_dataset.py in _map_single(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset, update_data)
1496 if update_data:
1497 batch = cast_to_python_objects(batch)
-> 1498 writer.write_batch(batch)
1499 if update_data:
1500 writer.finalize() # close_stream=bool(buf_writer is None)) # We only close if we are writing in a file
/usr/local/lib/python3.6/site-packages/datasets/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
271 typed_sequence = TypedSequence(batch_examples[col], type=col_type, try_type=col_try_type)
272 typed_sequence_examples[col] = typed_sequence
--> 273 pa_table = pa.Table.from_pydict(typed_sequence_examples)
274 self.write_table(pa_table)
275
/usr/local/lib/python3.6/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.from_pydict()
/usr/local/lib/python3.6/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.from_arrays()
/usr/local/lib/python3.6/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.validate()
/usr/local/lib/python3.6/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowInvalid: Column 4 named tokens expected length 999 but got length 1000
```
| 687 |
https://github.com/huggingface/datasets/issues/686 | Dataset browser url is still https://huggingface.co/nlp/viewer/ | [
"Yes! might do it with @srush one of these days. Hopefully it won't break too many links (we can always redirect from old url to new)",
"This was fixed but forgot to close the issue. cc @lhoestq @yjernite \r\n\r\nThanks @jarednielsen!"
] | Might be worth updating to https://huggingface.co/datasets/viewer/ | 686 |
https://github.com/huggingface/datasets/issues/678 | The download instructions for c4 datasets are not contained in the error message | [
"Good catch !\r\nIndeed the `@property` is missing.\r\n\r\nFeel free to open a PR :)",
"Also not that C4 is a dataset that needs an Apache Beam runtime to be generated.\r\nFor example Dataflow, Spark, Flink etc.\r\n\r\nUsually we generate the dataset on our side once and for all, but we haven't done it for C4 yet.\r\nMore info about beam datasets [here](https://huggingface.co/docs/datasets/beam_dataset.html)\r\n\r\nLet me know if you have any questions"
] | The manual download instructions are not clear
```The dataset c4 with config en requires manual data.
Please follow the manual download instructions: <bound method C4.manual_download_instructions of <datasets_modules.datasets.c4.830b0c218bd41fed439812c8dd19dbd4767d2a3faa385eb695cf8666c982b1b3.c4.C4 object at 0x7ff8c5969760>>.
Manual data can be loaded with `datasets.load_dataset(c4, data_dir='<path/to/manual/data>')
```
Either `@property` could be added to C4.manual_download_instrcutions (or make it a real property), or the manual_download_instructions function needs to be called I think.
Let me know if you want a PR for this, but I'm not sure which possible fix is the correct one. | 678 |
https://github.com/huggingface/datasets/issues/676 | train_test_split returns empty dataset item | [
"The problem still exists after removing the cache files.",
"Can you reproduce this example in a Colab so we can investigate? (or give more information on your software/hardware config)",
"Thanks for reporting.\r\nI just found the issue, I'm creating a PR",
"We'll do a release pretty soon to include the fix :)\r\nIn the meantime you can install the lib from source if you want to "
] | I try to split my dataset by `train_test_split`, but after that the item in `train` and `test` `Dataset` is empty.
The codes:
```
yelp_data = datasets.load_from_disk('/home/ssd4/huanglianzhe/test_yelp')
print(yelp_data[0])
yelp_data = yelp_data.train_test_split(test_size=0.1)
print(yelp_data)
print(yelp_data['test'])
print(yelp_data['test'][0])
```
The outputs:
```
{'stars': 2.0, 'text': 'xxxx'}
Loading cached split indices for dataset at /home/ssd4/huanglianzhe/test_yelp/cache-f9b22d8b9d5a7346.arrow and /home/ssd4/huanglianzhe/test_yelp/cache-4aa26fa4005059d1.arrow
DatasetDict({'train': Dataset(features: {'stars': Value(dtype='float64', id=None), 'text': Value(dtype='string', id=None)}, num_rows: 7219009), 'test': Dataset(features: {'stars': Value(dtype='float64', id=None), 'text': Value(dtype='string', id=None)}, num_rows: 802113)})
Dataset(features: {'stars': Value(dtype='float64', id=None), 'text': Value(dtype='string', id=None)}, num_rows: 802113)
{} # yelp_data['test'][0] is empty
``` | 676 |
https://github.com/huggingface/datasets/issues/675 | Add custom dataset to NLP? | [
"Yes you can have a look here: https://huggingface.co/docs/datasets/loading_datasets.html#csv-files",
"No activity, closing"
] | Is it possible to add a custom dataset such as a .csv to the NLP library?
Thanks. | 675 |
https://github.com/huggingface/datasets/issues/674 | load_dataset() won't download in Windows | [
"I have the same issue. Tried to download a few of them and not a single one is downloaded successfully.\r\n\r\nThis is the output:\r\n```\r\n>>> dataset = load_dataset('blended_skill_talk', split='train')\r\nUsing custom data configuration default <-- This step never ends\r\n```",
"This was fixed in #644 \r\nI'll do a new release soon :)\r\n\r\nIn the meantime you can run it by installing from source",
"Closing since version 1.1.0 got released with Windows support :) \r\nLet me know if it works for you now"
] | I don't know if this is just me or Windows. Maybe other Windows users can chime in if they don't have this problem. I've been trying to get some of the tutorials working on Windows, but when I use the load_dataset() function, it just stalls and the script keeps running indefinitely without downloading anything. I've waited upwards of 18 hours to download the 'multi-news' dataset (which isn't very big), and still nothing. I've tried running it through different IDE's and the command line, but it had the same behavior. I've also tried it with all virus and malware protection turned off. I've made sure python and all IDE's are exceptions to the firewall and all the requisite permissions are enabled.
Additionally, I checked to see if other packages could download content such as an nltk corpus, and they could. I've also run the same script using Ubuntu and it downloaded fine (and quickly). When I copied the downloaded datasets from my Ubuntu drive to my Windows .cache folder it worked fine by reusing the already-downloaded dataset, but it's cumbersome to do that for every dataset I want to try in my Windows environment.
Could this be a bug, or is there something I'm doing wrong or not thinking of?
Thanks. | 674 |
https://github.com/huggingface/datasets/issues/673 | blog_authorship_corpus crashed | [
"Thanks for reporting !\r\nWe'll free some memory"
] | This is just to report that When I pick blog_authorship_corpus in
https://huggingface.co/nlp/viewer/?dataset=blog_authorship_corpus
I get this:
| 673 |
https://github.com/huggingface/datasets/issues/672 | Questions about XSUM | [
"We should try to regenerate the data using the official script.\r\nBut iirc that's what we used in the first place, so not sure why it didn't match in the first place.\r\n\r\nI'll let you know when the dataset is updated",
"Thanks, looking forward to hearing your update on this thread. \r\n\r\nThis is a blocking issue for us; would appreciate any progress on this front. We can also help with the fix, if you deem it appropriately. ",
"I just started the generation on my side, I'll let you know how it goes :) ",
"Hmm after a first run I'm still missing 136668/226711 urls.\r\nI'll relaunch it tomorrow to try to get the remaining ones.",
"Update: I'm missing 36/226711 urls but I haven't managed to download them yet",
"Thanks! That sounds like a reasonable number! ",
"So I managed to download them all but when parsing only 226,181/226,711 worked.\r\nNot sure if it's worth digging and debugging parsing at this point :/ ",
"Maybe @sshleifer can help, I think he's already played with xsum at one point",
"Thanks @lhoestq\r\nIt would be great to improve coverage, but IDs are the really crucial part for us. We'd really appreciate an update to the dataset with IDs either way!",
"I gave up at an even earlier point. The dataset I use has 204,017 train examples.",
"@lhoestq @sshleifer like @jbragg said earlier, the main issue for us is that the current XSUM dataset (in your package) does not have IDs suggested by the original dataset ([here is the file](https://raw.githubusercontent.com/EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json).) Would appreciate if you update the XSUM dataset to include the instance IDs. \r\n\r\nThe missing instances is also a problem, but likely not worth pursuing given its relatively small scale. ",
">So I managed to download them all but when parsing only 226,181/226,711 worked.\r\n\r\n@lhoestq any chance we could update the HF-hosted dataset with the IDs in your new version? Happy to help if there's something I can do.",
"Well I couldn't parse what I downloaded.\r\nUnfortunately I think I won't be able to take a look at it this week.\r\nI can try to send you what I got if you want to give it a shot @jbragg \r\nOtherwise feel free to re-run the xsum download script, maybe you'll be luckier than me",
"Resolved via #754"
] | Hi there β
I'm looking into your `xsum` dataset and I have several questions on that.
So here is how I loaded the data:
```
>>> data = datasets.load_dataset('xsum', version='1.0.1')
>>> data['train']
Dataset(features: {'document': Value(dtype='string', id=None), 'summary': Value(dtype='string', id=None)}, num_rows: 204017)
>>> data['test']
Dataset(features: {'document': Value(dtype='string', id=None), 'summary': Value(dtype='string', id=None)}, num_rows: 11333)
```
The first issue is, the instance counts donβt match what I see on [the dataset's website](https://github.com/EdinburghNLP/XSum/tree/master/XSum-Dataset#what-builds-the-xsum-dataset) (11,333 vs 11,334 for test set; 204,017 vs 204,045 for training set)
```
β¦ training (90%, 204,045), validation (5%, 11,332), and test (5%, 11,334) set.
```
Any thoughts why? Perhaps @mariamabarham could help here, since she recently had a PR on this dataaset https://github.com/huggingface/datasets/pull/289 (reviewed by @patrickvonplaten)
Another issue is that the instances don't seem to have IDs. The original datasets provides IDs for the instances: https://github.com/EdinburghNLP/XSum/blob/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json but to be able to use them, the dataset sizes need to match.
CC @jbragg
| 672 |
https://github.com/huggingface/datasets/issues/671 | [BUG] No such file or directory | [] | This happens when both
1. Huggingface datasets cache dir does not exist
2. Try to load a local dataset script
builder.py throws an error when trying to create a filelock in a directory (cache/datasets) that does not exist
https://github.com/huggingface/datasets/blob/master/src/datasets/builder.py#L177
Tested on v1.0.2
@lhoestq | 671 |
https://github.com/huggingface/datasets/issues/669 | How to skip a example when running dataset.map | [
"Hi @xixiaoyao,\r\nDepending on what you want to do you can:\r\n- use a first step of `filter` to filter out the invalid examples: https://huggingface.co/docs/datasets/processing.html#filtering-rows-select-and-filter\r\n- or directly detect the invalid examples inside the callable used with `map` and return them unchanged or even remove them at the same time if you are using `map` in batched mode. Here is an example where we use `map` in batched mode to add new rows on the fly but you can also use it to remove examples on the fly (that's what `filter` actually do under-the-hood): https://huggingface.co/docs/datasets/processing.html#augmenting-the-dataset",
"Closing this one.\r\nFeel free to re-open if you have other questions",
"Letting finders-of-this-thread know that the new link is: https://huggingface.co/docs/datasets/process#data-augmentation\r\n"
] | in processing func, I process examples and detect some invalid examples, which I did not want it to be added into train dataset. However I did not find how to skip this recognized invalid example when doing dataset.map. | 669 |
https://github.com/huggingface/datasets/issues/668 | OverflowError when slicing with an array containing negative ids | [] | ```python
from datasets import Dataset
d = ds.Dataset.from_dict({"a": range(10)})
print(d[0])
# {'a': 0}
print(d[-1])
# {'a': 9}
print(d[[0, -1]])
# OverflowError
```
results in
```
---------------------------------------------------------------------------
OverflowError Traceback (most recent call last)
<ipython-input-5-863dc3555598> in <module>
----> 1 d[[0, -1]]
~/Desktop/hf/nlp/src/datasets/arrow_dataset.py in __getitem__(self, key)
1070 format_columns=self._format_columns,
1071 output_all_columns=self._output_all_columns,
-> 1072 format_kwargs=self._format_kwargs,
1073 )
1074
~/Desktop/hf/nlp/src/datasets/arrow_dataset.py in _getitem(self, key, format_type, format_columns, output_all_columns, format_kwargs)
1025 indices = key
1026
-> 1027 indices_array = pa.array([int(i) for i in indices], type=pa.uint64())
1028
1029 # Check if we need to convert indices
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/array.pxi in pyarrow.lib.array()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
OverflowError: can't convert negative value to unsigned int
``` | 668 |
https://github.com/huggingface/datasets/issues/667 | Loss not decrease with Datasets and Transformers | [
"And I tested it on T5ForConditionalGeneration, that works no problem.",
"Hi did you manage to fix your issue ?\r\n\r\nIf so feel free to share your fix and close this thread"
] | HI,
The following script is used to fine-tune a BertForSequenceClassification model on SST2.
The script is adapted from [this colab](https://colab.research.google.com/github/huggingface/datasets/blob/master/notebooks/Overview.ipynb) that presents an example of fine-tuning BertForQuestionAnswering using squad dataset. In that colab, loss works fine. When I adapt it to SST2, the loss fails to decrease as it should. I attach the adapted script below and appreciate anyone pointing out what I miss?
```python
import torch
from datasets import load_dataset
from transformers import BertForSequenceClassification
from transformers import BertTokenizerFast
# Load our training dataset and tokenizer
dataset = load_dataset("glue", 'sst2')
tokenizer = BertTokenizerFast.from_pretrained('bert-base-cased')
del dataset["test"] # let's remove it in this demo
# Tokenize our training dataset
def convert_to_features(example_batch):
encodings = tokenizer(example_batch["sentence"])
encodings.update({"labels": example_batch["label"]})
return encodings
encoded_dataset = dataset.map(convert_to_features, batched=True)
# Format our dataset to outputs torch.Tensor to train a pytorch model
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'labels']
encoded_dataset.set_format(type='torch', columns=columns)
# Instantiate a PyTorch Dataloader around our dataset
# Let's do dynamic batching (pad on the fly with our own collate_fn)
def collate_fn(examples):
return tokenizer.pad(examples, return_tensors='pt')
dataloader = torch.utils.data.DataLoader(encoded_dataset['train'], collate_fn=collate_fn, batch_size=8)
# Now let's train our model
device = 'cuda' if torch.cuda.is_available() else 'cpu'
# Let's load a pretrained Bert model and a simple optimizer
model = BertForSequenceClassification.from_pretrained('bert-base-cased', return_dict=True)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-5)
model.train().to(device)
for i, batch in enumerate(dataloader):
batch.to(device)
outputs = model(**batch)
loss = outputs.loss
loss.backward()
optimizer.step()
model.zero_grad()
print(f'Step {i} - loss: {loss:.3}')
```
In case needed.
- datasets == 1.0.2
- transformers == 3.2.0 | 667 |
https://github.com/huggingface/datasets/issues/666 | Does both 'bookcorpus' and 'wikipedia' belong to the same datasets which Google used for pretraining BERT? | [
"No they are other similar copies but they are not provided by the official Bert models authors."
] | 666 |
|
https://github.com/huggingface/datasets/issues/665 | runing dataset.map, it raises TypeError: can't pickle Tokenizer objects | [
"Hi !\r\nIt works on my side with both the LongFormerTokenizer and the LongFormerTokenizerFast.\r\n\r\nWhich version of transformers/datasets are you using ?",
"transformers and datasets are both the latest",
"Then I guess you need to give us more informations on your setup (OS, python, GPU, etc) or a Google Colab reproducing the error for us to be able to debug this error.",
"And your version of `dill` if possible :)",
"I have the same issue with `transformers/BertJapaneseTokenizer`.\r\n\r\n\r\n\r\n```python\r\n# train_ds = Dataset(features: {\r\n# 'title': Value(dtype='string', id=None), \r\n# 'score': Value(dtype='float64', id=None)\r\n# }, num_rows: 99999)\r\n\r\nt = BertJapaneseTokenizer.from_pretrained('bert-base-japanese-whole-word-masking')\r\nencoded = train_ds.map(lambda examples: {'tokens': t.encode(examples['title'])}, batched=True)\r\n```\r\n\r\n<details><summary>Error Message</summary>\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nTypeError Traceback (most recent call last)\r\n<ipython-input-35-2b7d66b291c1> in <module>\r\n 2 \r\n 3 encoded = train_ds.map(lambda examples:\r\n----> 4 {'tokens': t.encode(examples['title'])}, batched=True)\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/arrow_dataset.py in map(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint)\r\n 1242 fn_kwargs=fn_kwargs,\r\n 1243 new_fingerprint=new_fingerprint,\r\n-> 1244 update_data=update_data,\r\n 1245 )\r\n 1246 else:\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)\r\n 151 \"output_all_columns\": self._output_all_columns,\r\n 152 }\r\n--> 153 out: Union[\"Dataset\", \"DatasetDict\"] = func(self, *args, **kwargs)\r\n 154 if new_format[\"columns\"] is not None:\r\n 155 new_format[\"columns\"] = list(set(new_format[\"columns\"]) & set(out.column_names))\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)\r\n 156 kwargs_for_fingerprint[\"fingerprint_name\"] = fingerprint_name\r\n 157 kwargs[fingerprint_name] = update_fingerprint(\r\n--> 158 self._fingerprint, transform, kwargs_for_fingerprint\r\n 159 )\r\n 160 \r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in update_fingerprint(fingerprint, transform, transform_args)\r\n 103 for key in sorted(transform_args):\r\n 104 hasher.update(key)\r\n--> 105 hasher.update(transform_args[key])\r\n 106 return hasher.hexdigest()\r\n 107 \r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in update(self, value)\r\n 55 def update(self, value):\r\n 56 self.m.update(f\"=={type(value)}==\".encode(\"utf8\"))\r\n---> 57 self.m.update(self.hash(value).encode(\"utf-8\"))\r\n 58 \r\n 59 def hexdigest(self):\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in hash(cls, value)\r\n 51 return cls.dispatch[type(value)](cls, value)\r\n 52 else:\r\n---> 53 return cls.hash_default(value)\r\n 54 \r\n 55 def update(self, value):\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/fingerprint.py in hash_default(cls, value)\r\n 44 @classmethod\r\n 45 def hash_default(cls, value):\r\n---> 46 return cls.hash_bytes(dumps(value))\r\n 47 \r\n 48 @classmethod\r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/utils/py_utils.py in dumps(obj)\r\n 365 file = StringIO()\r\n 366 with _no_cache_fields(obj):\r\n--> 367 dump(obj, file)\r\n 368 return file.getvalue()\r\n 369 \r\n\r\n/usr/local/lib/python3.6/site-packages/datasets/utils/py_utils.py in dump(obj, file)\r\n 337 def dump(obj, file):\r\n 338 \"\"\"pickle an object to a file\"\"\"\r\n--> 339 Pickler(file, recurse=True).dump(obj)\r\n 340 return\r\n 341 \r\n\r\n/usr/local/lib/python3.6/site-packages/dill/_dill.py in dump(self, obj)\r\n 444 raise PicklingError(msg)\r\n 445 else:\r\n--> 446 StockPickler.dump(self, obj)\r\n 447 stack.clear() # clear record of 'recursion-sensitive' pickled objects\r\n 448 return\r\n\r\n/usr/local/lib/python3.6/pickle.py in dump(self, obj)\r\n 407 if self.proto >= 4:\r\n 408 self.framer.start_framing()\r\n--> 409 self.save(obj)\r\n 410 self.write(STOP)\r\n 411 self.framer.end_framing()\r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 474 f = self.dispatch.get(t)\r\n 475 if f is not None:\r\n--> 476 f(self, obj) # Call unbound method with explicit self\r\n 477 return\r\n 478 \r\n\r\n/usr/local/lib/python3.6/site-packages/dill/_dill.py in save_function(pickler, obj)\r\n 1436 globs, obj.__name__,\r\n 1437 obj.__defaults__, obj.__closure__,\r\n-> 1438 obj.__dict__, fkwdefaults), obj=obj)\r\n 1439 else:\r\n 1440 _super = ('super' in getattr(obj.func_code,'co_names',())) and (_byref is not None) and getattr(pickler, '_recurse', False)\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)\r\n 608 else:\r\n 609 save(func)\r\n--> 610 save(args)\r\n 611 write(REDUCE)\r\n 612 \r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 474 f = self.dispatch.get(t)\r\n 475 if f is not None:\r\n--> 476 f(self, obj) # Call unbound method with explicit self\r\n 477 return\r\n 478 \r\n\r\n/usr/local/lib/python3.6/pickle.py in save_tuple(self, obj)\r\n 749 write(MARK)\r\n 750 for element in obj:\r\n--> 751 save(element)\r\n 752 \r\n 753 if id(obj) in memo:\r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 474 f = self.dispatch.get(t)\r\n 475 if f is not None:\r\n--> 476 f(self, obj) # Call unbound method with explicit self\r\n 477 return\r\n 478 \r\n\r\n/usr/local/lib/python3.6/site-packages/dill/_dill.py in save_module_dict(pickler, obj)\r\n 931 # we only care about session the first pass thru\r\n 932 pickler._session = False\r\n--> 933 StockPickler.save_dict(pickler, obj)\r\n 934 log.info(\"# D2\")\r\n 935 return\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_dict(self, obj)\r\n 819 \r\n 820 self.memoize(obj)\r\n--> 821 self._batch_setitems(obj.items())\r\n 822 \r\n 823 dispatch[dict] = save_dict\r\n\r\n/usr/local/lib/python3.6/pickle.py in _batch_setitems(self, items)\r\n 850 k, v = tmp[0]\r\n 851 save(k)\r\n--> 852 save(v)\r\n 853 write(SETITEM)\r\n 854 # else tmp is empty, and we're done\r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 519 \r\n 520 # Save the reduce() output and finally memoize the object\r\n--> 521 self.save_reduce(obj=obj, *rv)\r\n 522 \r\n 523 def persistent_id(self, obj):\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)\r\n 632 \r\n 633 if state is not None:\r\n--> 634 save(state)\r\n 635 write(BUILD)\r\n 636 \r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 474 f = self.dispatch.get(t)\r\n 475 if f is not None:\r\n--> 476 f(self, obj) # Call unbound method with explicit self\r\n 477 return\r\n 478 \r\n\r\n/usr/local/lib/python3.6/site-packages/dill/_dill.py in save_module_dict(pickler, obj)\r\n 931 # we only care about session the first pass thru\r\n 932 pickler._session = False\r\n--> 933 StockPickler.save_dict(pickler, obj)\r\n 934 log.info(\"# D2\")\r\n 935 return\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_dict(self, obj)\r\n 819 \r\n 820 self.memoize(obj)\r\n--> 821 self._batch_setitems(obj.items())\r\n 822 \r\n 823 dispatch[dict] = save_dict\r\n\r\n/usr/local/lib/python3.6/pickle.py in _batch_setitems(self, items)\r\n 845 for k, v in tmp:\r\n 846 save(k)\r\n--> 847 save(v)\r\n 848 write(SETITEMS)\r\n 849 elif n:\r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 519 \r\n 520 # Save the reduce() output and finally memoize the object\r\n--> 521 self.save_reduce(obj=obj, *rv)\r\n 522 \r\n 523 def persistent_id(self, obj):\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)\r\n 632 \r\n 633 if state is not None:\r\n--> 634 save(state)\r\n 635 write(BUILD)\r\n 636 \r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 474 f = self.dispatch.get(t)\r\n 475 if f is not None:\r\n--> 476 f(self, obj) # Call unbound method with explicit self\r\n 477 return\r\n 478 \r\n\r\n/usr/local/lib/python3.6/site-packages/dill/_dill.py in save_module_dict(pickler, obj)\r\n 931 # we only care about session the first pass thru\r\n 932 pickler._session = False\r\n--> 933 StockPickler.save_dict(pickler, obj)\r\n 934 log.info(\"# D2\")\r\n 935 return\r\n\r\n/usr/local/lib/python3.6/pickle.py in save_dict(self, obj)\r\n 819 \r\n 820 self.memoize(obj)\r\n--> 821 self._batch_setitems(obj.items())\r\n 822 \r\n 823 dispatch[dict] = save_dict\r\n\r\n/usr/local/lib/python3.6/pickle.py in _batch_setitems(self, items)\r\n 845 for k, v in tmp:\r\n 846 save(k)\r\n--> 847 save(v)\r\n 848 write(SETITEMS)\r\n 849 elif n:\r\n\r\n/usr/local/lib/python3.6/pickle.py in save(self, obj, save_persistent_id)\r\n 494 reduce = getattr(obj, \"__reduce_ex__\", None)\r\n 495 if reduce is not None:\r\n--> 496 rv = reduce(self.proto)\r\n 497 else:\r\n 498 reduce = getattr(obj, \"__reduce__\", None)\r\n\r\nTypeError: can't pickle Tagger objects\r\n```\r\n\r\n</details>\r\n\r\ntrainsformers: 2.10.0\r\ndatasets: 1.0.2\r\ndill: 0.3.2\r\npython: 3.6.8\r\n\r\nOS: ubuntu 16.04 (Docker Image) on [Deep Learning VM](https://console.cloud.google.com/marketplace/details/click-to-deploy-images/deeplearning) (GCP)\r\nGPU: Tesla P100 (CUDA 10)\r\n",
"> I have the same issue with `transformers/BertJapaneseTokenizer`.\r\n\r\nIt looks like it this tokenizer is not supported unfortunately.\r\nThis is because `t.word_tokenizer.mecab` is a `fugashi.fugashi.GenericTagger` which is not compatible with pickle nor dill.\r\n\r\nWe need objects passes to `map` to be picklable for our caching system to work properly.\r\nHere it crashes because the caching system is not able to pickle the GenericTagger.\r\n\r\n\\> Maybe you can create an issue on [fugashi](https://github.com/polm/fugashi/issues) 's repo and ask to make `fugashi.fugashi.GenericTagger` compatible with pickle ?\r\n\r\nWhat you can do in the meantime is use a picklable wrapper of the tokenizer:\r\n\r\n\r\n```python\r\nfrom transformers import BertJapaneseTokenizer, MecabTokenizer\r\n\r\nclass PicklableTokenizer(BertJapaneseTokenizer):\r\n\r\n def __getstate__(self):\r\n state = dict(self.__dict__)\r\n state[\"do_lower_case\"] = self.word_tokenizer.do_lower_case\r\n state[\"never_split\"] = self.word_tokenizer.never_split \r\n del state[\"word_tokenizer\"]\r\n return state\r\n\r\n def __setstate__(self, state):\r\n do_lower_case = state.pop(\"do_lower_case\")\r\n never_split = state.pop(\"never_split\")\r\n self.__dict__ = state\r\n self.word_tokenizer = MecabTokenizer(\r\n do_lower_case=do_lower_case, never_split=never_split)\r\n )\r\n\r\nt = PicklableTokenizer.from_pretrained(\"cl-tohoku/bert-base-japanese-whole-word-masking\")\r\nencoded = train_ds.map(lambda examples: {'tokens': t.encode(examples['title'])}, batched=True) # it works\r\n```",
"We can also update the `BertJapaneseTokenizer` in `transformers` as you just shown @lhoestq to make it compatible with pickle. It will be faster than asking on fugashi 's repo and good for the other users of `transformers` as well.\r\n\r\nI'm currently working on `transformers` I'll include it in the https://github.com/huggingface/transformers/pull/7141 PR and the next release of `transformers`.",
"Thank you for the rapid and polite response!\r\n\r\n@lhoestq Thanks for the suggestion! I've passed the pickle phase, but another `ArrowInvalid` problem occored. I created another issue #687 .\r\n\r\n@thomwolf Wow, really fast work. I'm looking forward to the next release π€"
] | I load squad dataset. Then want to process data use following function with `Huggingface Transformers LongformerTokenizer`.
```
def convert_to_features(example):
# Tokenize contexts and questions (as pairs of inputs)
input_pairs = [example['question'], example['context']]
encodings = tokenizer.encode_plus(input_pairs, pad_to_max_length=True, max_length=512)
context_encodings = tokenizer.encode_plus(example['context'])
# Compute start and end tokens for labels using Transformers's fast tokenizers alignement methodes.
# this will give us the position of answer span in the context text
start_idx, end_idx = get_correct_alignement(example['context'], example['answers'])
start_positions_context = context_encodings.char_to_token(start_idx)
end_positions_context = context_encodings.char_to_token(end_idx-1)
# here we will compute the start and end position of the answer in the whole example
# as the example is encoded like this <s> question</s></s> context</s>
# and we know the postion of the answer in the context
# we can just find out the index of the sep token and then add that to position + 1 (+1 because there are two sep tokens)
# this will give us the position of the answer span in whole example
sep_idx = encodings['input_ids'].index(tokenizer.sep_token_id)
start_positions = start_positions_context + sep_idx + 1
end_positions = end_positions_context + sep_idx + 1
if end_positions > 512:
start_positions, end_positions = 0, 0
encodings.update({'start_positions': start_positions,
'end_positions': end_positions,
'attention_mask': encodings['attention_mask']})
return encodings
```
Then I run `dataset.map(convert_to_features)`, it raise
```
In [59]: a.map(convert_to_features)
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-59-c453b508761d> in <module>
----> 1 a.map(convert_to_features)
/opt/conda/lib/python3.7/site-packages/datasets/arrow_dataset.py in map(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint)
1242 fn_kwargs=fn_kwargs,
1243 new_fingerprint=new_fingerprint,
-> 1244 update_data=update_data,
1245 )
1246 else:
/opt/conda/lib/python3.7/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
151 "output_all_columns": self._output_all_columns,
152 }
--> 153 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
154 if new_format["columns"] is not None:
155 new_format["columns"] = list(set(new_format["columns"]) & set(out.column_names))
/opt/conda/lib/python3.7/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
156 kwargs_for_fingerprint["fingerprint_name"] = fingerprint_name
157 kwargs[fingerprint_name] = update_fingerprint(
--> 158 self._fingerprint, transform, kwargs_for_fingerprint
159 )
160
/opt/conda/lib/python3.7/site-packages/datasets/fingerprint.py in update_fingerprint(fingerprint, transform, transform_args)
103 for key in sorted(transform_args):
104 hasher.update(key)
--> 105 hasher.update(transform_args[key])
106 return hasher.hexdigest()
107
/opt/conda/lib/python3.7/site-packages/datasets/fingerprint.py in update(self, value)
55 def update(self, value):
56 self.m.update(f"=={type(value)}==".encode("utf8"))
---> 57 self.m.update(self.hash(value).encode("utf-8"))
58
59 def hexdigest(self):
/opt/conda/lib/python3.7/site-packages/datasets/fingerprint.py in hash(cls, value)
51 return cls.dispatch[type(value)](cls, value)
52 else:
---> 53 return cls.hash_default(value)
54
55 def update(self, value):
/opt/conda/lib/python3.7/site-packages/datasets/fingerprint.py in hash_default(cls, value)
44 @classmethod
45 def hash_default(cls, value):
---> 46 return cls.hash_bytes(dumps(value))
47
48 @classmethod
/opt/conda/lib/python3.7/site-packages/datasets/utils/py_utils.py in dumps(obj)
365 file = StringIO()
366 with _no_cache_fields(obj):
--> 367 dump(obj, file)
368 return file.getvalue()
369
/opt/conda/lib/python3.7/site-packages/datasets/utils/py_utils.py in dump(obj, file)
337 def dump(obj, file):
338 """pickle an object to a file"""
--> 339 Pickler(file, recurse=True).dump(obj)
340 return
341
/opt/conda/lib/python3.7/site-packages/dill/_dill.py in dump(self, obj)
444 raise PicklingError(msg)
445 else:
--> 446 StockPickler.dump(self, obj)
447 stack.clear() # clear record of 'recursion-sensitive' pickled objects
448 return
/opt/conda/lib/python3.7/pickle.py in dump(self, obj)
435 if self.proto >= 4:
436 self.framer.start_framing()
--> 437 self.save(obj)
438 self.write(STOP)
439 self.framer.end_framing()
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
502 f = self.dispatch.get(t)
503 if f is not None:
--> 504 f(self, obj) # Call unbound method with explicit self
505 return
506
/opt/conda/lib/python3.7/site-packages/dill/_dill.py in save_function(pickler, obj)
1436 globs, obj.__name__,
1437 obj.__defaults__, obj.__closure__,
-> 1438 obj.__dict__, fkwdefaults), obj=obj)
1439 else:
1440 _super = ('super' in getattr(obj.func_code,'co_names',())) and (_byref is not None) and getattr(pickler, '_recurse', False)
/opt/conda/lib/python3.7/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)
636 else:
637 save(func)
--> 638 save(args)
639 write(REDUCE)
640
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
502 f = self.dispatch.get(t)
503 if f is not None:
--> 504 f(self, obj) # Call unbound method with explicit self
505 return
506
/opt/conda/lib/python3.7/pickle.py in save_tuple(self, obj)
787 write(MARK)
788 for element in obj:
--> 789 save(element)
790
791 if id(obj) in memo:
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
502 f = self.dispatch.get(t)
503 if f is not None:
--> 504 f(self, obj) # Call unbound method with explicit self
505 return
506
/opt/conda/lib/python3.7/site-packages/dill/_dill.py in save_module_dict(pickler, obj)
931 # we only care about session the first pass thru
932 pickler._session = False
--> 933 StockPickler.save_dict(pickler, obj)
934 log.info("# D2")
935 return
/opt/conda/lib/python3.7/pickle.py in save_dict(self, obj)
857
858 self.memoize(obj)
--> 859 self._batch_setitems(obj.items())
860
861 dispatch[dict] = save_dict
/opt/conda/lib/python3.7/pickle.py in _batch_setitems(self, items)
883 for k, v in tmp:
884 save(k)
--> 885 save(v)
886 write(SETITEMS)
887 elif n:
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
547
548 # Save the reduce() output and finally memoize the object
--> 549 self.save_reduce(obj=obj, *rv)
550
551 def persistent_id(self, obj):
/opt/conda/lib/python3.7/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)
660
661 if state is not None:
--> 662 save(state)
663 write(BUILD)
664
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
502 f = self.dispatch.get(t)
503 if f is not None:
--> 504 f(self, obj) # Call unbound method with explicit self
505 return
506
/opt/conda/lib/python3.7/site-packages/dill/_dill.py in save_module_dict(pickler, obj)
931 # we only care about session the first pass thru
932 pickler._session = False
--> 933 StockPickler.save_dict(pickler, obj)
934 log.info("# D2")
935 return
/opt/conda/lib/python3.7/pickle.py in save_dict(self, obj)
857
858 self.memoize(obj)
--> 859 self._batch_setitems(obj.items())
860
861 dispatch[dict] = save_dict
/opt/conda/lib/python3.7/pickle.py in _batch_setitems(self, items)
883 for k, v in tmp:
884 save(k)
--> 885 save(v)
886 write(SETITEMS)
887 elif n:
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
547
548 # Save the reduce() output and finally memoize the object
--> 549 self.save_reduce(obj=obj, *rv)
550
551 def persistent_id(self, obj):
/opt/conda/lib/python3.7/pickle.py in save_reduce(self, func, args, state, listitems, dictitems, obj)
660
661 if state is not None:
--> 662 save(state)
663 write(BUILD)
664
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
502 f = self.dispatch.get(t)
503 if f is not None:
--> 504 f(self, obj) # Call unbound method with explicit self
505 return
506
/opt/conda/lib/python3.7/site-packages/dill/_dill.py in save_module_dict(pickler, obj)
931 # we only care about session the first pass thru
932 pickler._session = False
--> 933 StockPickler.save_dict(pickler, obj)
934 log.info("# D2")
935 return
/opt/conda/lib/python3.7/pickle.py in save_dict(self, obj)
857
858 self.memoize(obj)
--> 859 self._batch_setitems(obj.items())
860
861 dispatch[dict] = save_dict
/opt/conda/lib/python3.7/pickle.py in _batch_setitems(self, items)
883 for k, v in tmp:
884 save(k)
--> 885 save(v)
886 write(SETITEMS)
887 elif n:
/opt/conda/lib/python3.7/pickle.py in save(self, obj, save_persistent_id)
522 reduce = getattr(obj, "__reduce_ex__", None)
523 if reduce is not None:
--> 524 rv = reduce(self.proto)
525 else:
526 reduce = getattr(obj, "__reduce__", None)
TypeError: can't pickle Tokenizer objects
```
| 665 |
https://github.com/huggingface/datasets/issues/664 | load_dataset from local squad.py, raise error: TypeError: 'NoneType' object is not callable | [
"Hi !\r\nThanks for reporting.\r\nIt looks like no object inherits from `datasets.GeneratorBasedBuilder` (or more generally from `datasets.DatasetBuilder`) in your script.\r\n\r\nCould you check that there exist at least one dataset builder class ?",
"Hi @xixiaoyao did you manage to fix your issue ?",
"No activity, closing",
"It happened when try to change the old project which use 'nlp' to new project which use 'datasets'. You should check you old 'my_squad.py' file, change the inherit class from `nlp.xxx` to `datasets.xxx`. Otherwise datasets - load.py - import_main_class() `if inspect.isclass(obj) and issubclass(obj, main_cls_type):` can not find the main_cls."
] |
version: 1.0.2
```
train_dataset = datasets.load_dataset('squad')
```
The above code can works. However, when I download the squad.py from your server, and saved as `my_squad.py` to local. I run followings raise errors.
```
train_dataset = datasets.load_dataset('./my_squad.py')
```
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-28-25a84b4d1581> in <module>
----> 1 train_dataset = nlp.load_dataset('./my_squad.py')
/opt/conda/lib/python3.7/site-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)
602 hash=hash,
603 features=features,
--> 604 **config_kwargs,
605 )
606
TypeError: 'NoneType' object is not callable
| 664 |
https://github.com/huggingface/datasets/issues/657 | Squad Metric Description & Feature Mismatch | [
"Thanks for reporting !\r\nThere indeed a mismatch between the features and the kwargs description\r\n\r\nI believe `answer_start` was added to match the squad dataset format for consistency, even though it is not used in the metric computation. I think I'd rather keep it this way, so that you can just give `references=squad[\"answers\"]` to `.compute()`.\r\nMaybe we can just fix the description then.",
"But then providing the `answer_start` becomes mandatory since the format of the features is checked against the one provided in the squad [file](https://github.com/huggingface/datasets/pull/658/files)."
] | The [description](https://github.com/huggingface/datasets/blob/master/metrics/squad/squad.py#L39) doesn't mention `answer_start` in squad. However the `datasets.features` require [it](https://github.com/huggingface/datasets/blob/master/metrics/squad/squad.py#L68). It's also not used in the evaluation. | 657 |
https://github.com/huggingface/datasets/issues/651 | Problem with JSON dataset format | [
"Currently the `json` dataset doesn't support this format unfortunately.\r\nHowever you could load it with\r\n```python\r\nfrom datasets import Dataset\r\nimport pandas as pd\r\n\r\ndf = pd.read_json(\"path_to_local.json\", orient=\"index\")\r\ndataset = Dataset.from_pandas(df)\r\n```",
"or you can make a custom dataset script as explained in doc here: https://huggingface.co/docs/datasets/add_dataset.html"
] | I have a local json dataset with the following form.
{
'id01234': {'key1': value1, 'key2': value2, 'key3': value3},
'id01235': {'key1': value1, 'key2': value2, 'key3': value3},
.
.
.
'id09999': {'key1': value1, 'key2': value2, 'key3': value3}
}
Note that instead of a list of records it's basically a dictionary of key value pairs with the keys being the record_ids and the values being the corresponding record.
Reading this with json:
```
data = datasets.load('json', data_files='path_to_local.json')
```
Throws an error and asks me to chose a field. What's the right way to handle this? | 651 |
https://github.com/huggingface/datasets/issues/650 | dummy data testing can't test datasets using `dl_manager.extract` in `_split_generators` | [
"Hi :) \r\nIn your dummy data zip file you can just have `subset000.xz` as directories instead of compressed files.\r\nLet me know if it helps",
"Thanks for your comment @lhoestq ,\r\nJust for confirmation, changing dummy data like this won't make dummy test test the functionality to extract `subsetxxx.xz` but actually kind of circumvent it. But since we will test the real data so it is ok ?",
"Yes it's fine for now. We plan to add a job for slow tests.\r\nAnd at one point we'll also do another pass on the dummy data handling and consider extracting files.",
"Thanks for the confirmation.\r\nAlso the suggestion works. Thank you."
] | Hi, I recently want to add a dataset whose source data is like this
```
openwebtext.tar.xz
|__ openwebtext
|__subset000.xz
| |__ ....txt
| |__ ....txt
| ...
|__ subset001.xz
|
....
```
So I wrote `openwebtext.py` like this
```
def _split_generators(self, dl_manager):
dl_dir = dl_manager.download_and_extract(_URL)
owt_dir = os.path.join(dl_dir, 'openwebtext')
subset_xzs = [
os.path.join(owt_dir, file_name) for file_name in os.listdir(owt_dir) if file_name.endswith('xz') # filter out ...xz.lock
]
ex_dirs = dl_manager.extract(subset_xzs, num_proc=round(os.cpu_count()*0.75))
nested_txt_files = [
[
os.path.join(ex_dir,txt_file_name) for txt_file_name in os.listdir(ex_dir) if txt_file_name.endswith('txt')
] for ex_dir in ex_dirs
]
txt_files = chain(*nested_txt_files)
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN, gen_kwargs={"txt_files": txt_files}
),
]
```
All went good, I can load and use real openwebtext, except when I try to test with dummy data. The problem is `MockDownloadManager.extract` do nothing, so `ex_dirs = dl_manager.extract(subset_xzs)` won't decompress `subset_xxx.xz`s for me.
How should I do ? Or you can modify `MockDownloadManager` to make it like a real `DownloadManager` ? | 650 |
https://github.com/huggingface/datasets/issues/649 | Inconsistent behavior in map | [
"Thanks for reporting !\r\n\r\nThis issue must have appeared when we refactored type inference in `nlp`\r\nBy default the library tries to keep the same feature types when applying `map` but apparently it has troubles with nested structures. I'll try to fix that next week"
] | I'm observing inconsistent behavior when applying .map(). This happens specifically when I'm incrementally adding onto a feature that is a nested dictionary. Here's a simple example that reproduces the problem.
```python
import datasets
# Dataset with a single feature called 'field' consisting of two examples
dataset = datasets.Dataset.from_dict({'field': ['a', 'b']})
print(dataset[0])
# outputs
{'field': 'a'}
# Map this dataset to create another feature called 'otherfield', which is a dictionary containing a key called 'capital'
dataset = dataset.map(lambda example: {'otherfield': {'capital': example['field'].capitalize()}})
print(dataset[0])
# output is okay
{'field': 'a', 'otherfield': {'capital': 'A'}}
# Now I want to map again to modify 'otherfield', by adding another key called 'append_x' to the dictionary under 'otherfield'
print(dataset.map(lambda example: {'otherfield': {'append_x': example['field'] + 'x'}})[0])
# printing out the first example after applying the map shows that the new key 'append_x' doesn't get added
# it also messes up the value stored at 'capital'
{'field': 'a', 'otherfield': {'capital': None}}
# Instead, I try to do the same thing by using a different mapped fn
print(dataset.map(lambda example: {'otherfield': {'append_x': example['field'] + 'x', 'capital': example['otherfield']['capital']}})[0])
# this preserves the value under capital, but still no 'append_x'
{'field': 'a', 'otherfield': {'capital': 'A'}}
# Instead, I try to pass 'otherfield' to remove_columns
print(dataset.map(lambda example: {'otherfield': {'append_x': example['field'] + 'x', 'capital': example['otherfield']['capital']}}, remove_columns=['otherfield'])[0])
# this still doesn't fix the problem
{'field': 'a', 'otherfield': {'capital': 'A'}}
# Alternately, here's what happens if I just directly map both 'capital' and 'append_x' on a fresh dataset.
# Recreate the dataset
dataset = datasets.Dataset.from_dict({'field': ['a', 'b']})
# Now map the entire 'otherfield' dict directly, instead of incrementally as before
print(dataset.map(lambda example: {'otherfield': {'append_x': example['field'] + 'x', 'capital': example['field'].capitalize()}})[0])
# This looks good!
{'field': 'a', 'otherfield': {'append_x': 'ax', 'capital': 'A'}}
```
This might be a new issue, because I didn't see this behavior in the `nlp` library.
Any help is appreciated! | 649 |
https://github.com/huggingface/datasets/issues/648 | offset overflow when multiprocessing batched map on large datasets. | [
"This should be fixed with #645 ",
"Feel free to re-open if it still occurs"
] | It only happened when "multiprocessing" + "batched" + "large dataset" at the same time.
```
def bprocess(examples):
examples['len'] = []
for text in examples['text']:
examples['len'].append(len(text))
return examples
wiki.map(brpocess, batched=True, num_proc=8)
```
```
---------------------------------------------------------------------------
RemoteTraceback Traceback (most recent call last)
RemoteTraceback:
"""
Traceback (most recent call last):
File "/home/yisiang/miniconda3/envs/ml/lib/python3.7/multiprocessing/pool.py", line 121, in worker
result = (True, func(*args, **kwds))
File "/home/yisiang/datasets/src/datasets/arrow_dataset.py", line 153, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/home/yisiang/datasets/src/datasets/fingerprint.py", line 163, in wrapper
out = func(self, *args, **kwargs)
File "/home/yisiang/datasets/src/datasets/arrow_dataset.py", line 1486, in _map_single
batch = self[i : i + batch_size]
File "/home/yisiang/datasets/src/datasets/arrow_dataset.py", line 1071, in __getitem__
format_kwargs=self._format_kwargs,
File "/home/yisiang/datasets/src/datasets/arrow_dataset.py", line 972, in _getitem
data_subset = self._data.take(indices_array)
File "pyarrow/table.pxi", line 1145, in pyarrow.lib.Table.take
File "/home/yisiang/miniconda3/envs/ml/lib/python3.7/site-packages/pyarrow/compute.py", line 268, in take
return call_function('take', [data, indices], options)
File "pyarrow/_compute.pyx", line 298, in pyarrow._compute.call_function
File "pyarrow/_compute.pyx", line 192, in pyarrow._compute.Function.call
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: offset overflow while concatenating arrays
"""
The above exception was the direct cause of the following exception:
ArrowInvalid Traceback (most recent call last)
in
30 owt = datasets.load_dataset('/home/yisiang/datasets/datasets/openwebtext/openwebtext.py', cache_dir='./datasets')['train']
31 print('load/create data from OpenWebText Corpus for ELECTRA')
---> 32 e_owt = ELECTRAProcessor(owt, apply_cleaning=False).map(cache_file_name=f"electra_owt_{c.max_length}.arrow")
33 dsets.append(e_owt)
34
~/Reexamine_Attention/electra_pytorch/_utils/utils.py in map(self, **kwargs)
126 writer_batch_size=10**4,
127 num_proc=num_proc,
--> 128 **kwargs
129 )
130
~/hugdatafast/hugdatafast/transform.py in my_map(self, *args, **kwargs)
21 if not cache_file_name.endswith('.arrow'): cache_file_name += '.arrow'
22 if '/' not in cache_file_name: cache_file_name = os.path.join(self.cache_directory(), cache_file_name)
---> 23 return self.map(*args, cache_file_name=cache_file_name, **kwargs)
24
25 @patch
~/datasets/src/datasets/arrow_dataset.py in map(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint)
1285 logger.info("Spawning {} processes".format(num_proc))
1286 results = [pool.apply_async(self.__class__._map_single, kwds=kwds) for kwds in kwds_per_shard]
-> 1287 transformed_shards = [r.get() for r in results]
1288 logger.info("Concatenating {} shards from multiprocessing".format(num_proc))
1289 result = concatenate_datasets(transformed_shards)
~/datasets/src/datasets/arrow_dataset.py in (.0)
1285 logger.info("Spawning {} processes".format(num_proc))
1286 results = [pool.apply_async(self.__class__._map_single, kwds=kwds) for kwds in kwds_per_shard]
-> 1287 transformed_shards = [r.get() for r in results]
1288 logger.info("Concatenating {} shards from multiprocessing".format(num_proc))
1289 result = concatenate_datasets(transformed_shards)
~/miniconda3/envs/ml/lib/python3.7/multiprocessing/pool.py in get(self, timeout)
655 return self._value
656 else:
--> 657 raise self._value
658
659 def _set(self, i, obj):
ArrowInvalid: offset overflow while concatenating arrays
``` | 648 |
https://github.com/huggingface/datasets/issues/647 | Cannot download dataset_info.json | [
"Thanks for reporting !\r\nWe should add support for servers without internet connection indeed\r\nI'll do that early next week",
"Thanks, @lhoestq !\r\nPlease let me know when it is available. ",
"Right now the recommended way is to create the dataset on a server with internet connection and then to save it and copy the serialized dataset to the server without internet connection.",
"#652 should allow you to load text/json/csv/pandas datasets without an internet connection **IF** you've the dataset script locally.\r\n\r\nExample: \r\nIf you have `datasets/text/text.py` locally, then you can do `load_dataset(\"./datasets/text\", data_files=...)`"
] | I am running my job on a cloud server where does not provide for connections from the standard compute nodes to outside resources. Hence, when I use `dataset.load_dataset()` to load data, I got an error like this:
```
ConnectionError: Couldn't reach https://storage.googleapis.com/huggingface-nlp/cache/datasets/text/default-53ee3045f07ba8ca/0.0.0/dataset_info.json
```
I tried to open this link manually, but I cannot access this file. How can I download this file and pass it through `dataset.load_dataset()` manually?
Versions:
Python version 3.7.3
PyTorch version 1.6.0
TensorFlow version 2.3.0
datasets version: 1.0.1
| 647 |
https://github.com/huggingface/datasets/issues/643 | Caching processed dataset at wrong folder | [
"Thanks for reporting !\r\nIt uses a temporary file to write the data.\r\nHowever it looks like the temporary file is not placed in the right directory during the processing",
"Well actually I just tested and the temporary file is placed in the same directory, so it should work as expected.\r\nWhich version of `datasets` are you using ?",
"`datasets-1.0.1`\r\nHere you can reproduce it here:\r\nhttps://colab.research.google.com/drive/1O0KcepTFsmpkBbrbLLMq42iwTKmQh8d5?usp=sharing\r\n",
"It looks like a pyarrow issue with google colab.\r\nFor some reason this code increases the disk usage of google colab while it actually writes into google drive:\r\n\r\n```python\r\nimport pyarrow as pa\r\n\r\nstream = pa.OSFile(\"/content/drive/My Drive/path/to/file.arrow\", \"wb\")\r\nwriter = pa.RecordBatchStreamWriter(stream, schema=pa.schema({\"text\": pa.string()}))\r\nwriter.write_table(pa.Table.from_pydict({\"text\": [\"a\"*511 + \"\\n\"] * ((1 << 30) // 512)})) # 1GiB\r\nwriter.close()\r\nstream.close()\r\n```\r\n\r\nMoreover if I `rm` the file on google drive, it frees disk space on google colab.",
"It looks like replacing `pa.OSFile` by `open` fixes it, I'm going to open a PR",
"Ok. Thank you so much!",
"Actually I did more tests it doesn't >.<\r\nI'll let you know if I find a way to fix that",
"Actually I also have the issue when writing a regular text file\r\n\r\n```python\r\nf = open(\"/content/drive/My Drive/path/to/file\", \"w\")\r\nf.write((\"a\"*511 + \"\\n\") * ((1 << 30) // 512)) # 1GiB\r\nf.close()\r\n```\r\n\r\nIs that supposed to happen ?",
"The code you wrote should write a 1GB file in the Google Drive folder. Doesn't it? ",
"Yes it does, but the disk usage of google colab also increases by 1GB",
"I could check it and as you say as I write to te Drive disk the colab disk also increases...",
"To reproduce it: \r\n```bash\r\n!df -h | grep sda1\r\n```\r\n```python\r\nf = open(\"/content/drive/My Drive/test_to_remove.txt\", \"w\")\r\nf.write((\"a\"*511 + \"\\n\") * ((1 << 30) // 512)) # 1GiB\r\nf.write((\"a\"*511 + \"\\n\") * ((1 << 30) // 512)) # 1GiB\r\nf.close()\r\n```\r\n```bash\r\n!ls -lh /content/drive/My\\ Drive/test_to_remove.txt\r\n\r\n!df -h | grep sda1\r\n\r\n!rm -rf /content/drive/My\\ Drive/test_to_remove.txt\r\n\r\n```\r\n[Colab](https://colab.research.google.com/drive/1D0UiweCYQwwWZ65EEhuqqbaDDbhJYXfm?usp=sharing)\r\n\r\n\r\n",
"Apparently, Colab uses a local cache of the data files read/written from Google Drive. See:\r\n- https://github.com/googlecolab/colabtools/issues/2087#issuecomment-860818457\r\n- https://github.com/googlecolab/colabtools/issues/1915#issuecomment-804234540\r\n- https://github.com/googlecolab/colabtools/issues/2147#issuecomment-885052636"
] | Hi guys, I run this on my Colab (PRO):
```python
from datasets import load_dataset
dataset = load_dataset('text', data_files='/content/corpus.txt', cache_dir='/content/drive/My Drive', split='train')
def encode(examples):
return tokenizer(examples['text'], truncation=True, padding='max_length')
dataset = dataset.map(encode, batched=True)
```
The file is about 4 GB, so I cannot process it on the Colab HD because there is no enough space. So I decided to mount my Google Drive fs and do it on it.
The dataset is cached in the right place but by processing it (applying `encode` function) seems to use a different folder because Colab HD starts to grow and it crashes when it should be done in the Drive fs.
What gets me crazy, it prints it is processing/encoding the dataset in the right folder:
```
Testing the mapped function outputs
Testing finished, running the mapping function on the dataset
Caching processed dataset at /content/drive/My Drive/text/default-ad3e69d6242ee916/0.0.0/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7/cache-b16341780a59747d.arrow
``` | 643 |
https://github.com/huggingface/datasets/issues/638 | GLUE/QQP dataset: NonMatchingChecksumError | [
"Hi ! Sure I'll take a look"
] | Hi @lhoestq , I know you are busy and there are also other important issues. But if this is easy to be fixed, I am shamelessly wondering if you can give me some help , so I can evaluate my models and restart with my developing cycle asap. π
datasets version: editable install of master at 9/17
`datasets.load_dataset('glue','qqp', cache_dir='./datasets')`
```
Downloading and preparing dataset glue/qqp (download: 57.73 MiB, generated: 107.02 MiB, post-processed: Unknown size, total: 164.75 MiB) to ./datasets/glue/qqp/1.0.0/7c99657241149a24692c402a5c3f34d4c9f1df5ac2e4c3759fadea38f6cb29c4...
---------------------------------------------------------------------------
NonMatchingChecksumError Traceback (most recent call last)
in
----> 1 datasets.load_dataset('glue','qqp', cache_dir='./datasets')
~/datasets/src/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)
609 download_config=download_config,
610 download_mode=download_mode,
--> 611 ignore_verifications=ignore_verifications,
612 )
613
~/datasets/src/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
467 if not downloaded_from_gcs:
468 self._download_and_prepare(
--> 469 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
470 )
471 # Sync info
~/datasets/src/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
527 if verify_infos:
528 verify_checksums(
--> 529 self.info.download_checksums, dl_manager.get_recorded_sizes_checksums(), "dataset source files"
530 )
531
~/datasets/src/datasets/utils/info_utils.py in verify_checksums(expected_checksums, recorded_checksums, verification_name)
37 if len(bad_urls) > 0:
38 error_msg = "Checksums didn't match" + for_verification_name + ":\n"
---> 39 raise NonMatchingChecksumError(error_msg + str(bad_urls))
40 logger.info("All the checksums matched successfully" + for_verification_name)
41
NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://dl.fbaipublicfiles.com/glue/data/QQP-clean.zip']
``` | 638 |
https://github.com/huggingface/datasets/issues/633 | Load large text file for LM pre-training resulting in OOM | [
"Not sure what could cause that on the `datasets` side. Could this be a `Trainer` issue ? cc @julien-c @sgugger ?",
"There was a memory leak issue fixed recently in master. You should install from source and see if it fixes your problem.",
"@lhoestq @sgugger Thanks for your comments. I have install from source code as you told, but the problem is still there.\r\nTo reproduce the issue, just replace [these lines](https://github.com/huggingface/transformers/blob/master/examples/language-modeling/run_language_modeling.py#L241-L258) with: \r\n(load_dataset and DataCollatorForDatasetsLanguageModeling as [above mentioned](https://github.com/huggingface/datasets/issues/633#issue-702440484))\r\n```python\r\n dataset = load_dataset(\"bookcorpus\")\r\n dataset = dataset.train_test_split(test_size=0.1)\r\n train_dataset = dataset['train']\r\n eval_dataset = dataset['test'] if training_args.do_eval else None\r\n\r\n data_collator = DataCollatorForDatasetsLanguageModeling(\r\n tokenizer=tokenizer,\r\n mlm=data_args.mlm,\r\n mlm_probability=data_args.mlm_probability,\r\n block_size=data_args.block_size\r\n )\r\n```\r\nand run by:\r\n```bash\r\npython run_language_modeling.py\r\n--output_dir=output \\\r\n--model_type=bert \\\r\n--model_name_or_path=bert-base-uncased \\\r\n--do_train \\\r\n--do_eval \\\r\n--mlm \r\n```",
"Same here. Pre-training on wikitext-103 to do some test. At the end of the training it takes 32GB of RAM + ~30GB of SWAP. I installed dataset==1.1.0, not built from source. I will try uninstalling and building from source when it finish.",
"This seems to be on the `transformers` library side.\r\n\r\nIf you have more informations (pip env) or even better, a colab reproducing the error we can investigate.",
"It seems like it's solved with freshed versions of transformers. I have tried to replicate the error doing a fresh pip install transformers & datasets on colab and the error doesn't continue. On colab it keeps stable on 5GB! (Y)\r\n\r\nEdit: **Thanks for your great work**. Have a good day.",
"@gaceladri witch version transformers and datasets are you using now? I want to try again. Thanks.",
"transformers==3.3.1\r\ndatasets==1.1.0\r\ntokenizers==0.8.1rc2\r\n",
"doing some modifications to mobilebert\r\nhttps://colab.research.google.com/drive/1ba09ZOpyHGAOQLcsxiQAHRXl10qnMU5o?usp=sharing ",
"It does not happen to me anymore. Can we close? @leethu2012 ",
"It's happening to me again. After 4 hours of pre-training, my ram memory gets full and the kernel dies. I am using the last transformers version as today. 4.4.0 and the last version of datasets 1.2.1, both installed from master. The memory consumption keeps increasing.",
"It looks like it is something from pytorch/python itself :face_with_head_bandage: https://github.com/pytorch/pytorch/issues/13246 ",
"Thanks for the investigation @gaceladri \r\n\r\nApparently this happens when `num_workers>0` and has to do with objects being copied-on-write.\r\nDid you try setting num_workers to 0 @gaceladri ?\r\nIf the issue doesn't happen with `num_workers=0` then this would confirm that it's indeed related to this python/pytorch issue.\r\n\r\nSince a `Dataset` object is a wrapper of a pyarrow Table, we should investigate if the data being copied comes from the Table itself or from metadata in the `Dataset` object. If it comes from the metadata in the `Dataset` object, we should be able to implement a workaround. But if it comes from the Table, we'll need to see with the pyarrow team what we can do... ",
"@lhoestq I have tried and it keeps increasing also with `dataloader_num_workers=0`",
"Hmmm so this might come from another issue...\r\nSince it doesn't seem to be related to multiprocessing it should be easier to investigate though.\r\nDo you have some ideas @gaceladri ?",
"@lhoestq I looked quickly to a previously spoted bug in my env wandb /sdk/interface/interface.py, because sometimes when I load the dataset I got a multiprocessing error at line 510 in wandb...interface.py\r\n\r\nThis bug is reported here https://github.com/huggingface/datasets/issues/847\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nAssertionError Traceback (most recent call last)\r\n<timed eval> in <module>\r\n\r\n~/anaconda3/envs/tfm/lib/python3.6/site-packages/transformers/trainer.py in train(self, model_path, trial)\r\n 877 print(len(epoch_iterator))\r\n 878 \r\n--> 879 for step, inputs in enumerate(epoch_iterator):\r\n 880 \r\n 881 start_step = time.time()\r\n\r\n~/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/dataloader.py in __next__(self)\r\n 433 if self._sampler_iter is None:\r\n 434 self._reset()\r\n--> 435 data = self._next_data()\r\n 436 self._num_yielded += 1\r\n 437 if self._dataset_kind == _DatasetKind.Iterable and \\\r\n\r\n~/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/dataloader.py in _next_data(self)\r\n 1083 else:\r\n 1084 del self._task_info[idx]\r\n-> 1085 return self._process_data(data)\r\n 1086 \r\n 1087 def _try_put_index(self):\r\n\r\n~/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/dataloader.py in _process_data(self, data)\r\n 1109 self._try_put_index()\r\n 1110 if isinstance(data, ExceptionWrapper):\r\n-> 1111 data.reraise()\r\n 1112 return data\r\n 1113 \r\n\r\n~/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/_utils.py in reraise(self)\r\n 426 # have message field\r\n 427 raise self.exc_type(message=msg)\r\n--> 428 raise self.exc_type(msg)\r\n 429 \r\n 430 \r\n\r\nAssertionError: Caught AssertionError in DataLoader worker process 0.\r\nOriginal Traceback (most recent call last):\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py\", line 198, in _worker_loop\r\n data = fetcher.fetch(index)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in fetch\r\n data = [self.dataset[idx] for idx in possibly_batched_index]\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/torch/utils/data/_utils/fetch.py\", line 44, in <listcomp>\r\n data = [self.dataset[idx] for idx in possibly_batched_index]\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1083, in __getitem__\r\n format_kwargs=self._format_kwargs,\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1070, in _getitem\r\n format_kwargs=format_kwargs,\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 886, in _convert_outputs\r\n v = map_nested(command, v, **map_nested_kwargs)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/datasets/utils/py_utils.py\", line 216, in map_nested\r\n return function(data_struct)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 847, in command\r\n return torch.tensor(x, **format_kwargs)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/warnings.py\", line 101, in _showwarnmsg\r\n _showwarnmsg_impl(msg)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/warnings.py\", line 30, in _showwarnmsg_impl\r\n file.write(text)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/wandb/sdk/lib/redirect.py\", line 100, in new_write\r\n cb(name, data)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/wandb/sdk/wandb_run.py\", line 729, in _console_callback\r\n self._backend.interface.publish_output(name, data)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/wandb/sdk/interface/interface.py\", line 186, in publish_output\r\n self._publish_output(o)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/wandb/sdk/interface/interface.py\", line 191, in _publish_output\r\n self._publish(rec)\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/site-packages/wandb/sdk/interface/interface.py\", line 510, in _publish\r\n if self._process and not self._process.is_alive():\r\n File \"/home/ad/anaconda3/envs/tfm/lib/python3.6/multiprocessing/process.py\", line 134, in is_alive\r\n assert self._parent_pid == os.getpid(), 'can only test a child process'\r\nAssertionError: can only test a child process\r\n```\r\n\r\nMy workaround was to just comment those lines without looking to much into consecuences:\r\n\r\n```\r\ndef _publish(self, record: pb.Record, local: bool = None) -> None:\r\n #if self._process and not self._process.is_alive():\r\n # raise Exception(\"The wandb backend process has shutdown\")\r\n```\r\n\r\nIt worked so far... I need to try running without wandb and see if it could be causing something wrong with multiprocessing. I am going to try to launch the training setting wandb to false and I will let you know again.",
"@lhoestq But despite this, I got lost into the [class Dataset()](https://huggingface.co/docs/datasets/_modules/datasets/arrow_dataset.html#Dataset) reading the pyarrow files.\r\n\r\nEdit: but you should be rigth, that it does not have to be related to multiprocessing since it keeps happening when `num_workers=0` ",
"Or maybe wandb uses multiprocessing ? One process for wandb logging and one for actual training ? If this is the case then even setting `num_workers=0` would cause the process to be forked for wandb and therefore cause the memory issue.",
"@lhoestq could be, but if we set wandb to false this should not happen. I am going to try.",
"@lhoestq It keeps happening. I have uninstalled wandb from my env, setted `%env WANDB_DISABLED=true` on my notebook, and commented this func:\r\n\r\n```\r\ndef get_available_reporting_integrations():\r\n integrations = []\r\n if is_azureml_available():\r\n integrations.append(\"azure_ml\")\r\n if is_comet_available():\r\n integrations.append(\"comet_ml\")\r\n if is_mlflow_available():\r\n integrations.append(\"mlflow\")\r\n if is_tensorboard_available():\r\n integrations.append(\"tensorboard\")\r\n # if is_wandb_available():\r\n # integrations.append(\"wandb\")\r\n return integrations\r\n```\r\nAs a fast test and it keeps increasing the ram memory. Wandb could not be the blameworthy here.",
"Thanks for checking @gaceladri . Let's investigate the single process setting then.\r\nIf you have some sort of colab notebook with a minimal code example that shows this behavior feel free to share it @gaceladri so that we can play around with it to find what causes this. Otherwise I'll probably try to reproduce on my side at one point",
"@lhoestq sure. Here you have https://colab.research.google.com/drive/1ba09ZOpyHGAOQLcsxiQAHRXl10qnMU5o?usp=sharing let me know if the link works and it reproduces the issue. To me, it reproduces the issue, since if you start the training the ram memory keeps increasing.\r\n\r\nLet me know. Thanks!",
"Could the bug be comming from tokenizers?\r\n\r\nI got this warning at the terminal from my jupyter notebook: \r\n```\r\nhuggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...\r\nTo disable this warning, you can either:\r\n\t- Avoid using `tokenizers` before the fork if possible\r\n\t- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\r\n```",
"I've never experienced memory issues with tokenizers so I don't know\r\nCc @n1t0 are you aware of any issue that would cause memory to keep increasing when the tokenizer is used in the Data Collator for language modeling ?",
"@lhoestq Thanks for pointing to n1t0, just to clarify. That warning was doing fine-tuning, without collator:\r\n```\r\n\r\nfrom datasets import load_dataset, load_metric\r\nimport numpy as np\r\n\r\nGLUE_TASKS = [\r\n \"cola\",\r\n \"mnli\",\r\n \"mnli-mm\",\r\n \"mrpc\",\r\n \"qnli\",\r\n \"qqp\",\r\n \"rte\",\r\n \"sst2\",\r\n \"stsb\",\r\n \"wnli\",\r\n]\r\ntask = \"mnli\"\r\nactual_task = \"mnli\" if task == \"mnli-mm\" else task\r\ndataset = load_dataset(\"glue\", actual_task)\r\nmetric = load_metric(\"glue\", actual_task)\r\nbatch_size = 16\r\nattention_type = \"linear\"\r\n\r\nfrom transformers.models.mobilebert_mod import (\r\n MobileBertForSequenceClassification,\r\n MobileBertTokenizerFast,\r\n)\r\nfrom transformers.models.mobilebert_mod.configuration_mobilebert import (\r\n MobileBertConfigMod,\r\n)\r\nfrom transformers import TrainingArguments, Trainer\r\n\r\nnum_labels = 3 if task.startswith(\"mnli\") else 1 if task == \"stsb\" else 2\r\ntokenizer = MobileBertTokenizerFast.from_pretrained(\r\n \"/media/ad/00b5422b-9d54-4449-8b5d-08eab5cdac8c/training_trfm/big_linear_layerdrop_shared/checkpoint-23000/\",\r\n max_len=512,\r\n)\r\nmodel = MobileBertForSequenceClassification.from_pretrained(\r\n \"/media/ad/00b5422b-9d54-4449-8b5d-08eab5cdac8c/training_trfm/big_linear_layerdrop_shared/checkpoint-23000/\",\r\n num_labels=num_labels,\r\n)\r\nprint(model.num_parameters())\r\n\r\ntask_to_keys = {\r\n \"cola\": (\"sentence\", None),\r\n \"mnli\": (\"premise\", \"hypothesis\"),\r\n \"mnli-mm\": (\"premise\", \"hypothesis\"),\r\n \"mrpc\": (\"sentence1\", \"sentence2\"),\r\n \"qnli\": (\"question\", \"sentence\"),\r\n \"qqp\": (\"question1\", \"question2\"),\r\n \"rte\": (\"sentence1\", \"sentence2\"),\r\n \"sst2\": (\"sentence\", None),\r\n \"stsb\": (\"sentence1\", \"sentence2\"),\r\n \"wnli\": (\"sentence1\", \"sentence2\"),\r\n}\r\n\r\nsentence1_key, sentence2_key = task_to_keys[task]\r\nif sentence2_key is None:\r\n print(f\"Sentence: {dataset['train'][0][sentence1_key]}\")\r\nelse:\r\n print(f\"Sentence 1: {dataset['train'][0][sentence1_key]}\")\r\n print(f\"Sentence 2: {dataset['train'][0][sentence2_key]}\")\r\n\r\n\r\ndef preprocess_function(examples):\r\n if sentence2_key is None:\r\n return tokenizer(examples[sentence1_key], truncation=True)\r\n return tokenizer(examples[sentence1_key], examples[sentence2_key], truncation=True)\r\n\r\n\r\nencoded_dataset = dataset.map(preprocess_function, batched=True)\r\nmetric_name = (\r\n \"pearson\"\r\n if task == \"stsb\"\r\n else \"matthews_correlation\"\r\n if task == \"cola\"\r\n else \"accuracy\"\r\n)\r\n\r\nargs = TrainingArguments(\r\n f\"test-glue/{task}_{attention_type}\",\r\n evaluation_strategy=\"steps\",\r\n learning_rate=1e-5,\r\n per_device_train_batch_size=batch_size,\r\n per_device_eval_batch_size=batch_size,\r\n logging_steps=200,\r\n num_train_epochs=5,\r\n gradient_accumulation_steps=1,\r\n warmup_steps=10000,\r\n fp16=True,\r\n dataloader_num_workers=10,\r\n weight_decay=0.1,\r\n load_best_model_at_end=True,\r\n metric_for_best_model=metric_name,\r\n)\r\n\r\n\r\ndef compute_metrics(eval_pred):\r\n predictions, labels = eval_pred\r\n if task != \"stsb\":\r\n predictions = np.argmax(predictions, axis=1)\r\n else:\r\n predictions = predictions[:, 0]\r\n return metric.compute(predictions=predictions, references=labels)\r\n\r\n\r\nvalidation_key = (\r\n \"validation_mismatched\"\r\n if task == \"mnli-mm\"\r\n else \"validation_matched\"\r\n if task == \"mnli\"\r\n else \"validation\"\r\n)\r\n\r\ntrainer = Trainer(\r\n model,\r\n args,\r\n train_dataset=encoded_dataset[\"train\"],\r\n eval_dataset=encoded_dataset[validation_key],\r\n tokenizer=tokenizer,\r\n compute_metrics=compute_metrics,\r\n)\r\n\r\ntrainer.train()\r\n```\r\n\r\nNow, I have come back to pre-training. The changes that I think I have done are: not formatting the dataset to torch: ~~`big_dataset.set_format(type='torch', columns=[\"text\", \"input_ids\", \"attention_mask\", \"token_type_ids\"])`~~ so maybe some column is dropped and not freezed in memory and now I have not setted any validation dataset in the trainer. \r\n\r\nMy validation dataset before:\r\n```\r\nbook_corpus_eval = load_dataset(\r\n \"bookcorpus\",\r\n \"plain_text\",\r\n cache_dir=\"/home/ad/Desktop/bookcorpus\",\r\n split=\"train[98:99%]\",\r\n)\r\nbook_corpus_eval = book_corpus_eval.map(encode, batched=True)\r\nbook_corpus_eval.set_format(\r\n type=\"torch\", columns=[\"text\", \"input_ids\", \"attention_mask\", \"token_type_ids\"]\r\n)\r\n**book_corpus_eval = book_corpus_eval.select([i for i in range(1500)])**\r\n```\r\nMaybe _selecting_ or indexing the dataset before feeding it to the trainer, do something strange.\r\n\r\nMy trainer now:\r\n```\r\n\r\nbig_dataset = load_from_disk(\"/home/ad/Desktop/35percent_data.arrow/\")\r\n\r\nfrom transformers import DataCollatorForWholeWordMask\r\n\r\ndata_collator = DataCollatorForWholeWordMask(\r\n tokenizer=tokenizer, mlm=True, mlm_probability=0.15)\r\n\r\nfrom transformers import Trainer, TrainingArguments\r\n\r\ntraining_args = TrainingArguments(\r\n output_dir=\"./big_linear_layerdrop_shared_silu_secondtry\",\r\n overwrite_output_dir=True,\r\n per_device_train_batch_size=60,\r\n per_device_eval_batch_size=60,\r\n save_steps=500,\r\n save_total_limit=10,\r\n logging_first_step=True,\r\n logging_steps=100,\r\n# evaluation_strategy='steps',\r\n# eval_steps=250,\r\n gradient_accumulation_steps=8,\r\n fp16=True,\r\n dataloader_num_workers=10,\r\n warmup_steps=15000,\r\n learning_rate=6e-4,\r\n adam_epsilon=1e-6,\r\n adam_beta2=0.98,\r\n weight_decay=0.01,\r\n max_grad_norm=1.0,\r\n max_steps=500000, \r\n)\r\n\r\ntrainer = Trainer(\r\n model=model,\r\n args=training_args,\r\n data_collator=data_collator,\r\n train_dataset=big_dataset,\r\n# eval_dataset=book_corpus_eval,\r\n tokenizer=tokenizer)\r\n\r\nimport wandb\r\nwandb.login()\r\n\r\ntrainer.train()\r\n```\r\n\r\nAnd surprisingly, the ram now keeps going up and down. The training is up now for 12h without collapse the ram. I don't know what could cause the leakage. :mag: \r\n\r\nEdit: I didn't see the swap memory, that keeps increasing. So the problem persist. ",
"Thanks for sharing your results.\r\nSo you still had the issue for fine-tuning ?\r\nAnd the issue still appears with a bare-bone dataset from an arrow file...",
"Yes, on both cases. Fine-tuning a pre-trained model and pre-training from scratch with a local arrow file already pre-processed."
] | I tried to pretrain Longformer using transformers and datasets. But I got OOM issues with loading a large text file. My script is almost like this:
```python
from datasets import load_dataset
@dataclass
class DataCollatorForDatasetsLanguageModeling(DataCollatorForLanguageModeling):
"""
Data collator used for language modeling based on DataCollatorForLazyLanguageModeling
- collates batches of tensors, honoring their tokenizer's pad_token
- preprocesses batches for masked language modeling
"""
block_size: int = 512
def __call__(self, examples: List[dict]) -> Dict[str, torch.Tensor]:
examples = [example['text'] for example in examples]
batch, attention_mask = self._tensorize_batch(examples)
if self.mlm:
inputs, labels = self.mask_tokens(batch)
return {"input_ids": inputs, "labels": labels}
else:
labels = batch.clone().detach()
if self.tokenizer.pad_token_id is not None:
labels[labels == self.tokenizer.pad_token_id] = -100
return {"input_ids": batch, "labels": labels}
def _tensorize_batch(self, examples: List[str]) -> Tuple[torch.Tensor, torch.Tensor]:
if self.tokenizer._pad_token is None:
raise ValueError(
"You are attempting to pad samples but the tokenizer you are using"
f" ({self.tokenizer.__class__.__name__}) does not have one."
)
tensor_examples = self.tokenizer.batch_encode_plus(
[ex for ex in examples if ex],
max_length=self.block_size,
return_tensors="pt",
pad_to_max_length=True,
return_attention_mask=True,
truncation=True,
)
input_ids, attention_mask = tensor_examples["input_ids"], tensor_examples["attention_mask"]
return input_ids, attention_mask
dataset = load_dataset('text', data_files='train.txt',cache_dir="./", , split='train')
data_collator = DataCollatorForDatasetsLanguageModeling(tokenizer=tokenizer, mlm=True,
mlm_probability=0.15, block_size=tokenizer.max_len)
trainer = Trainer(model=model, args=args, data_collator=data_collator,
train_dataset=train_dataset, prediction_loss_only=True, )
trainer.train(model_path=model_path)
```
This train.txt is about 1.1GB and has 90k lines where each line is a sequence of 4k words.
During training, the memory usage increased fast as the following graph and resulted in OOM before the finish of training.
Could you please give me any suggestions on why this happened and how to fix it?
Thanks. | 633 |
https://github.com/huggingface/datasets/issues/630 | Text dataset not working with large files | [
"Seems like it works when setting ```block_size=2100000000``` or something arbitrarily large though.",
"Can you give us some stats on the data files you use as inputs?",
"Basically ~600MB txt files(UTF-8) * 59. \r\ncontents like ```μλ
νμΈμ, μ΄κ²μ μμ λ‘ νλ² λ§ν΄λ³΄λ ν
μ€νΈμ
λλ€. κ·Έλ₯ μ΄λ λ€κ³ μ.<|endoftext|>\\n```\r\n\r\nAlso, it gets stuck for a loooong time at ```Testing the mapped function outputs```, for more than 12 hours(currently ongoing)",
"It gets stuck while doing `.map()` ? Are you using multiprocessing ?\r\nIf you could provide a code snippet it could be very useful",
"From transformers/examples/language-modeling/run-language-modeling.py :\r\n```\r\ndef get_dataset(\r\n args: DataTrainingArguments,\r\n tokenizer: PreTrainedTokenizer,\r\n evaluate: bool = False,\r\n cache_dir: Optional[str] = None,\r\n):\r\n file_path = args.eval_data_file if evaluate else args.train_data_file\r\n if True:\r\n dataset = load_dataset(\"text\", data_files=glob.glob(file_path), split='train', use_threads=True, \r\n ignore_verifications=True, save_infos=True, block_size=104857600)\r\n dataset = dataset.map(lambda ex: tokenizer(ex[\"text\"], add_special_tokens=True,\r\n truncation=True, max_length=args.block_size), batched=True)\r\n dataset.set_format(type='torch', columns=['input_ids'])\r\n return dataset\r\n if args.line_by_line:\r\n return LineByLineTextDataset(tokenizer=tokenizer, file_path=file_path, block_size=args.block_size)\r\n else:\r\n return TextDataset(\r\n tokenizer=tokenizer,\r\n file_path=file_path,\r\n block_size=args.block_size,\r\n overwrite_cache=args.overwrite_cache,\r\n cache_dir=cache_dir,\r\n )\r\n```\r\n\r\nNo, I'm not using multiprocessing.",
"I am not able to reproduce on my side :/\r\n\r\nCould you send the version of `datasets` and `pyarrow` you're using ?\r\nCould you try to update the lib and try again ?\r\nOr do you think you could try to reproduce it on google colab ?",
"Huh, weird. It's fixed on my side too.\r\nBut now ```Caching processed dataset``` is taking forever - how can I disable it? Any flags?",
"Right after `Caching processed dataset`, your function is applied to the dataset and there's a progress bar that shows how much time is left. How much time does it take for you ?\r\n\r\nAlso caching isn't supposed to slow down your processing. But if you still want to disable it you can do `.map(..., load_from_cache_file=False)`",
"Ah, itβs much faster now(Takes around 15~20min). \r\nBTW, any way to set default tensor output as plain tensors with distributed training? The ragged tensors are incompatible with tpustrategy :(",
"> Ah, itβs much faster now(Takes around 15~20min).\r\n\r\nGlad to see that it's faster now. What did you change exactly ?\r\n\r\n> BTW, any way to set default tensor output as plain tensors with distributed training? The ragged tensors are incompatible with tpustrategy :(\r\n\r\nOh I didn't know about that. Feel free to open an issue to mention that.\r\nI guess what you can do for now is set the dataset format to numpy instead of tensorflow, and use a wrapper of the dataset that converts the numpy arrays to tf tensors.\r\n\r\n",
">>> Glad to see that it's faster now. What did you change exactly ?\r\nI don't know, it just worked...? Sorry I couldn't be more helpful.\r\n\r\nSetting with numpy array is a great idea! Thanks."
] | ```
Traceback (most recent call last):
File "examples/language-modeling/run_language_modeling.py", line 333, in <module>
main()
File "examples/language-modeling/run_language_modeling.py", line 262, in main
get_dataset(data_args, tokenizer=tokenizer, cache_dir=model_args.cache_dir) if training_args.do_train else None
File "examples/language-modeling/run_language_modeling.py", line 144, in get_dataset
dataset = load_dataset("text", data_files=file_path, split='train+test')
File "/home/ksjae/.local/lib/python3.7/site-packages/datasets/load.py", line 611, in load_dataset
ignore_verifications=ignore_verifications,
File "/home/ksjae/.local/lib/python3.7/site-packages/datasets/builder.py", line 469, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/ksjae/.local/lib/python3.7/site-packages/datasets/builder.py", line 546, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/ksjae/.local/lib/python3.7/site-packages/datasets/builder.py", line 888, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False, disable=not_verbose):
File "/home/ksjae/.local/lib/python3.7/site-packages/tqdm/std.py", line 1129, in __iter__
for obj in iterable:
File "/home/ksjae/.cache/huggingface/modules/datasets_modules/datasets/text/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7/text.py", line 104, in _generate_tables
convert_options=self.config.convert_options,
File "pyarrow/_csv.pyx", line 714, in pyarrow._csv.read_csv
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
```
**pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)**
It gives the same message for both 200MB, 10GB .tx files but not for 700MB file.
Can't upload due to size & copyright problem. sorry. | 630 |
https://github.com/huggingface/datasets/issues/629 | straddling object straddles two block boundaries | [
"sorry it's an apache arrow issue."
] | I am trying to read json data (it's an array with lots of dictionaries) and getting block boundaries issue as below :
I tried calling read_json with readOptions but no luck .
```
table = json.read_json(fn)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "pyarrow/_json.pyx", line 246, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)
```
| 629 |
https://github.com/huggingface/datasets/issues/625 | dtype of tensors should be preserved | [
"Indeed we convert tensors to list to be able to write in arrow format. Because of this conversion we lose the dtype information. We should add the dtype detection when we do type inference. However it would require a bit of refactoring since currently the conversion happens before the type inference..\r\n\r\nAnd then for your information, when reading from arrow format we have to cast from arrow to numpy (which is fast since pyarrow has a numpy integration), and then to torch.\r\n\r\nHowever there's one thing that can help you: we make sure that the dtypes correspond to what is defined in `features`.\r\nTherefore what you can do is provide `features` in `.map(preprocess, feature=...)` to specify the output types.\r\n\r\nFor example in your case:\r\n```python\r\nfrom datasets import Features, Value, Sequence\r\n\r\nfeatures = Features({\r\n \"input_ids\": Sequence(Value(\"int32\")),\r\n \"sembedding\": Sequence(Value(\"float32\"))\r\n})\r\npreprocessed_dataset = dataset.map(preprocess, features=features)\r\n\r\npreprocessed_dataset.set_format(\"torch\", columns=[\"input_ids\", \"sembedding\"])\r\nprint(preprocessed_dataset[0][\"sembedding\"].dtype)\r\n# \"torch.float32\"\r\n```\r\n\r\nLet me know if it helps",
"If the arrow format is basically lists, why is the intermediate step to numpy necessary? I am a bit confused about that part.\r\n\r\nThanks for your suggestion. as I have currently implemented this, I cast to torch.Tensor in my collate_fn to save disk space (so I do not have to save padded tensors to max_len but can pad up to max batch len in collate_fn) at the cost of a bit slower processing. So for me this is not relevant anymore, but I am sure it is for others!",
"I'm glad you managed to figure something out :)\r\n\r\nCasting from arrow to numpy can be 100x faster than casting from arrow to list.\r\nThis is because arrow has an integration with numpy that allows it to instantiate numpy arrays with zero-copy from arrow.\r\nOn the other hand to create python lists it is slow since it has to recreate the list object by iterating through each element in python.",
"Ah that is interesting. I have no direct experience with arrow so I didn't know. ",
"I encountered a simliar issue: `datasets` converted my float numpy array to `torch.float64` tensors, while many pytorch operations require `torch.float32` inputs and it's very troublesome. \r\n\r\nI tried @lhoestq 's solution, but since it's mixed with the preprocess function, it's not very intuitive. \r\n\r\nI just want to share another possible simpler solution: directly cast the dtype of the processed dataset.\r\n\r\nNow I want to change the type of `labels` in `train_dataset` from float64 to float32, I can do this.\r\n\r\n```\r\nfrom datasets import Value, Sequence, Features\r\nfeats = train_dataset.features.copy()\r\nfeats['labels'].feature = Value(dtype='float32')\r\nfeats = Features(feats)\r\ntrain_dataset.cast_(feats)\r\n```\r\n",
"Reopening since @bhavitvyamalik started looking into it !\r\n\r\nAlso I'm posting here a function that could be helpful to support preserving the dtype of tensors.\r\n\r\nIt's used to build a pyarrow array out of a numpy array and:\r\n- it doesn't convert the numpy array to a python list\r\n- it keeps the precision of the numpy array for the pyarrow array\r\n- it works with multidimensional arrays (while `pa.array` can only take a 1D array as input)\r\n- it builds the pyarrow ListArray from offsets created on-the-fly and values that come from the flattened numpy array\r\n\r\n```python\r\nfrom functools import reduce\r\nfrom operator import mul\r\n\r\nimport numpy as np\r\nimport pyarrow as pa\r\n\r\ndef pa_ndarray(a):\r\n \"\"\"Build a PyArrow ListArray from a multidimensional NumPy array\"\"\"\r\n values = pa.array(a.flatten()) \r\n for i in range(a.ndim - 1): \r\n n_offsets = reduce(mul, a.shape[:a.ndim - i - 1], 1) \r\n step_offsets = a.shape[a.ndim - i - 1] \r\n offsets = pa.array(np.arange(n_offsets + 1) * step_offsets, type=pa.int32()) \r\n values = pa.ListArray.from_arrays(offsets, values) \r\n return values \r\n\r\nnarr = np.arange(42).reshape(7, 2, 3).astype(np.uint8)\r\nparr = pa_ndarray(narr)\r\nassert isinstance(parr, pa.Array)\r\nassert parr.type == pa.list_(pa.list_(pa.uint8()))\r\nassert narr.tolist() == parr.to_pylist()\r\n```\r\n\r\nThe only costly operation is the offsets computations. Since it doesn't iterate on the numpy array values this function is pretty fast.",
"@lhoestq Have you thought about this further?\r\n\r\nWe have a use case where we're attempting to load data containing numpy arrays using the `datasets` library.\r\n\r\nWhen using one of the \"standard\" methods (`[Value(...)]` or `Sequence()`) we see ~200 samples processed per second during the call to `_prepare_split`. This slowdown is caused by the vast number of calls to `encode_nested_example` (each sequence is converted to a list, and each element in the sequence...). \r\n\r\nUsing the `Feature` `ArrayND` improves this somewhat to ~500/s as it now uses numpy's `tolist()` rather than iterating over each value in the array and converting them individually.\r\n\r\nHowever, it's still pretty slow and in theory it should be possible to avoid the `numpy -> python -> arrow` dance altogether. To demonstrate this, if you keep the `Feature` set to an `ArrayND` but instead return a `pa_ndarray(...)` in `_generate_examples` it skips the conversion (`return obj, False`) and hits ~11_000/s. Two orders of magnitude speed up! The problem is this then fails later on when the `ArrowWriter` tries to write the examples to disk :-( \r\n\r\nIt would be nice to have first-class support for user-defined PyArrow objects. Is this a possibility? We have _large_ datasets where even an order of magnitude difference is important so settling on the middle ~500/s is less than ideal! \r\n\r\nIs there a workaround for this or another method that should be used instead that gets near-to or equal performance to returning PyArrow arrays?",
"Note that manually generating the table using `pyarrow` achieves ~30_000/s",
"Hi !\r\n\r\nIt would be awesome to achieve this speed for numpy arrays !\r\nFor now we have to use `encode_nested_example` to convert numpy arrays to python lists since pyarrow doesn't support multidimensional numpy arrays (only 1D).\r\n\r\nMaybe let's start a new PR from your PR @bhavitvyamalik (idk why we didn't answer your PR at that time, sorry about that).\r\nBasically the idea is to allow `TypedSequence` to support numpy arrays as you did, and remove the numpy->python casting in `_cast_to_python_objects`.\r\n\r\nThis is really important since we are starting to have a focus on other modalities than text as well (audio, images).\r\n\r\nThough until then @samgd, there is another feature that may interest you and that may give you the speed you want:\r\n\r\nIn a dataset script you can subclass either a GeneratorBasedBuilder (with the `_generate_examples ` method) or an ArrowBasedBuilder if you want. the ArrowBasedBuilder allows to yield arrow data by implementing the `_generate_tables` method (it's the same as `_generate_examples` except you must yield arrow tables). Since the data are already in arrow format, it doesn't call `encode_nested_example`. Let me know if that helps."
] | After switching to `datasets` my model just broke. After a weekend of debugging, the issue was that my model could not handle the double that the Dataset provided, as it expected a float (but didn't give a warning, which seems a [PyTorch issue](https://discuss.pytorch.org/t/is-it-required-that-input-and-hidden-for-gru-have-the-same-dtype-float32/96221)).
As a user I did not expect this bug. I have a `map` function that I call on the Dataset that looks like this:
```python
def preprocess(sentences: List[str]):
token_ids = [[vocab.to_index(t) for t in s.split()] for s in sentences]
sembeddings = stransformer.encode(sentences)
print(sembeddings.dtype)
return {"input_ids": token_ids, "sembedding": sembeddings}
```
Given a list of `sentences` (`List[str]`), it converts those into token_ids on the one hand (list of lists of ints; `List[List[int]]`) and into sentence embeddings on the other (Tensor of dtype `torch.float32`). That means that I actually set the column "sembedding" to a tensor that I as a user expect to be a float32.
It appears though that behind the scenes, this tensor is converted into a **list**. I did not find this documented anywhere but I might have missed it. From a user's perspective this is incredibly important though, because it means you cannot do any data_type or tensor casting yourself in a mapping function! Furthermore, this can lead to issues, as was my case.
My model expected float32 precision, which I thought `sembedding` was because that is what `stransformer.encode` outputs. But behind the scenes this tensor is first cast to a list, and when we then set its format, as below, this column is cast not to float32 but to double precision float64.
```python
dataset.set_format(type="torch", columns=["input_ids", "sembedding"])
```
This happens because apparently there is an intermediate step of casting to a **numpy** array (?) **whose dtype creation/deduction is different from torch dtypes** (see the snippet below). As you can see, this means that the dtype is not preserved: if I got it right, the dataset goes from torch.float32 -> list -> float64 (numpy) -> torch.float64.
```python
import torch
import numpy as np
l = [-0.03010837361216545, -0.035979013890028, -0.016949838027358055]
torch_tensor = torch.tensor(l)
np_array = np.array(l)
np_to_torch = torch.from_numpy(np_array)
print(torch_tensor.dtype)
# torch.float32
print(np_array.dtype)
# float64
print(np_to_torch.dtype)
# torch.float64
```
This might lead to unwanted behaviour. I understand that the whole library is probably built around casting from numpy to other frameworks, so this might be difficult to solve. Perhaps `set_format` should include a `dtypes` option where for each input column the user can specify the wanted precision.
The alternative is that the user needs to cast manually after loading data from the dataset but that does not seem user-friendly, makes the dataset less portable, and might use more space in memory as well as on disk than is actually needed. | 625 |
https://github.com/huggingface/datasets/issues/624 | Add learningq dataset | [] | Hi,
Thank you again for this amazing repo.
Would it be possible for y'all to add the LearningQ dataset - https://github.com/AngusGLChen/LearningQ ?
| 624 |
https://github.com/huggingface/datasets/issues/623 | Custom feature types in `load_dataset` from CSV | [
"Currently `csv` doesn't support the `features` attribute (unlike `json`).\r\nWhat you can do for now is cast the features using the in-place transform `cast_`\r\n\r\n```python\r\nfrom datasets import load_dataset\r\n\r\ndataset = load_dataset('csv', data_files=file_dict, delimiter=';', column_names=['text', 'label'])\r\ndataset.cast_(emotion_features)\r\n```\r\n",
"Thanks for the clarification!",
"Hi @lhoestq we've tried out your suggestion but are now running into the following error:\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nValueError Traceback (most recent call last)\r\n<ipython-input-163-81ffd5ac18c9> in <module>\r\n----> 1 dataset.cast_(emotion_features)\r\n\r\n/usr/local/lib/python3.6/dist-packages/datasets/dataset_dict.py in cast_(self, features)\r\n 125 self._check_values_type()\r\n 126 for dataset in self.values():\r\n--> 127 dataset.cast_(features=features)\r\n 128 \r\n 129 def remove_columns_(self, column_names: Union[str, List[str]]):\r\n\r\n/usr/local/lib/python3.6/dist-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)\r\n 161 # Call actual function\r\n 162 \r\n--> 163 out = func(self, *args, **kwargs)\r\n 164 \r\n 165 # Update fingerprint of in-place transforms + update in-place history of transforms\r\n\r\n/usr/local/lib/python3.6/dist-packages/datasets/arrow_dataset.py in cast_(self, features)\r\n 602 self._info.features = features\r\n 603 schema = pa.schema(features.type)\r\n--> 604 self._data = self._data.cast(schema)\r\n 605 \r\n 606 @fingerprint(inplace=True)\r\n\r\n/usr/local/lib/python3.6/dist-packages/pyarrow/table.pxi in pyarrow.lib.Table.cast()\r\n\r\nValueError: Target schema's field names are not matching the table's field names: ['text', 'label'], ['label', 'text']\r\n```\r\n\r\nLooking at the types in `emotion_features` we see that `label` and `text` appear to be swapped in the Arrow table:\r\n\r\n```\r\nemotion_features.type\r\nStructType(struct<label: int64, text: string>)\r\n```\r\n\r\nDid we define the `emotion_features` incorrectly? We just followed the instructions from the [docs](https://huggingface.co/docs/datasets/features.html?highlight=features#dataset-features), but perhaps we misunderstood something π¬ \r\n\r\n",
"In general, I don't think there is any hard reason we don't allow to use `features` in the csv script, right @lhoestq?\r\n\r\nShould I add it?",
"> In general, I don't think there is any hard reason we don't allow to use `features` in the csv script, right @lhoestq?\r\n> \r\n> Should I add it?\r\n\r\nSure let's add it. Setting the convert options should do the job\r\n\r\n> Hi @lhoestq we've tried out your suggestion but are now running into the following error:\r\n> \r\n> ```\r\n> ---------------------------------------------------------------------------\r\n> ValueError Traceback (most recent call last)\r\n> <ipython-input-163-81ffd5ac18c9> in <module>\r\n> ----> 1 dataset.cast_(emotion_features)\r\n>\r\n> /usr/local/lib/python3.6/dist-packages/pyarrow/table.pxi in pyarrow.lib.Table.cast()\r\n> \r\n> ValueError: Target schema's field names are not matching the table's field names: ['text', 'label'], ['label', 'text']\r\n> ```\r\n>\r\n> Did we define the `emotion_features` incorrectly? We just followed the instructions from the [docs](https://huggingface.co/docs/datasets/features.html?highlight=features#dataset-features), but perhaps we misunderstood something π¬\r\n\r\nThanks for reporting, that's a bug :) I'm fixing it right now",
"PR is open for the `ValueError: Target schema's field names are not matching the table's field names` error.\r\n\r\nI'm adding the features parameter to csv",
"Thanks a lot for the PR and quick fix @lhoestq!"
] | I am trying to load a local file with the `load_dataset` function and I want to predefine the feature types with the `features` argument. However, the types are always the same independent of the value of `features`.
I am working with the local files from the emotion dataset. To get the data you can use the following code:
```Python
from pathlib import Path
import wget
EMOTION_PATH = Path("./data/emotion")
DOWNLOAD_URLS = [
"https://www.dropbox.com/s/1pzkadrvffbqw6o/train.txt?dl=1",
"https://www.dropbox.com/s/2mzialpsgf9k5l3/val.txt?dl=1",
"https://www.dropbox.com/s/ikkqxfdbdec3fuj/test.txt?dl=1",
]
if not Path.is_dir(EMOTION_PATH):
Path.mkdir(EMOTION_PATH)
for url in DOWNLOAD_URLS:
wget.download(url, str(EMOTION_PATH))
```
The first five lines of the train set are:
```
i didnt feel humiliated;sadness
i can go from feeling so hopeless to so damned hopeful just from being around someone who cares and is awake;sadness
im grabbing a minute to post i feel greedy wrong;anger
i am ever feeling nostalgic about the fireplace i will know that it is still on the property;love
i am feeling grouchy;anger
```
Here the code to reproduce the issue:
```Python
from datasets import Features, Value, ClassLabel, load_dataset
class_names = ["sadness", "joy", "love", "anger", "fear", "surprise"]
emotion_features = Features({'text': Value('string'), 'label': ClassLabel(names=class_names)})
file_dict = {'train': EMOTION_PATH/'train.txt'}
dataset = load_dataset('csv', data_files=file_dict, delimiter=';', column_names=['text', 'label'], features=emotion_features)
```
**Observed behaviour:**
```Python
dataset['train'].features
```
```Python
{'text': Value(dtype='string', id=None),
'label': Value(dtype='string', id=None)}
```
**Expected behaviour:**
```Python
dataset['train'].features
```
```Python
{'text': Value(dtype='string', id=None),
'label': ClassLabel(num_classes=6, names=['sadness', 'joy', 'love', 'anger', 'fear', 'surprise'], names_file=None, id=None)}
```
**Things I've tried:**
- deleting the cache
- trying other types such as `int64`
Am I missing anything? Thanks for any pointer in the right direction. | 623 |
https://github.com/huggingface/datasets/issues/622 | load_dataset for text files not working | [
"Can you give us more information on your os and pip environments (pip list)?",
"@thomwolf Sure. I'll try downgrading to 3.7 now even though Arrow say they support >=3.5.\r\n\r\nLinux (Ubuntu 18.04) - Python 3.8\r\n======================\r\nPackage - Version\r\n---------------------\r\ncertifi 2020.6.20\r\nchardet 3.0.4\r\nclick 7.1.2\r\ndatasets 1.0.1\r\ndill 0.3.2\r\nfasttext 0.9.2\r\nfilelock 3.0.12\r\nfuture 0.18.2\r\nidna 2.10\r\njoblib 0.16.0\r\nnltk 3.5\r\nnumpy 1.19.1\r\npackaging 20.4\r\npandas 1.1.2\r\npip 20.0.2\r\nprotobuf 3.13.0\r\npyarrow 1.0.1\r\npybind11 2.5.0\r\npyparsing 2.4.7\r\npython-dateutil 2.8.1\r\npytz 2020.1\r\nregex 2020.7.14\r\nrequests 2.24.0\r\nsacremoses 0.0.43\r\nscikit-learn 0.23.2\r\nscipy 1.5.2\r\nsentence-transformers 0.3.6\r\nsentencepiece 0.1.91\r\nsetuptools 46.1.3\r\nsix 1.15.0\r\nstanza 1.1.1\r\nthreadpoolctl 2.1.0\r\ntokenizers 0.8.1rc2\r\ntorch 1.6.0+cu101\r\ntqdm 4.48.2\r\ntransformers 3.1.0\r\nurllib3 1.25.10\r\nwheel 0.34.2\r\nxxhash 2.0.0\r\n\r\nWindows 10 - Python 3.8\r\n================\r\nPackage - Version\r\n----------------------------\r\ncertifi 2020.6.20\r\nchardet 3.0.4\r\nclick 7.1.2\r\ndatasets 1.0.1\r\ndill 0.3.2\r\nfasttext 0.9.2\r\nfilelock 3.0.12\r\nfuture 0.18.2\r\nidna 2.10\r\njoblib 0.16.0\r\nnlp 0.4.0\r\nnltk 3.5\r\nnumpy 1.19.1\r\npackaging 20.4\r\npandas 1.1.1\r\npip 20.0.2\r\nprotobuf 3.13.0\r\npyarrow 1.0.1\r\npybind11 2.5.0\r\npyparsing 2.4.7\r\npython-dateutil 2.8.1\r\npytz 2020.1\r\nregex 2020.7.14\r\nrequests 2.24.0\r\nsacremoses 0.0.43\r\nscikit-learn 0.23.2\r\nscipy 1.5.2\r\nsentence-transformers 0.3.5.1\r\nsentencepiece 0.1.91\r\nsetuptools 46.1.3\r\nsix 1.15.0\r\nstanza 1.1.1\r\nthreadpoolctl 2.1.0\r\ntokenizers 0.8.1rc1\r\ntorch 1.6.0+cu101\r\ntqdm 4.48.2\r\ntransformers 3.0.2\r\nurllib3 1.25.10\r\nwheel 0.34.2\r\nxxhash 2.0.0",
"Downgrading to 3.7 does not help. Here is a dummy text file:\r\n\r\n```text\r\nVerzekering weigert vaker te betalen\r\nBedrijven van verzekeringen erkennen steeds minder arbeidsongevallen .\r\nIn 2012 weigerden de bedrijven te betalen voor 21.055 ongevallen op het werk .\r\nDat is 11,8 % van alle ongevallen op het werk .\r\nNog nooit weigerden verzekeraars zoveel zaken .\r\nIn 2012 hadden 135.118 mensen een ongeval op het werk .\r\nDat zijn elke werkdag 530 mensen .\r\nBij die ongevallen stierven 67 mensen .\r\nBijna 12.000 hebben een handicap na het ongeval .\r\nGeen echt arbeidsongeval Bedrijven moeten een verzekering hebben voor hun werknemers .\r\n```\r\n\r\nA temporary work around for the \"text\" type, is\r\n\r\n```python\r\ndataset = Dataset.from_dict({\"text\": Path(dataset_f).read_text().splitlines()})\r\n```",
"\r\n\r\neven i am facing the same issue.",
"@banunitte Please do not post screenshots in the future but copy-paste your code and the errors. That allows others to copy-and-paste your code and test it. You may also want to provide the Python version that you are using.",
"I have the exact same problem in Windows 10, Python 3.8.\r\n",
"I have the same problem on Linux of the script crashing with a CSV error. This may be caused by 'CRLF', when changed 'CRLF' to 'LF', the problem solved.",
"I pushed a fix for `pyarrow.lib.ArrowInvalid: CSV parse error`. Let me know if you still have this issue.\r\n\r\nNot sure about the windows one yet",
"To complete what @lhoestq is saying, I think that to use the new version of the `text` processing script (which is on master right now) you need to either specify the version of the script to be the `master` one or to install the lib from source (in which case it uses the `master` version of the script by default):\r\n```python\r\ndataset = load_dataset('text', script_version='master', data_files=XXX)\r\n```\r\nWe do versioning by default, i.e. your version of the dataset lib will use the script with the same version by default (i.e. only the `1.0.1` version of the script if you have the PyPI version `1.0.1` of the lib).",
"\r\nwin10, py3.6\r\n\r\n\r\n```\r\nfrom datasets import Features, Value, ClassLabel, load_dataset\r\n\r\n\r\nfeatures = Features({'text': Value('string'), 'ctext': Value('string')})\r\nfile_dict = {'train': PATH/'summary.csv'}\r\n\r\ndataset = load_dataset('csv', data_files=file_dict, script_version='master', delimiter='\\t', column_names=['text', 'ctext'], features=features)\r\n```",
"```python\r\nTraceback` (most recent call last):\r\n File \"main.py\", line 281, in <module>\r\n main()\r\n File \"main.py\", line 190, in main\r\n train_data, test_data = data_factory(\r\n File \"main.py\", line 129, in data_factory\r\n train_data = load_dataset('text', \r\n File \"/home/me/Downloads/datasets/src/datasets/load.py\", line 608, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/home/me/Downloads/datasets/src/datasets/builder.py\", line 468, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/me/Downloads/datasets/src/datasets/builder.py\", line 546, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/home/me/Downloads/datasets/src/datasets/builder.py\", line 888, in _prepare_split\r\n for key, table in utils.tqdm(generator, unit=\" tables\", leave=False, disable=not_verbose):\r\n File \"/home/me/.local/lib/python3.8/site-packages/tqdm/std.py\", line 1130, in __iter__\r\n for obj in iterable:\r\n File \"/home/me/.cache/huggingface/modules/datasets_modules/datasets/text/512f465342e4f4cd07a8791428a629c043bb89d55ad7817cbf7fcc649178b014/text.py\", line 103, in _generate_tables\r\n pa_table = pac.read_csv(\r\n File \"pyarrow/_csv.pyx\", line 617, in pyarrow._csv.read_csv\r\n File \"pyarrow/error.pxi\", line 123, in pyarrow.lib.pyarrow_internal_check_status\r\n File \"pyarrow/error.pxi\", line 85, in pyarrow.lib.check_status\r\npyarrow.lib.ArrowInvalid: CSV parse error: Expected 1 columns, got 2\r\n```\r\n\r\nUnfortunately i am still getting this issue on Linux. I installed datasets from source and specified script_version to master.\r\n\r\n",
"> \r\n> win10, py3.6\r\n> \r\n> ```\r\n> from datasets import Features, Value, ClassLabel, load_dataset\r\n> \r\n> \r\n> features = Features({'text': Value('string'), 'ctext': Value('string')})\r\n> file_dict = {'train': PATH/'summary.csv'}\r\n> \r\n> dataset = load_dataset('csv', data_files=file_dict, script_version='master', delimiter='\\t', column_names=['text', 'ctext'], features=features)\r\n> ```\r\n\r\nSince #644 it should now work on windows @ScottishFold007 \r\n\r\n> Trying the following snippet, I get different problems on Linux and Windows.\r\n> \r\n> ```python\r\n> dataset = load_dataset(\"text\", data_files=\"data.txt\")\r\n> # or \r\n> dataset = load_dataset(\"text\", data_files=[\"data.txt\"])\r\n> ```\r\n>\r\n> Windows just seems to get stuck. Even with a tiny dataset of 10 lines, it has been stuck for 15 minutes already at this message:\r\n> \r\n> ```\r\n> Checking C:\\Users\\bramv\\.cache\\huggingface\\datasets\\b1d50a0e74da9a7b9822cea8ff4e4f217dd892e09eb14f6274a2169e5436e2ea.30c25842cda32b0540d88b7195147decf9671ee442f4bc2fb6ad74016852978e.py for additional imports.\r\n> Found main folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\\Users\\bramv\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\text\r\n> Found specific version folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\\Users\\bramv\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\text\\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7\r\n> Found script file from https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py to C:\\Users\\bramv\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\text\\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7\\text.py\r\n> Couldn't find dataset infos file at https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text\\dataset_infos.json\r\n> Found metadata file for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\\Users\\bramv\\.cache\\huggingface\\modules\\datasets_modules\\datasets\\text\\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7\\text.json\r\n> Using custom data configuration default\r\n> ```\r\n\r\nSame for you @BramVanroy .\r\n\r\nNot sure about the one on linux though",
"> To complete what @lhoestq is saying, I think that to use the new version of the `text` processing script (which is on master right now) you need to either specify the version of the script to be the `master` one or to install the lib from source (in which case it uses the `master` version of the script by default):\r\n> \r\n> ```python\r\n> dataset = load_dataset('text', script_version='master', data_files=XXX)\r\n> ```\r\n> \r\n> We do versioning by default, i.e. your version of the dataset lib will use the script with the same version by default (i.e. only the `1.0.1` version of the script if you have the PyPI version `1.0.1` of the lib).\r\n\r\nLinux here:\r\n\r\nI was using the 0.4.0 nlp library load_dataset to load a text dataset of 9-10Gb without collapsing the RAM memory. However, today I got the csv error message mentioned in this issue. After installing the new (datasets) library from source and specifying the script_verson = 'master' I'm still having this same error message. Furthermore, I cannot use the dictionary \"trick\" to load the dataset since the system kills the process due to a RAM out of memory problem. Is there any other solution to this error? Thank you in advance. ",
"Hi @raruidol \r\nTo fix the RAM issue you'll need to shard your text files into smaller files (see https://github.com/huggingface/datasets/issues/610#issuecomment-691672919 for example)\r\n\r\nI'm not sure why you're having the csv error on linux.\r\nDo you think you could to to reproduce it on google colab for example ?\r\nOr send me a dummy .txt file that reproduces the issue ?",
"@lhoestq \r\n\r\nThe crash message shows up when loading the dataset:\r\n```\r\nprint('Loading corpus...') \r\nfiles = glob.glob('corpora/shards/*') \r\n-> dataset = load_dataset('text', script_version='master', data_files=files) \r\nprint('Corpus loaded.')\r\n```\r\nAnd this is the exact message:\r\n```\r\nTraceback (most recent call last):\r\n File \"run_language_modeling.py\", line 27, in <module>\r\n dataset = load_dataset('text', script_version='master', data_files=files)\r\n File \"/home/jupyter-raruidol/DebatAnalyser/env/lib/python3.7/site-packages/datasets/load.py\", line 611, in load_dataset\r\n ignore_verifications=ignore_verifications,\r\n File \"/home/jupyter-raruidol/DebatAnalyser/env/lib/python3.7/site-packages/datasets/builder.py\", line 471, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/home/jupyter-raruidol/DebatAnalyser/env/lib/python3.7/site-packages/datasets/builder.py\", line 548, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/home/jupyter-raruidol/DebatAnalyser/env/lib/python3.7/site-packages/datasets/builder.py\", line 892, in _prepare_split\r\n for key, table in utils.tqdm(generator, unit=\" tables\", leave=False, disable=not_verbose):\r\n File \"/home/jupyter-raruidol/DebatAnalyser/env/lib/python3.7/site-packages/tqdm/std.py\", line 1130, in __iter__\r\n for obj in iterable:\r\n File \"/home/jupyter-raruidol/.cache/huggingface/modules/datasets_modules/datasets/text/512f465342e4f4cd07a8791428a629c043bb89d55ad7817cbf7fcc649178b014/text.py\", line 107, in _generate_tables\r\n convert_options=self.config.convert_options,\r\n File \"pyarrow/_csv.pyx\", line 714, in pyarrow._csv.read_csv\r\n File \"pyarrow/error.pxi\", line 122, in pyarrow.lib.pyarrow_internal_check_status\r\n File \"pyarrow/error.pxi\", line 84, in pyarrow.lib.check_status\r\npyarrow.lib.ArrowInvalid: CSV parse error: Expected 1 columns, got 2\r\n```\r\n\r\nAnd these are the pip packages I have atm and their versions:\r\n\r\n```\r\nPackage Version Location \r\n--------------- --------- -------------------------------------------------------------\r\ncertifi 2020.6.20 \r\nchardet 3.0.4 \r\nclick 7.1.2 \r\ndatasets 1.0.2 \r\ndill 0.3.2 \r\nfilelock 3.0.12 \r\nfuture 0.18.2 \r\nidna 2.10 \r\njoblib 0.16.0 \r\nnumpy 1.19.1 \r\npackaging 20.4 \r\npandas 1.1.1 \r\npip 19.0.3 \r\npyarrow 1.0.1 \r\npyparsing 2.4.7 \r\npython-dateutil 2.8.1 \r\npytz 2020.1 \r\nregex 2020.7.14 \r\nrequests 2.24.0 \r\nsacremoses 0.0.43 \r\nsentencepiece 0.1.91 \r\nsetuptools 40.8.0 \r\nsix 1.15.0 \r\ntokenizers 0.8.1rc2 \r\ntorch 1.6.0 \r\ntqdm 4.48.2 \r\ntransformers 3.0.2 /home/jupyter-raruidol/DebatAnalyser/env/src/transformers/src\r\n```\r\n\r\n\r\n",
"I tested on google colab which is also linux using this code:\r\n\r\n- first download an arbitrary text file\r\n```bash\r\nwget https://raw.githubusercontent.com/abisee/cnn-dailymail/master/url_lists/all_train.txt\r\n```\r\n- then run\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nd = load_dataset(\"text\", data_files=\"all_train.txt\", script_version='master')\r\n```\r\nAnd I don't get this issue.\r\n\r\n\\> Could you test on your side if these lines work @raruidol ?\r\n\r\nalso cc @Skyy93 as it seems you have the same issue\r\n\r\nIf it works:\r\nIt could mean that the issue could come from unexpected patterns in the files you want to use.\r\nIn that case we should find a way to handle them.\r\n\r\nAnd if it doesn't work:\r\nIt could mean that it comes from the way pyarrow reads text files on linux.\r\nIn that case we should report it to pyarrow and find a workaround in the meantime\r\n\r\nEither way it should help to find where this bug comes from and fix it :)\r\n\r\nThank you in advance !",
"Update: also tested the above code in a docker container from [jupyter/minimal-notebook](https://hub.docker.com/r/jupyter/minimal-notebook/) (based on ubuntu) and still not able to reproduce",
"It looks like with your text input file works without any problem. I have been doing some experiments this morning with my input files and I'm almost certain that the crash is caused by some unexpected pattern in the files. However, I've not been able to spot the main cause of it. What I find strange is that this same corpus was being loaded by the nlp 0.4.0 library without any problem... Where can I find the code where you structure the input text data in order to use it with pyarrow?",
"Under the hood it does\r\n```python\r\nimport pyarrow as pa\r\nimport pyarrow.csv\r\n\r\n# Use csv reader from Pyarrow with one column for text files\r\n\r\n# To force the one-column setting, we set an arbitrary character\r\n# that is not in text files as delimiter, such as \\b or \\v.\r\n# The bell character, \\b, was used to make beeps back in the days\r\nparse_options = pa.csv.ParseOptions( \r\n delimiter=\"\\b\", \r\n quote_char=False, \r\n double_quote=False, \r\n escape_char=False, \r\n newlines_in_values=False, \r\n ignore_empty_lines=False, \r\n)\r\n\r\nread_options= pa.csv.ReadOptions(use_threads=True, column_names=[\"text\"])\r\n\r\npa_table = pa.csv.read_csv(\"all_train.txt\", read_options=read_options, parse_options=parse_options)\r\n```\r\n\r\nNote that we changed the parse options with datasets 1.0\r\nIn particular the delimiter used to be `\\r` but this delimiter doesn't work on windows.",
"Could you try with `\\a` instead of `\\b` ? It looks like the bell character is \\a in python and not \\b",
"I was just exploring if the crash was happening in every shard or not, and which shards were generating the error message. With \\b I got the following list of shards crashing:\r\n\r\n```\r\nErrors on files: ['corpora/shards/shard_0069', 'corpora/shards/shard_0043', 'corpora/shards/shard_0014', 'corpora/shards/shard_0032', 'corpora/shards/shard_0088', 'corpora/shards/shard_0018', 'corpora/shards/shard_0073', 'corpora/shards/shard_0079', 'corpora/shards/shard_0038', 'corpora/shards/shard_0041', 'corpora/shards/shard_0007', 'corpora/shards/shard_0004', 'corpora/shards/shard_0102', 'corpora/shards/shard_0096', 'corpora/shards/shard_0030', 'corpora/shards/shard_0076', 'corpora/shards/shard_0067', 'corpora/shards/shard_0052', 'corpora/shards/shard_0026', 'corpora/shards/shard_0024', 'corpora/shards/shard_0064', 'corpora/shards/shard_0044', 'corpora/shards/shard_0013', 'corpora/shards/shard_0062', 'corpora/shards/shard_0057', 'corpora/shards/shard_0097', 'corpora/shards/shard_0094', 'corpora/shards/shard_0078', 'corpora/shards/shard_0075', 'corpora/shards/shard_0039', 'corpora/shards/shard_0077', 'corpora/shards/shard_0021', 'corpora/shards/shard_0040', 'corpora/shards/shard_0009', 'corpora/shards/shard_0023', 'corpora/shards/shard_0095', 'corpora/shards/shard_0107', 'corpora/shards/shard_0063', 'corpora/shards/shard_0086', 'corpora/shards/shard_0047', 'corpora/shards/shard_0089', 'corpora/shards/shard_0037', 'corpora/shards/shard_0101', 'corpora/shards/shard_0093', 'corpora/shards/shard_0082', 'corpora/shards/shard_0091', 'corpora/shards/shard_0065', 'corpora/shards/shard_0020', 'corpora/shards/shard_0070', 'corpora/shards/shard_0008', 'corpora/shards/shard_0058', 'corpora/shards/shard_0060', 'corpora/shards/shard_0022', 'corpora/shards/shard_0059', 'corpora/shards/shard_0100', 'corpora/shards/shard_0027', 'corpora/shards/shard_0072', 'corpora/shards/shard_0098', 'corpora/shards/shard_0019', 'corpora/shards/shard_0066', 'corpora/shards/shard_0042', 'corpora/shards/shard_0053']\r\n```\r\n\r\nI also tried with \\a and the list decreased but there were still several crashes:\r\n\r\n```\r\nErrors on files: ['corpora/shards/shard_0069', 'corpora/shards/shard_0055', 'corpora/shards/shard_0043', 'corpora/shards/shard_0014', 'corpora/shards/shard_0073', 'corpora/shards/shard_0025', 'corpora/shards/shard_0068', 'corpora/shards/shard_0102', 'corpora/shards/shard_0096', 'corpora/shards/shard_0076', 'corpora/shards/shard_0067', 'corpora/shards/shard_0026', 'corpora/shards/shard_0024', 'corpora/shards/shard_0044', 'corpora/shards/shard_0087', 'corpora/shards/shard_0092', 'corpora/shards/shard_0074', 'corpora/shards/shard_0094', 'corpora/shards/shard_0078', 'corpora/shards/shard_0039', 'corpora/shards/shard_0077', 'corpora/shards/shard_0040', 'corpora/shards/shard_0009', 'corpora/shards/shard_0107', 'corpora/shards/shard_0063', 'corpora/shards/shard_0103', 'corpora/shards/shard_0047', 'corpora/shards/shard_0033', 'corpora/shards/shard_0089', 'corpora/shards/shard_0037', 'corpora/shards/shard_0082', 'corpora/shards/shard_0071', 'corpora/shards/shard_0091', 'corpora/shards/shard_0065', 'corpora/shards/shard_0070', 'corpora/shards/shard_0058', 'corpora/shards/shard_0081', 'corpora/shards/shard_0060', 'corpora/shards/shard_0002', 'corpora/shards/shard_0059', 'corpora/shards/shard_0027', 'corpora/shards/shard_0072', 'corpora/shards/shard_0098', 'corpora/shards/shard_0019', 'corpora/shards/shard_0045', 'corpora/shards/shard_0036', 'corpora/shards/shard_0066', 'corpora/shards/shard_0053']\r\n```\r\n\r\nWhich means that it is quite possible that the assumption of that some unexpected pattern in the files is causing the crashes is true. If I am able to reach any conclusion I will post It here asap.",
"Hmmm I was expecting it to work with \\a, not sure why they appear in your text files though",
"Hi @lhoestq, is there any input length restriction which was not before the update of the nlp library?",
"No we never set any input length restriction on our side (maybe arrow but I don't think so)",
"@lhoestq Can you ever be certain that a delimiter character is not present in a plain text file? In other formats (e.g. CSV) , rules are set of what is allowed and what isn't so that it actually constitutes a CSV file. In a text file you basically have \"anything goes\", so I don't think you can ever be entirely sure that the chosen delimiter does not exist in the text file, or am I wrong? \r\n\r\nIf I understand correctly you choose a delimiter that we hope does not exist in the file, so that when the CSV parser starts splitting into columns, it will only ever create one column? Why can't we use a newline character though?",
"Okay, I have splitted the crashing shards into individual sentences and some examples of the inputs that are causing the crashes are the following ones:\r\n\r\n\r\n_4.βDE LβORGANITZACIΓ ESTAMENTAL A LβORGANITZACIΓ EN CLASSES A mesura que es desenvolupava un sistema econΓ²mic capitalista i naixia una classe burgesa cada vegada mΓ©s preparada per a substituir els dirigents de les velles monarquies absolutistes, es qΓΌestionava lβabundΓ ncia de bΓ©ns amortitzats, que com sβha dit estaven fora del mercat i no pagaven tributs, pels perjudicis que ocasionaven a les finances pΓΊbliques i a lβeconomia en general. Aquest estat dβopiniΓ³ revolucionari va desembocar en un conjunt de mesures prΓ ctiques de carΓ cter liberal. Dβuna banda, les que intentaven desposseir les mans mortes del domini de bΓ©ns acumulats, procΓ©s que acostumem a denominar desamortitzaciΓ³, i que no Γ©s mΓ©s que la nacionalitzaciΓ³ i venda dβaquests bΓ©ns eclesiΓ stics o civils en subhasta pΓΊblica al millor postor. Dβaltra banda, les que redimien o reduΓ―en els censos i delmes o aixecaven les prohibicions de venda, Γ©s a dir, les vinculacions. La desamortitzaciΓ³, que va afectar bΓ©ns dels ordes religiosos, dels pobles i dβalgunes corporacions civils, no va ser un camΓ fΓ cil, perquΓ¨ costava i costa trobar algΓΊ que sigui indiferent a la pΓ¨rdua de bΓ©ns, drets i privilegis. I tΓ© una gran transcendΓ¨ncia, va privar els antics estaments de les Espanyes, clero i pobles βla noblesa en queda al margeβ, de la forΓ§a econΓ²mica que els donaven bona part de les seves terres i, en ΓΊltima instΓ ncia, va preparar el terreny per a la substituciΓ³ de la vella societat estamental per la nova societat classista. En aquesta societat, en teoria, les agrupacions socials sΓ³n obertes, no tenen cap estatut jurΓdic privilegiat i estan definides per la possessiΓ³ o no dβuns bΓ©ns econΓ²mics que sΓ³n lliurement alienables. A les Espanyes la transformaciΓ³ va afectar poc lβaristocrΓ cia latifundista, allΓ on nβhi havia. Aquesta situaciΓ³ va afavorir, en part, la persistΓ¨ncia de la vella cultura de la societat estamental en determinats ambients, i aixΓ² ha influΓ―t decisivament en la manca de democrΓ cia que caracteritza la majoria de rΓ¨gims polΓtics que sβhan anat succeint. Una manera de pensar que sempre sura en un moment o altre, i que de fet no acaba de desaparΓ¨ixer del tot. 5.βINICI DE LA DESAMORTITZACIΓ A LES ESPANYES Durant el segle xviii, dins dβaquesta visiΓ³ lliberal, va agafar forΓ§a en alguns cercles de les Espanyes el corrent dβopiniΓ³ contrari a les mans mortes. Durant el regnat de Carles III, sβarbitraren les primeres mesures desamortitzadores proposades per alguns ministres ilΒ·lustrats. Aquestes disposicions foren modestes i poc eficaces, no van aturar lβacumulaciΓ³ de terres per part dels estaments que constituΓ―en les mans mortes i varen afectar principalment bΓ©ns dels pobles. LβEsglΓ©sia no va ser tocada, excepte en el cas de 110_\r\n\r\n_la revoluciΓ³ liberal, perquΓ¨, encara que havia perdut els seus drets jurisdiccionals, havia conservat la majoria de terres i fins i tot les havia incrementat amb dβaltres que procedien de la desamortitzaciΓ³. En la nova situaciΓ³, les mans mortes del bosc pΓΊblic eren lβEstat, que no cerca mai lβautofinanΓ§ament de les despeses de gestiΓ³; els diners que manquin ja els posarΓ lβEstat. 9.βDEFENSA I INTENTS DE RECUPERACIΓ DELS BΓNS COMUNALS DESAMORTITZATS El procΓ©s de centralitzaciΓ³ no era senzill, perquΓ¨, dβuna banda, la nova organitzaciΓ³ apartava de la gestiΓ³ moltes corporacions locals i molts veΓ―ns que lβhavien portada des de lβedat mitjana, i, de lβaltra, era difΓcil de coordinar la nova silvicultura amb moltes prΓ ctiques forestals i drets tradicionals, com la pastura, fer llenya o tallar un arbre aquΓ i un altre allΓ quan tenia el gruix suficient, les prΓ ctiques que sβhavien fet sempre. Les primeres passes de la nova organitzaciΓ³ centralitzada varen tenir moltes dificultats en aquells indrets en quΓ¨ els terrenys municipals i comunals tenien un paper important en lβeconomia local. La desobediΓ¨ncia a determinades normes imposades varen prendre formes diferents. Algunes institucions, com, per exemple, la DiputaciΓ³ de Lleida, varen retardar la tramitaciΓ³ dβalguns expedients i varen evitar la venda de bΓ©ns municipals. Molts pobles permeteren deixar que els veΓ―ns continuessin amb les seves prΓ ctiques tradicionals, dβaltres varen boicotejar les subhastes dβaprofitaments. LβEstat va reaccionar encomanant a la GuΓ rdia Civil el compliment de les noves directrius. Imposar el nou rΓ¨gim va costar a lβAdministraciΓ³ un grapat dβanys, perΓ² de mica en mica, amb molta, molta guarderia i gens de negociaciΓ³, ho va aconseguir. La nova gestiΓ³ estatal dels bΓ©ns municipals va deixar, com hem comentat, molta gent sense uns recursos necessaris per a la supervivΓ¨ncia, sobre tot en Γ rees on predominaven les grans propietats, i on els pagesos sense terra treballaven de jornalers temporers. AixΓ² va afavorir que, a bona part de les Espanyes, les primeres lluites camperoles de la segona meitat del segle xix defensessin la recuperaciΓ³ dels comunals desamortitzats; per a molts aquella expropiaciΓ³ i venda dirigida pels governs monΓ rquics era la causa de molta misΓ¨ria. Dβaltres, mΓ©s radicalitzats, varen entendre que lβeliminaciΓ³ de la propietat colΒ·lectiva i la gestiΓ³ estatal dels boscos no desamortitzats suposava una usurpaciΓ³ pura i dura. En les zones mΓ©s afectades per la desamortitzaciΓ³ aixΓ² va donar lloc a un imaginari centrat en la defensa del comunal. La Segona RepΓΊblica va arribar en una conjuntura econΓ²mica de crisi, generada pel crac del 1929. Al camp, aquesta situaciΓ³ va produir una forta caiguda dels preus dels productes agraris i un increment important de lβatur. QUADERNS AGRARIS 42β(juny 2017), p. 105-126_\r\n\r\nI think that the main difference between the crashing samples and the rest is their length. Therefore, couldn't the length be causing the message errors? I hope with these samples you can identify what is causing the crashes considering that the 0.4.0 nlp library was loading them properly.",
"So we're using the csv reader to read text files because arrow doesn't have a text reader.\r\nTo workaround the fact that text files are just csv with one column, we want to set a delimiter that doesn't appear in text files.\r\nUntil now I thought that it would do the job but unfortunately it looks like even characters like \\a appear in text files.\r\n\r\nSo we have to option:\r\n- find another delimiter that does the job (maybe `\\x1b` esc or `\\x18` cancel)\r\n- don't use the csv reader from arrow but the text reader from pandas instead (or any other reader). The only important thing is that it must be fast (arrow's reader has a nice and fast multithreaded for csv that we're using now but hopefully we can find an alternative)\r\n\r\n\r\n\r\n> @lhoestq Can you ever be certain that a delimiter character is not present in a plain text file? In other formats (e.g. CSV) , rules are set of what is allowed and what isn't so that it actually constitutes a CSV file. In a text file you basically have \"anything goes\", so I don't think you can ever be entirely sure that the chosen delimiter does not exist in the text file, or am I wrong?\r\n\r\nAs long as the text file follows some encoding it wouldn't make sense to have characters such as the bell character. However I agree it can happen.\r\n\r\n> If I understand correctly you choose a delimiter that we hope does not exist in the file, so that when the CSV parser starts splitting into columns, it will only ever create one column? Why can't we use a newline character though?\r\n\r\nExactly. Arrow doesn't allow the newline character unfortunately.",
"> Okay, I have splitted the crashing shards into individual sentences and some examples of the inputs that are causing the crashes are the following ones\r\n\r\nThanks for digging into it !\r\n\r\nCharacters like \\a or \\b are not shown when printing the text, so as it is I can't tell if it contains unexpected characters.\r\nMaybe could could open the file in python and check if `\"\\b\" in open(\"path/to/file\", \"r\").read()` ?\r\n\r\n> I think that the main difference between the crashing samples and the rest is their length. Therefore, couldn't the length be causing the message errors? I hope with these samples you can identify what is causing the crashes considering that the 0.4.0 nlp library was loading them properly.\r\n\r\nTo check that you could try to run \r\n\r\n```python\r\nimport pyarrow as pa\r\nimport pyarrow.csv\r\n\r\nopen(\"dummy.txt\", \"w\").write(((\"a\" * 10_000) + \"\\n\") * 4) # 4 lines of 10 000 'a'\r\n\r\nparse_options = pa.csv.ParseOptions( \r\n delimiter=\"\\b\", \r\n quote_char=False, \r\n double_quote=False, \r\n escape_char=False, \r\n newlines_in_values=False, \r\n ignore_empty_lines=False, \r\n)\r\n\r\nread_options= pa.csv.ReadOptions(use_threads=True, column_names=[\"text\"])\r\n\r\npa_table = pa.csv.read_csv(\"dummy.txt\", read_options=read_options, parse_options=parse_options)\r\n```\r\n\r\non my side it runs without error though",
"That's true, It was my error printing the text that way. Maybe as a workaround, I can force all my input samples to have \"\\b\" at the end?",
"> That's true, It was my error printing the text that way. Maybe as a workaround, I can force all my input samples to have \"\\b\" at the end?\r\n\r\nI don't think it would work since we only want one column, and \"\\b\" is set to be the delimiter between two columns, so it will raise the same issue again. Pyarrow would think that there is more than one column if the delimiter is found somewhere.\r\n\r\nAnyway, I I'll work on a new text reader if we don't find the right workaround about this delimiter issue."
] | Trying the following snippet, I get different problems on Linux and Windows.
```python
dataset = load_dataset("text", data_files="data.txt")
# or
dataset = load_dataset("text", data_files=["data.txt"])
```
(ps [This example](https://huggingface.co/docs/datasets/loading_datasets.html#json-files) shows that you can use a string as input for data_files, but the signature is `Union[Dict, List]`.)
The problem on Linux is that the script crashes with a CSV error (even though it isn't a CSV file). On Windows the script just seems to freeze or get stuck after loading the config file.
Linux stack trace:
```
PyTorch version 1.6.0+cu101 available.
Checking /home/bram/.cache/huggingface/datasets/b1d50a0e74da9a7b9822cea8ff4e4f217dd892e09eb14f6274a2169e5436e2ea.30c25842cda32b0540d88b7195147decf9671ee442f4bc2fb6ad74016852978e.py for additional imports.
Found main folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at /home/bram/.cache/huggingface/modules/datasets_modules/datasets/text
Found specific version folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at /home/bram/.cache/huggingface/modules/datasets_modules/datasets/text/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7
Found script file from https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py to /home/bram/.cache/huggingface/modules/datasets_modules/datasets/text/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7/text.py
Couldn't find dataset infos file at https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/dataset_infos.json
Found metadata file for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at /home/bram/.cache/huggingface/modules/datasets_modules/datasets/text/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7/text.json
Using custom data configuration default
Generating dataset text (/home/bram/.cache/huggingface/datasets/text/default-0907112cc6cd2a38/0.0.0/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7)
Downloading and preparing dataset text/default-0907112cc6cd2a38 (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/bram/.cache/huggingface/datasets/text/default-0907112cc6cd2a38/0.0.0/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7...
Dataset not on Hf google storage. Downloading and preparing it from source
Downloading took 0.0 min
Checksum Computation took 0.0 min
Unable to verify checksums.
Generating split train
Traceback (most recent call last):
File "/home/bram/Python/projects/dutch-simplification/utils.py", line 45, in prepare_data
dataset = load_dataset("text", data_files=dataset_f)
File "/home/bram/.local/share/virtualenvs/dutch-simplification-NcpPZtDF/lib/python3.8/site-packages/datasets/load.py", line 608, in load_dataset
builder_instance.download_and_prepare(
File "/home/bram/.local/share/virtualenvs/dutch-simplification-NcpPZtDF/lib/python3.8/site-packages/datasets/builder.py", line 468, in download_and_prepare
self._download_and_prepare(
File "/home/bram/.local/share/virtualenvs/dutch-simplification-NcpPZtDF/lib/python3.8/site-packages/datasets/builder.py", line 546, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/bram/.local/share/virtualenvs/dutch-simplification-NcpPZtDF/lib/python3.8/site-packages/datasets/builder.py", line 888, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False, disable=not_verbose):
File "/home/bram/.local/share/virtualenvs/dutch-simplification-NcpPZtDF/lib/python3.8/site-packages/tqdm/std.py", line 1130, in __iter__
for obj in iterable:
File "/home/bram/.cache/huggingface/modules/datasets_modules/datasets/text/7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7/text.py", line 100, in _generate_tables
pa_table = pac.read_csv(
File "pyarrow/_csv.pyx", line 714, in pyarrow._csv.read_csv
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: CSV parse error: Expected 1 columns, got 2
```
Windows just seems to get stuck. Even with a tiny dataset of 10 lines, it has been stuck for 15 minutes already at this message:
```
Checking C:\Users\bramv\.cache\huggingface\datasets\b1d50a0e74da9a7b9822cea8ff4e4f217dd892e09eb14f6274a2169e5436e2ea.30c25842cda32b0540d88b7195147decf9671ee442f4bc2fb6ad74016852978e.py for additional imports.
Found main folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\Users\bramv\.cache\huggingface\modules\datasets_modules\datasets\text
Found specific version folder for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\Users\bramv\.cache\huggingface\modules\datasets_modules\datasets\text\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7
Found script file from https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py to C:\Users\bramv\.cache\huggingface\modules\datasets_modules\datasets\text\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7\text.py
Couldn't find dataset infos file at https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text\dataset_infos.json
Found metadata file for dataset https://raw.githubusercontent.com/huggingface/datasets/1.0.1/datasets/text/text.py at C:\Users\bramv\.cache\huggingface\modules\datasets_modules\datasets\text\7e13bc0fa76783d4ef197f079dc8acfe54c3efda980f2c9adfab046ede2f0ff7\text.json
Using custom data configuration default
```
| 622 |
https://github.com/huggingface/datasets/issues/620 | map/filter multiprocessing raises errors and corrupts datasets | [
"It seems that I ran into the same problem\r\n```\r\ndef tokenize(cols, example):\r\n for in_col, out_col in cols.items():\r\n example[out_col] = hf_tokenizer.convert_tokens_to_ids(hf_tokenizer.tokenize(example[in_col]))\r\n return example\r\ncola = datasets.load_dataset('glue', 'cola')\r\ntokenized_cola = cola.map(partial(tokenize, {'sentence': 'text_idxs'}),\r\n num_proc=2,)\r\n```\r\nand it outpus (exceprts)\r\n```\r\nConcatenating 2 shards from multiprocessing\r\nSet __getitem__(key) output type to python objects for ['idx', 'label', 'sentence', 'text_idxs'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nTesting the mapped function outputs\r\nTesting finished, running the mapping function on the dataset\r\nDone writing 532 indices in 4256 bytes .\r\nDone writing 531 indices in 4248 bytes .\r\nProcess #0 will write at /home/yisiang/.cache/huggingface/datasets/glue/cola/1.0.0/930e9d141872db65102cabb9fa8ac01c11ffc8a1b72c2e364d8cdda4610df542/tokenized_test_00000_of_00002.arrow\r\nProcess #1 will write at /home/yisiang/.cache/huggingface/datasets/glue/cola/1.0.0/930e9d141872db65102cabb9fa8ac01c11ffc8a1b72c2e364d8cdda4610df542/tokenized_test_00001_of_00002.arrow\r\nSpawning 2 processes\r\n```\r\nand then the program never stop.",
"same problem.\r\n`encoded_dataset = core_data.map(lambda examples: tokenizer(examples[\"query\"], examples[\"document\"], padding=True, truncation='longest_first', return_tensors=\"pt\", max_length=384), num_proc=16, keep_in_memory=True)`\r\nit outputs:\r\n```\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787500 indices in 25568400000 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nDone writing 1787499 indices in 25568385696 bytes .\r\nSet __getitem__(key) output type to python objects for ['document', 'is_random', 'query'] columns (when key is int or slice) and don't output other (un-formatted) columns.\r\nSpawning 16 processes\r\n```",
"Thanks for reporting.\r\n\r\n\r\nWhich tokenizers are you using ? What platform are you on ? Can you tell me which version of datasets and pyarrow you're using ? @timothyjlaurent @richarddwang @HuangLianzhe \r\n\r\nAlso if you're able to reproduce the issue on google colab that would be very helpful.\r\n\r\nI tried to run your code @richarddwang with the bert tokenizer and I wasn't able to reproduce",
"Hi, Sorry that I forgot to see what my version was.\r\nBut after updating datasets to master (editable install), and latest pyarrow. \r\nIt works now ~",
"Sorry, I just noticed this.\r\nI'm running this on MACOS the version of datasets I'm was 1.0.0 but I've also tried it on 1.0.2. `pyarrow==1.0.1`, Python 3.6\r\n\r\nConsider this code:\r\n```python\r\n\r\n loader_path = str(Path(__file__).parent / \"prodigy_dataset_builder.py\")\r\n ds = load_dataset(\r\n loader_path, name=\"prodigy-ds\", data_files=list(file_paths), cache_dir=cache_dir\r\n )[\"train\"]\r\n valid_relations = set(vocabulary.relation_types.keys())\r\n\r\n ds = ds.filter(filter_good_rows, fn_kwargs=dict(valid_rel_labels=valid_relations))\r\n\r\n ds = ds.map(map_bpe_encodings, batched=True, fn_kwargs=dict(tokenizer=vocabulary.tokenizer), num_proc=10)\r\n\r\n # add all feature data\r\n ner_ds: Dataset = ds.map(\r\n add_bio_tags,\r\n fn_kwargs=dict(ner_label_map=vocabulary.ner_labels, tokenizer=vocabulary.tokenizer),\r\n )\r\n rel_ds: Dataset = ner_ds.map(\r\n relation_ds_factory,\r\n batched=True,\r\n writer_batch_size=100,\r\n fn_kwargs=dict(tokenizer=vocabulary.tokenizer, vocabulary=vocabulary),\r\n )\r\n```\r\nThe loader is essentially a jsonloader with some extra error handling. The data is a jsonlines format with text field and a list of span objects and relation objects. \r\n\r\nIn the `ner_ds` a field, `ner_labels` is added, this is used in the downstream `relation_ds_factory`. It all runs fine in a single process but I get a KeyError error if run with num_proc set\r\n:\r\n\r\n```\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/dataset.py\", line 348, in relation_ds_factory\r\n ner_labels = example[\"ner_labels\"]\r\nKeyError: 'ner_labels'\r\n``` \r\n\r\nThis is just one example of what goes wrong. I've started just saving the dataset as arrow at the end because it takes a long time to map/filter/shuffle and the caching isn't working (tracked it down to byte differences in the pickled functions). \r\n\r\n^^ Interestingly if I heed the warning from Tokenizers and set the environment variable, `TOKENIZERS_PARALLELISM=true` the map just hangs:\r\n\r\n```\r\n[I 200921 21:43:18 filelock:318] Lock 5694118768 released on /Users/timothy.laurent/.cache/huggingface/datasets/_Users_timothy.laurent_.cache_huggingface_datasets_prodigy_dataset_builder_prodigy-ds-5f34378723c4e83f_0.0.0_e67d9b43d5cd82c50b1eae8f2097daf95b601a04dc03ddd504f2b234a5fa247a.lock\r\n100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 1.34ba/s]\r\n#0: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#1: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#2: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#3: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#4: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#5: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#6: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#7: 0%| | 0/1 [00:00<?, ?ba/s]\r\n#8: 0%| | 0/1 [00:00<?, ?ba/s]\r\n```",
"Thank you, I was able to reproduce :)\r\nI'm on it",
"#659 should fix the `KeyError` issue. It was due to the formatting not getting updated the right way",
"Also maybe @n1t0 knows why setting `TOKENIZERS_PARALLELISM=true` creates deadlock issues when calling `map` with multiprocessing ?",
"@lhoestq \r\n\r\nThanks for taking a look. I pulled the master but I still see the key error.\r\n\r\n```\r\nTo disable this warning, you can either:\r\n - Avoid using `tokenizers` before the fork if possible\r\n - Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\r\n#0: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 21.56ba/s]\r\n#1: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 17.71ba/s]\r\n#2: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 20.45ba/s]\r\n#3: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 26.05ba/s]\r\n#4: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 26.83ba/s]\r\n#5: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 27.00ba/s]\r\n#6: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 27.40ba/s]\r\n#7: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 25.91ba/s]\r\n#8: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 22.46ba/s]\r\n#9: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 20.15ba/s]\r\n#10: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 26.81ba/s]\r\n#11: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 27.45ba/s]\r\n100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 322/322 [00:00<00:00, 1462.85ex/s]\r\nTraceback (most recent call last): | 0/1 [00:00<?, ?ba/s]\r\n File \"text2struct/run_model.py\", line 372, in <module>\r\n main()\r\n File \"text2struct/run_model.py\", line 368, in main | 0/1 [00:00<?, ?ba/s]\r\n run_model(auto_envvar_prefix=\"GFB_CIES\") # pragma: no cover\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 829, in __call__\r\n return self.main(*args, **kwargs) | 0/1 [00:00<?, ?ba/s]\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 782, in main\r\n rv = self.invoke(ctx)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 1236, in invoke\r\n return Command.invoke(self, ctx)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 1066, in invoke\r\n return ctx.invoke(self.callback, **ctx.params)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 610, in invoke\r\n return callback(*args, **kwargs)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/decorators.py\", line 21, in new_func\r\n return f(get_current_context(), *args, **kwargs)\r\n File \"text2struct/run_model.py\", line 136, in run_model\r\n ctx.invoke(ctx.command.commands[config_dict[\"mode\"]])\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/core.py\", line 610, in invoke\r\n return callback(*args, **kwargs)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/click/decorators.py\", line 21, in new_func\r\n return f(get_current_context(), *args, **kwargs)\r\n File \"text2struct/run_model.py\", line 187, in train\r\n run_train_model(_parse_subcommand(ctx))\r\n File \"text2struct/run_model.py\", line 241, in run_train_model\r\n checkpoint_steps=config.train.checkpoint_steps,\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/train.py\", line 153, in alternate_training\r\n max_len=config.model.dim.max_len,\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/dataset.py\", line 466, in load_prodigy_tf_datasets\r\n folder, file_patterns, vocabulary, cache_dir=cache_dir, test_pct=test_pct\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/dataset.py\", line 447, in load_prodigy_arrow_datasets\r\n fn_kwargs=dict(tokenizer=vocabulary.tokenizer, vocabulary=vocabulary),\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1224, in map\r\n update_data = does_function_return_dict(test_inputs, test_indices)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1195, in does_function_return_dict\r\n function(*fn_args, indices, **fn_kwargs) if with_indices else function(*fn_args, **fn_kwargs)\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/dataset.py\", line 348, in relation_ds_factory\r\n ner_labels = example[\"ner_labels\"]\r\nKeyError: 'ner_labels'\r\n\r\n```",
"The parallelism is automatically disabled on `tokenizers` when the process gets forked, while we already used the parallelism capabilities of a tokenizer. We have to do it in order to avoid having the process hang, because we cannot safely fork a multithreaded process (cf https://github.com/huggingface/tokenizers/issues/187).\r\nSo if possible, the tokenizers shouldn't be used before the fork, so that each process can then make use of the parallelism. Otherwise using `TOKENIZERS_PARALLELISM=false` is the way to go.",
"> Thanks for taking a look. I pulled the master but I still see the key error.\r\n\r\nI am no longer able to get the error since #659 was merged. Not sure why you still have it @timothyjlaurent \r\nMaybe it is a cache issue ? Could you try to use `load_from_cache_file=False` in your `.map()` calls ?",
"> The parallelism is automatically disabled on `tokenizers` when the process gets forked, while we already used the parallelism capabilities of a tokenizer. We have to do it in order to avoid having the process hang, because we cannot safely fork a multithreaded process (cf [huggingface/tokenizers#187](https://github.com/huggingface/tokenizers/issues/187)).\r\n> So if possible, the tokenizers shouldn't be used before the fork, so that each process can then make use of the parallelism. Otherwise using `TOKENIZERS_PARALLELISM=false` is the way to go.\r\n\r\nOk thanks :)\r\n\r\nIs there something we should do on the `datasets` side to avoid that that the program hangs ?\r\n\r\nAlso when doing `.map` with a tokenizer, the tokenizer is called once on the first examples of the dataset to check the function output before spawning the processes. Is that compatible with how tokenizers are supposed to be used with multiprocessing ?",
"#659 fixes the empty dict issue\r\n#688 fixes the hang issue",
"Hmmm I pulled the latest commit, `b93c5517f70a480533a44e0c42638392fd53d90`, and I'm still seeing both the hanging and the key error. ",
"Hi @timothyjlaurent \r\n\r\nThe hanging fix just got merged, that why you still had it.\r\n\r\nFor the key error it's possible that the code you ran reused cached datasets from where the KeyError bug was still there.\r\nCould you try to clear your cache or make sure that it doesn't reuse cached data with `.map(..., load_from_cache=False)` ?\r\nLet me know if it it helps",
"Hi @lhoestq , \r\n\r\nThanks for letting me know about the update.\r\n\r\nSo I don't think it's the caching - because hashing mechanism isn't stable for me -- but that's a different issue. In any case I `rm -rf ~/.cache/huggingface` to make a clean slate.\r\n\r\nI synced with master and I see the key error has gone away, I tried with and without the `TOKENIZERS_PARALLELISM` variable set and see the log line for setting the value false before the map.\r\n\r\nNow I'm seeing an issue with `.train_test_split()` on datasets that are the product of a multiprocess map.\r\n\r\nHere is the stack trace\r\n\r\n```\r\n File \"/Users/timothy.laurent/src/inv-text2struct/text2struct/model/dataset.py\", line 451, in load_prodigy_arrow_datasets\r\n ner_ds_dict = ner_ds.train_test_split(test_size=test_pct, shuffle=True, seed=seed)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/src/datasets/src/datasets/arrow_dataset.py\", line 168, in wrapper\r\n dataset.set_format(**new_format)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/src/datasets/src/datasets/fingerprint.py\", line 163, in wrapper\r\n out = func(self, *args, **kwargs)\r\n File \"/Users/timothy.laurent/.virtualenvs/inv-text2struct/src/datasets/src/datasets/arrow_dataset.py\", line 794, in set_format\r\n list(filter(lambda col: col not in self._data.column_names, columns)), self._data.column_names\r\nValueError: Columns ['train', 'test'] not in the dataset. Current columns in the dataset: ['_input_hash', '_task_hash', '_view_id', 'answer', 'encoding__ids', 'encoding__offsets', 'encoding__overflowing', 'encoding__tokens', 'encoding__words', 'ner_ids', 'ner_labels', 'relations', 'spans', 'text', 'tokens']\r\n```\r\n\r\n\r\n",
"Thanks for reporting.\r\nI'm going to fix that and add a test case so that it doesn't happen again :) \r\nI'll let you know when it's done\r\n\r\nIn the meantime if you could make a google colab that reproduces the issue it would be helpful ! @timothyjlaurent ",
"Sure thing, @lhoestq.\r\n\r\nhttps://colab.research.google.com/drive/1lg4fbyrUO6m8ssQ2dNdVFaUqMUfA2zZ3?usp=sharing",
"Thanks @timothyjlaurent ! I just merged a fix on master. I also checked your notebook and it looks like it's working now.\r\nI added some tests to make sure it works as expected now :)",
"Great, @lhoestq . I'm trying to verify in the colab:\r\nchanged\r\n```\r\n!pip install datasets\r\n```\r\nto \r\n\r\n```\r\n!pip install git+https://github.com/huggingface/datasets@master\r\n```\r\n\r\nBut I'm still seeing the error - I wonder why?",
"It works on my side @timothyjlaurent on google colab.\r\nDid you try to uninstall datasets first, before updating it to master's version ?",
"I didn't -- it was a new sessions --- buuut - look like it's working today -- woot! I'll close this issue. Thanks @lhoestq "
] | After upgrading to the 1.0 started seeing errors in my data loading script after enabling multiprocessing.
```python
...
ner_ds_dict = ner_ds.train_test_split(test_size=test_pct, shuffle=True, seed=seed)
ner_ds_dict["validation"] = ner_ds_dict["test"]
rel_ds_dict = rel_ds.train_test_split(test_size=test_pct, shuffle=True, seed=seed)
rel_ds_dict["validation"] = rel_ds_dict["test"]
return ner_ds_dict, rel_ds_dict
```
The first train_test_split, `ner_ds`/`ner_ds_dict`, returns a `train` and `test` split that are iterable.
The second, `rel_ds`/`rel_ds_dict` in this case, returns a Dataset dict that has rows but if selected from or sliced into into returns an empty dictionary. eg `rel_ds_dict['train'][0] == {}` and `rel_ds_dict['train'][0:100] == {}`.
Ok I think I know the problem -- the rel_ds was mapped though a mapper with `num_proc=12`. If I remove `num_proc`. The dataset loads.
I also see errors with other map and filter functions when `num_proc` is set.
```
Done writing 67 indices in 536 bytes .
Done writing 67 indices in 536 bytes .
Fatal Python error: PyCOND_WAIT(gil_cond) failed
``` | 620 |
https://github.com/huggingface/datasets/issues/619 | Mistakes in MLQA features names | [
"Indeed you're right ! Thanks for reporting that\r\n\r\nCould you open a PR to fix the features names ?"
] | I think the following features in MLQA shouldn't be named the way they are:
1. `questions` (should be `question`)
2. `ids` (should be `id`)
3. `start` (should be `answer_start`)
The reasons I'm suggesting these features be renamed are:
* To make them consistent with other QA datasets like SQuAD, XQuAD, TyDiQA etc. and hence make it easier to concatenate multiple QA datasets.
* The features names are not the same as the ones provided in the original MLQA datasets (it uses the names I suggested).
I know these columns can be renamed using using `Dataset.rename_column_`, `questions` and `ids` can be easily renamed but `start` on the other hand is annoying to rename since it's nested inside the feature `answers`.
| 619 |
https://github.com/huggingface/datasets/issues/617 | Compare different Rouge implementations | [
"Updates - the differences between the following three\r\n(1) https://github.com/bheinzerling/pyrouge (previously popular. The one I trust the most)\r\n(2) https://github.com/google-research/google-research/tree/master/rouge\r\n(3) https://github.com/pltrdy/files2rouge (used in fairseq)\r\ncan be explained by two things, stemming and handling multiple sentences.\r\n\r\nStemming: \r\n(1), (2): default is no stemming. (3): default is with stemming ==> No stemming is the correct default as you did [here](https://github.com/huggingface/datasets/blob/master/metrics/rouge/rouge.py#L84)\r\n\r\nMultiple sentences:\r\n(1) `rougeL` splits text using `\\n`\r\n(2) `rougeL` ignores `\\n`. \r\n(2) `rougeLsum` splits text using `\\n`\r\n(3) `rougeL` splits text using `.`\r\n\r\nFor (2), `rougeL` and `rougeLsum` are identical if the sequence doesn't contain `\\n`. With `\\n`, it is `rougeLsum` that matches (1) not `rougeL`. \r\n\r\nOverall, and as far as I understand, for your implementation here https://github.com/huggingface/datasets/blob/master/metrics/rouge/rouge.py#L65 to match the default, you only need to change `rougeL` [here](https://github.com/huggingface/datasets/blob/master/metrics/rouge/rouge.py#L86) to `rougeLsum` to correctly compute metrics for text with newlines.\r\n\r\nTagging @sshleifer who might be interested.",
"Thanks for the clarification !\r\nWe're adding Rouge Lsum in #701 ",
"This is a real issue, sorry for missing the mention @ibeltagy\r\n\r\nWe implemented a more involved [solution](https://github.com/huggingface/transformers/blob/99cb924bfb6c4092bed9232bea3c242e27c6911f/examples/seq2seq/utils.py#L481) that enforces that sentences are split with `\\n` so that rougeLsum scores match papers even if models don't generate newlines. \r\n\r\nUnfortunately, the best/laziest way I found to do this introduced an `nltk` dependency (For sentence splitting, all sentences don't end in `.`!!!), but this might be avoidable with some effort.\r\n\r\n#### Sidebar: Wouldn't Deterministic Be Better?\r\n\r\n`rouge_scorer.scoring.BootstrapAggregator` is well named but is not deterministic which I would like to change for my mental health, unless there is some really good reason to sample 500 observations before computing f-scores.\r\n\r\nI have a fix on a branch, but I wanted to get some context before introducting a 4th way to compute rouge. Scores are generally within .03 Rouge2 of boostrap after multiplying by 100, e.g 22.05 vs 22.08 Rouge2.\r\n\r\n",
"> This is a real issue, sorry for missing the mention @ibeltagy\r\n> \r\n> We implemented a more involved [solution](https://github.com/huggingface/transformers/blob/99cb924bfb6c4092bed9232bea3c242e27c6911f/examples/seq2seq/utils.py#L481) that enforces that sentences are split with `\\n` so that rougeLsum scores match papers even if models don't generate newlines.\r\n> \r\n> Unfortunately, the best/laziest way I found to do this introduced an `nltk` dependency (For sentence splitting, all sentences don't end in `.`!!!), but this might be avoidable with some effort.\r\n\r\nThanks for the details, I didn't know about that. Maybe we should consider adding this processing step or at least mention it somewhere in the library or the documentation\r\n\r\n> #### Sidebar: Wouldn't Deterministic Be Better?\r\n> `rouge_scorer.scoring.BootstrapAggregator` is well named but is not deterministic which I would like to change for my mental health, unless there is some really good reason to sample 500 observations before computing f-scores.\r\n> \r\n> I have a fix on a branch, but I wanted to get some context before introducting a 4th way to compute rouge. Scores are generally within .03 Rouge2 of boostrap after multiplying by 100, e.g 22.05 vs 22.08 Rouge2.\r\n\r\nI think the default `n_samples` of the aggregator is 1000. We could increase it or at least allow users to change it if they want more precise results.",
"Hi, thanks for the solution. \r\n\r\nI am not sure if this is a bug, but on line [510](https://github.com/huggingface/transformers/blob/99cb924bfb6c4092bed9232bea3c242e27c6911f/examples/seq2seq/utils.py#L510), are pred, tgt supposed to be swapped?",
"This looks like a bug in an old version of the examples in `transformers`",
"Hi, so I took this example from the HF implementation. What I can see is that the precision of `Hello there` being summarized to `general kenobi` is 1. I don't understand how this calculation is correct.\r\nIs the comparison just counting the words?\r\nand if Yes, then how does this translates to summarization evaluation?\r\n```\r\n >>> rouge = datasets.load_metric('rouge')\r\n >>> predictions = [\"hello there\", \"general kenobi\"]\r\n >>> references = [\"hello there\", \"general kenobi\"]\r\n >>> results = rouge.compute(predictions=predictions, references=references)\r\n >>> print(list(results.keys()))\r\n ['rouge1', 'rouge2', 'rougeL', 'rougeLsum']\r\n >>> print(results[\"rouge1\"])\r\n AggregateScore(low=Score(precision=1.0, recall=1.0, fmeasure=1.0), mid=Score(precision=1.0, recall=1.0, fmeasure=1.0), high=Score(precision=1.0, recall=1.0, fmeasure=1.0))\r\n >>> print(results[\"rouge1\"].mid.fmeasure)\r\n 1.0\r\n\"\"\", stored examples: 0)\r\n```\r\n\r\n\r\n"
] | I used RougeL implementation provided in `datasets` [here](https://github.com/huggingface/datasets/blob/master/metrics/rouge/rouge.py) and it gives numbers that match those reported in the pegasus paper but very different from those reported in other papers, [this](https://arxiv.org/pdf/1909.03186.pdf) for example.
Can you make sure the google-research implementation you are using matches the official perl implementation?
There are a couple of python wrappers around the perl implementation, [this](https://pypi.org/project/pyrouge/) has been commonly used, and [this](https://github.com/pltrdy/files2rouge) is used in fairseq).
There's also a python reimplementation [here](https://github.com/pltrdy/rouge) but its RougeL numbers are way off.
| 617 |
https://github.com/huggingface/datasets/issues/616 | UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors | [
"I have the same issue",
"Same issue here when Trying to load a dataset from disk.",
"I am also experiencing this issue, and don't know if it's affecting my training.",
"Same here. I hope the dataset is not being modified in-place.",
"I think the only way to avoid this warning would be to do a copy of the numpy array before providing it.\r\n\r\nThis would slow down a bit the iteration over the dataset but maybe it would be safer. We could disable the copy with a flag on the `set_format` command.\r\n\r\nIn most typical cases of training a NLP model, PyTorch shouldn't modify the input so it's ok to have a non-writable array but I can understand the warning is a bit scary so maybe we could choose the side of non-warning/slower by default and have an option to speedup.\r\n\r\nWhat do you think @lhoestq ? ",
"@thomwolf Would it be possible to have the array look writeable, but raise an error if it is actually written to?\r\n\r\nI would like to keep my code free of warning, but I also wouldn't like to slow down the program because of unnecessary copy operations. ",
"@AndreasMadsen probably not I would guess (no free lunch hahah)",
"@thomwolf Why not? Writable is checked with `arr.flags.writeable`, and writing is done via magic methods.",
"Well because I don't know the internal of numpy as well as you I guess hahahah, do you want to try to open a PR proposing a solution?",
"@thomwolf @AndreasMadsen I think this is a terrible idea, n/o, and I am very much against it. Modifying internals of an array in such a hacky way is bound to run into other (user) issues down the line. To users it would not be clear at all what is going on e.g. when they check for write access (which will return True) but then they get a warning that the array is not writeable. That's extremely confusing. \r\n\r\nIf your only goal is to get rid of warnings in your code, then you can just use a [simplefilter](https://docs.python.org/3.8/library/warnings.html#temporarily-suppressing-warnings) for UserWarnings in your own code. Changing the code-base in such an intuitive way does not seem like a good way to go and sets a bad precedent, imo. \r\n\r\n(Feel free to disagree, of course.)\r\n\r\nIMO a warning can stay (as they can be filtered by users anyway), but it can be clarified why the warning takes place.",
"> To users it would not be clear at all what is going on e.g. when they check for write access (which will return True) but then they get a warning that the array is not writeable. That's extremely confusing.\r\n\r\nConfusion can be resolved with a helpful error message. In this case, that error message can be controlled by huggingface/datasets. The right argument here is that if code depends on `.flags.writable` being truthful (not just for warnings), then it will cause unavoidable errors. Although, I can't imagine such a use-case.\r\n\r\n> If your only goal is to get rid of warnings in your code, then you can just use a simplefilter for UserWarnings in your own code. Changing the code-base in such an intuitive way does not seem like a good way to go and sets a bad precedent, imo.\r\n\r\nI don't want to ignore all `UserWarnings`, nor all warnings regarding non-writable arrays. Ignoring warnings leads to hard to debug issues.\r\n\r\n> IMO a warning can stay (as they can be filtered by users anyway), but it can be clarified why the warning takes place.\r\n\r\nPlain use cases should really not generate warnings. It teaches developers to ignore warnings which is a terrible practice.\r\n\r\n---\r\n\r\nThe best solution would be to allow non-writable arrays in `DataLoader`, but that is a PyTorch issue.",
"> The right argument here is that if code depends on `.flags.writable` being truthful (not just for warnings), then it will cause unavoidable errors. Although, I can't imagine such a use-case.\r\n\r\nThat's exactly the argument in my first sentence. Too often someone \"cannot think of a use-case\", but you can not foresee the use-cases of a whole research community.\r\n \r\n> I don't want to ignore all `UserWarnings`, nor all warnings regarding non-writable arrays. Ignoring warnings leads to hard to debug issues.\r\n\r\nThat's fair.\r\n\r\n> Plain use cases should really not generate warnings. It teaches developers to ignore warnings which is a terrible practice.\r\n\r\nBut this is not a plain use-case (because Pytorch does not support these read-only tensors). Manually setting the flag to writable will solve the issue on the surface but is basically just a hack to compensate for something that is not allowed in another library. \r\n\r\nWhat about an \"ignore_warnings\" flag in `set_format` that when True wraps the offending code in a block to ignore userwarnings at that specific step in [_convert_outputs](https://github.com/huggingface/datasets/blob/880c2c76a8223a00c303eab2909371e857113063/src/datasets/arrow_dataset.py#L821)? Something like:\r\n\r\n```python\r\ndef _convert_outputs(..., ignore_warnings=True):\r\n ...\r\n with warnings.catch_warnings():\r\n if ignore_warnings:\r\n warnings.simplefilter(\"ignore\", UserWarning)\r\n return torch.tensor(...)\r\n# continues without warning filter after context manager...\r\n```",
"> But this is not a plain use-case (because Pytorch does not support these read-only tensors).\r\n\r\nBy \"plain\", I mean the recommended way to use `datasets` with PyTorch according to the `datasets` documentation.",
"This error is what I see when I run the first lines of the Pytorch Quickstart. It should also say that it should be ignored and/or how to fix it. BTW, this is a Pytorch error message -- not a Huggingface error message. My code runs anyway."
] | I am trying out the library and want to load in pickled data with `from_dict`. In that dict, one column `text` should be tokenized and the other (an embedding vector) should be retained. All other columns should be removed. When I eventually try to set the format for the columns with `set_format` I am getting this strange Userwarning without a stack trace:
> Set __getitem__(key) output type to torch for ['input_ids', 'sembedding'] columns (when key is int or slice) and don't output other (un-formatted) columns.
> C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\datasets\arrow_dataset.py:835: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at ..\torch\csrc\utils\tensor_numpy.cpp:141.)
> return torch.tensor(x, **format_kwargs)
The first one might not be related to the warning, but it is odd that it is shown, too. It is unclear whether that is something that I should do or something that that the program is doing at that moment.
Snippet:
```
dataset = Dataset.from_dict(torch.load("data/dummy.pt.pt"))
print(dataset)
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
keys_to_retain = {"input_ids", "sembedding"}
dataset = dataset.map(lambda example: tokenizer(example["text"], padding='max_length'), batched=True)
dataset.remove_columns_(set(dataset.column_names) - keys_to_retain)
dataset.set_format(type="torch", columns=["input_ids", "sembedding"])
dataloader = torch.utils.data.DataLoader(dataset, batch_size=2)
print(next(iter(dataloader)))
```
PS: the input type for `remove_columns_` should probably be an Iterable rather than just a List. | 616 |
https://github.com/huggingface/datasets/issues/615 | Offset overflow when slicing a big dataset with an array of indices in Pyarrow >= 1.0.0 | [
"Related: https://issues.apache.org/jira/browse/ARROW-9773\r\n\r\nIt's definitely a size thing. I took a smaller dataset with 87000 rows and did:\r\n```\r\nfor i in range(10,1000,20):\r\n table = pa.concat_tables([dset._data]*i)\r\n table.take([0])\r\n```\r\nand it broke at around i=300.\r\n\r\nAlso when `_indices` is not None, this breaks indexing by slice. E.g. `dset.shuffle()[:1]` breaks.\r\n\r\nLuckily so far I haven't seen `_indices.column(0).take` break, which means it doesn't break `select` or anything like that which is where the speed really matters, it's just `_getitem`. So I'm currently working around it by just doing the arrow v0 method in `_getitem`:\r\n```\r\n#if PYARROW_V0:\r\ndata_subset = pa.concat_tables(\r\n self._data.slice(indices_array[i].as_py(), 1) for i in range(len(indices_array))\r\n)\r\n#else:\r\n #data_subset = self._data.take(indices_array)\r\n```",
"Let me know if you meet other offset overflow issues @joeddav ",
"Will this problem be solved in newer version?",
"This specific issue has been fixed in https://github.com/huggingface/datasets/pull/645\r\n\r\nIf you still have this error, could you open a new issue and explain how to reproduce the error ?",
"same error here in version 2.1.0",
"Facing the same issue. \r\nSteps to reproduce: (dataset is a few GB big so try in colab maybe)\r\nDatasets version - 2.11.0\r\n```\r\nimport datasets\r\nimport re\r\n\r\nds = datasets.load_dataset('nishanthc/dnd_map_dataset_v0.1', split = 'train')\r\n\r\ndef get_text_caption(example):\r\n regex_pattern = r'\\s\\d+x\\d+|,\\sLQ|,\\sgrid|\\.\\w+$'\r\n example['text_caption'] = re.sub(regex_pattern, '', example['picture_text'])\r\n return example\r\n\r\nds = ds.map(get_text_caption)\r\n```\r\n\r\nI am trying to apply a regex to remove certain patterns from a text column. Not sure why this error is showing up.",
"Got this error on a very large data set (900m rows, 35 cols) performing a similar batch map operation.",
"There is a solution that has been proposed here: https://github.com/huggingface/datasets/issues/5783",
"@lhoestq I ran into this problem with load_dataset. What should I do\r\n",
"What version of `datasets` are you using ? Feel free to open a new issue with some details (e.g. what dataset you loaded, what code you ran etc)",
"@lhoestq It's been solved,thanks",
"I am facing this problem.\r\n\r\nHere's my code:\r\n\r\n```python\r\nmodel.eval()\r\nmodel.to('cuda')\r\nblock_size = tokenizer.model_max_length\r\ndef group_texts(examples):\r\n # Concatenate all texts.\r\n concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}\r\n total_length = len(concatenated_examples[list(examples.keys())[0]])\r\n # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can\r\n # customize this part to your needs.\r\n total_length = (total_length // block_size) * block_size\r\n # Split by chunks of max_len.\r\n result = {\r\n k: [t[i : i + block_size] for i in range(0, total_length, block_size)]\r\n for k, t in concatenated_examples.items()\r\n }\r\n with torch.no_grad():\r\n input_ids = torch.tensor(result[\"input_ids\"]).to('cuda')\r\n attention_mask = torch.tensor(result[\"attention_mask\"]).to('cuda')\r\n r = model.forward(input_ids=input_ids, attention_mask=attention_mask)\r\n result[\"labels\"] = r.logits.cpu().numpy().tolist()\r\n return result\r\n\r\n\r\nlm_datasets = tokenized_datasets.map(\r\n group_texts,\r\n batched=True,\r\n batch_size=1000,\r\n num_proc=1,\r\n)\r\n```\r\n\r\nThis works for a few iterations and then gives the error:\r\n```sh\r\nTraceback (most recent call last):\r\n File \"/home/jpiabrantes/rosetta/tmp.py\", line 45, in <module>\r\n lm_datasets = tokenized_datasets.map(\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/dataset_dict.py\", line 868, in map\r\n {\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/dataset_dict.py\", line 869, in <dictcomp>\r\n k: dataset.map(\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_dataset.py\", line 593, in wrapper\r\n out: Union[\"Dataset\", \"DatasetDict\"] = func(self, *args, **kwargs)\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_dataset.py\", line 558, in wrapper\r\n out: Union[\"Dataset\", \"DatasetDict\"] = func(self, *args, **kwargs)\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_dataset.py\", line 3105, in map\r\n for rank, done, content in Dataset._map_single(**dataset_kwargs):\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_dataset.py\", line 3501, in _map_single\r\n writer.write_batch(batch)\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py\", line 571, in write_batch\r\n self.write_table(pa_table, writer_batch_size)\r\n File \"/home/jpiabrantes/rosetta/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py\", line 583, in write_table\r\n pa_table = pa_table.combine_chunks()\r\n File \"pyarrow/table.pxi\", line 3638, in pyarrow.lib.Table.combine_chunks\r\n File \"pyarrow/error.pxi\", line 154, in pyarrow.lib.pyarrow_internal_check_status\r\n File \"pyarrow/error.pxi\", line 91, in pyarrow.lib.check_status\r\npyarrow.lib.ArrowInvalid: offset overflow while concatenating arrays\r\n```",
"Hi ! What version of `pyarrow` are you using ? Also what's the lengths of your texts ?",
"@lhoestq pyarrow version: 15.0.2\r\n\r\nlengths of texts are 1024 tokens.",
"```\r\nimport pandas as pd\r\nfrom datasets import Dataset,Image\r\n\r\n# Read the CSV file\r\ndf = pd.read_csv(\"MedMQ-2k/metadata.csv\")\r\n# Create a Hugging Face Dataset\r\ndataset = Dataset.from_pandas(df)\r\n\r\ndataset = dataset.map(lambda example: {\"image\": example[\"file_name\"]}, batched=True)\r\n# Convert the file_name column to Image type\r\ndataset = dataset.cast_column(\"image\", Image())\r\n\r\n# Upload to Hugging Face Hub (make sure authentication is set up)\r\ndataset.push_to_hub(\"MedMLLM-attack/3MAD-24K\", num_shards=16)\r\n```\r\n<img width=\"1143\" alt=\"ζͺε±2024-05-02 13 04 09\" src=\"https://github.com/huggingface/datasets/assets/48406770/a722eadc-f8f1-4094-a38b-eaeebfa11b83\">\r\n\r\n<img width=\"176\" alt=\"ζͺε±2024-05-02 13 03 40\" src=\"https://github.com/huggingface/datasets/assets/48406770/3f6fbeff-2a43-4d39-8ffa-fd21769608bc\">\r\n\r\nsame problem here\r\n\r\n\r\n- datasets 2.12.0\r\n- pyarrow 11.0.0\r\n",
"problem solved by using dataset split, but i don't know what's different between \"split and subset\""
] | How to reproduce:
```python
from datasets import load_dataset
wiki = load_dataset("wikipedia", "20200501.en", split="train")
wiki[[0]]
---------------------------------------------------------------------------
ArrowInvalid Traceback (most recent call last)
<ipython-input-13-381aedc9811b> in <module>
----> 1 wikipedia[[0]]
~/Desktop/hf/nlp/src/datasets/arrow_dataset.py in __getitem__(self, key)
1069 format_columns=self._format_columns,
1070 output_all_columns=self._output_all_columns,
-> 1071 format_kwargs=self._format_kwargs,
1072 )
1073
~/Desktop/hf/nlp/src/datasets/arrow_dataset.py in _getitem(self, key, format_type, format_columns, output_all_columns, format_kwargs)
1037 )
1038 else:
-> 1039 data_subset = self._data.take(indices_array)
1040
1041 if format_type is not None:
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.take()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/compute.py in take(data, indices, boundscheck)
266 """
267 options = TakeOptions(boundscheck)
--> 268 return call_function('take', [data, indices], options)
269
270
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/_compute.pyx in pyarrow._compute.call_function()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/_compute.pyx in pyarrow._compute.Function.call()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowInvalid: offset overflow while concatenating arrays
```
It seems to work fine with small datasets or with pyarrow 0.17.1 | 615 |
https://github.com/huggingface/datasets/issues/611 | ArrowCapacityError: List array cannot contain more than 2147483646 child elements, have 2147483648 | [
"Can you give us stats/information on your pandas DataFrame?",
"```\r\n<class 'pandas.core.frame.DataFrame'>\r\nInt64Index: 17136104 entries, 0 to 17136103\r\nData columns (total 6 columns):\r\n # Column Dtype \r\n--- ------ ----- \r\n 0 item_id int64 \r\n 1 item_titl object \r\n 2 start_price float64\r\n 3 shipping_fee float64\r\n 4 picture_url object \r\n 5 embeddings object \r\ndtypes: float64(2), int64(1), object(3)\r\nmemory usage: 915.2+ MB\r\n```",
"Thanks and some more on the `embeddings` and `picture_url` would be nice as well (type and max lengths of the elements)",
"`embedding` is `np.array` of shape `(128,)`. `picture_url` is url, such as 'https://i.ebayimg.com/00/s/MTE5OVgxNjAw/z/ZOsAAOSwAG9fHQq5/$_12.JPG?set_id=880000500F;https://i.ebayimg.com/00/s/MTE5OVgxNjAw/z/OSgAAOSwokBfHQq8/$_12.JPG?set_id=880000500F'",
"It looks like a Pyarrow limitation.\r\nI was able to reproduce the error with \r\n\r\n```python\r\nimport pandas as pd\r\nimport numpy as np\r\nimport pyarrow as pa\r\n\r\n n = 1713614\r\ndf = pd.DataFrame.from_dict({\"a\": list(np.zeros((n, 128))), \"b\": range(n)})\r\npa.Table.from_pandas(df)\r\n```\r\n\r\nI also tried with 50% of the dataframe and it actually works.\r\nI created an issue on Apache Arrow's JIRA [here](https://issues.apache.org/jira/browse/ARROW-9976)\r\n\r\nOne way to fix that would be to chunk the dataframe and concatenate arrow tables.",
"It looks like it's going to be fixed in pyarrow 2.0.0 :)\r\n\r\nIn the meantime I suggest to chunk big dataframes to create several small datasets, and then concatenate them using [concatenate_datasets](https://huggingface.co/docs/datasets/package_reference/main_classes.html?highlight=concatenate#datasets.concatenate_datasets)"
] | Hi, I'm trying to load a dataset from Dataframe, but I get the error:
```bash
---------------------------------------------------------------------------
ArrowCapacityError Traceback (most recent call last)
<ipython-input-7-146b6b495963> in <module>
----> 1 dataset = Dataset.from_pandas(emb)
~/miniconda3/envs/dev/lib/python3.7/site-packages/nlp/arrow_dataset.py in from_pandas(cls, df, features, info, split)
223 info.features = features
224 pa_table: pa.Table = pa.Table.from_pandas(
--> 225 df=df, schema=pa.schema(features.type) if features is not None else None
226 )
227 return cls(pa_table, info=info, split=split)
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.from_pandas()
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/pandas_compat.py in dataframe_to_arrays(df, schema, preserve_index, nthreads, columns, safe)
591 for i, maybe_fut in enumerate(arrays):
592 if isinstance(maybe_fut, futures.Future):
--> 593 arrays[i] = maybe_fut.result()
594
595 types = [x.type for x in arrays]
~/miniconda3/envs/dev/lib/python3.7/concurrent/futures/_base.py in result(self, timeout)
426 raise CancelledError()
427 elif self._state == FINISHED:
--> 428 return self.__get_result()
429
430 self._condition.wait(timeout)
~/miniconda3/envs/dev/lib/python3.7/concurrent/futures/_base.py in __get_result(self)
382 def __get_result(self):
383 if self._exception:
--> 384 raise self._exception
385 else:
386 return self._result
~/miniconda3/envs/dev/lib/python3.7/concurrent/futures/thread.py in run(self)
55
56 try:
---> 57 result = self.fn(*self.args, **self.kwargs)
58 except BaseException as exc:
59 self.future.set_exception(exc)
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/pandas_compat.py in convert_column(col, field)
557
558 try:
--> 559 result = pa.array(col, type=type_, from_pandas=True, safe=safe)
560 except (pa.ArrowInvalid,
561 pa.ArrowNotImplementedError,
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/array.pxi in pyarrow.lib.array()
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/array.pxi in pyarrow.lib._ndarray_to_array()
~/miniconda3/envs/dev/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowCapacityError: List array cannot contain more than 2147483646 child elements, have 2147483648
```
My code is :
```python
from nlp import Dataset
dataset = Dataset.from_pandas(emb)
``` | 611 |
https://github.com/huggingface/datasets/issues/610 | Load text file for RoBERTa pre-training. | [
"Could you try\r\n```python\r\nload_dataset('text', data_files='test.txt',cache_dir=\"./\", split=\"train\")\r\n```\r\n?\r\n\r\n`load_dataset` returns a dictionary by default, like {\"train\": your_dataset}",
"Hi @lhoestq\r\nThanks for your suggestion.\r\n\r\nI tried \r\n```\r\ndataset = load_dataset('text', data_files='test.txt',cache_dir=\"./\", split=\"train\")\r\nprint(dataset)\r\ndataset.set_format(type='torch',columns=[\"text\"])\r\ndataloader = torch.utils.data.DataLoader(dataset, batch_size=8)\r\nnext(iter(dataloader))\r\n```\r\n\r\nBut it still doesn't work and got error:\r\n```\r\n---------------------------------------------------------------------------\r\nTypeError Traceback (most recent call last)\r\n<ipython-input-7-388aca337e2f> in <module>\r\n----> 1 next(iter(dataloader))\r\n\r\n/Library/Python/3.7/site-packages/torch/utils/data/dataloader.py in __next__(self)\r\n 361 \r\n 362 def __next__(self):\r\n--> 363 data = self._next_data()\r\n 364 self._num_yielded += 1\r\n 365 if self._dataset_kind == _DatasetKind.Iterable and \\\r\n\r\n/Library/Python/3.7/site-packages/torch/utils/data/dataloader.py in _next_data(self)\r\n 401 def _next_data(self):\r\n 402 index = self._next_index() # may raise StopIteration\r\n--> 403 data = self._dataset_fetcher.fetch(index) # may raise StopIteration\r\n 404 if self._pin_memory:\r\n 405 data = _utils.pin_memory.pin_memory(data)\r\n\r\n/Library/Python/3.7/site-packages/torch/utils/data/_utils/fetch.py in fetch(self, possibly_batched_index)\r\n 42 def fetch(self, possibly_batched_index):\r\n 43 if self.auto_collation:\r\n---> 44 data = [self.dataset[idx] for idx in possibly_batched_index]\r\n 45 else:\r\n 46 data = self.dataset[possibly_batched_index]\r\n\r\n/Library/Python/3.7/site-packages/torch/utils/data/_utils/fetch.py in <listcomp>(.0)\r\n 42 def fetch(self, possibly_batched_index):\r\n 43 if self.auto_collation:\r\n---> 44 data = [self.dataset[idx] for idx in possibly_batched_index]\r\n 45 else:\r\n 46 data = self.dataset[possibly_batched_index]\r\n\r\n/Library/Python/3.7/site-packages/datasets-0.4.0-py3.7.egg/datasets/arrow_dataset.py in __getitem__(self, key)\r\n 1069 format_columns=self._format_columns,\r\n 1070 output_all_columns=self._output_all_columns,\r\n-> 1071 format_kwargs=self._format_kwargs,\r\n 1072 )\r\n 1073 \r\n\r\n/Library/Python/3.7/site-packages/datasets-0.4.0-py3.7.egg/datasets/arrow_dataset.py in _getitem(self, key, format_type, format_columns, output_all_columns, format_kwargs)\r\n 1056 format_columns=format_columns,\r\n 1057 output_all_columns=output_all_columns,\r\n-> 1058 format_kwargs=format_kwargs,\r\n 1059 )\r\n 1060 return outputs\r\n\r\n/Library/Python/3.7/site-packages/datasets-0.4.0-py3.7.egg/datasets/arrow_dataset.py in _convert_outputs(self, outputs, format_type, format_columns, output_all_columns, format_kwargs)\r\n 872 continue\r\n 873 if format_columns is None or k in format_columns:\r\n--> 874 v = map_nested(command, v, **map_nested_kwargs)\r\n 875 output_dict[k] = v\r\n 876 return output_dict\r\n\r\n/Library/Python/3.7/site-packages/datasets-0.4.0-py3.7.egg/datasets/utils/py_utils.py in map_nested(function, data_struct, dict_only, map_list, map_tuple, map_numpy, num_proc, types)\r\n 214 # Singleton\r\n 215 if not isinstance(data_struct, dict) and not isinstance(data_struct, types):\r\n--> 216 return function(data_struct)\r\n 217 \r\n 218 disable_tqdm = bool(logger.getEffectiveLevel() > INFO)\r\n\r\n/Library/Python/3.7/site-packages/datasets-0.4.0-py3.7.egg/datasets/arrow_dataset.py in command(x)\r\n 833 if x.dtype == np.object: # pytorch tensors cannot be instantied from an array of objects\r\n 834 return [map_nested(command, i, **map_nested_kwargs) for i in x]\r\n--> 835 return torch.tensor(x, **format_kwargs)\r\n 836 \r\n 837 elif format_type == \"tensorflow\":\r\n\r\nTypeError: new(): invalid data type 'str'\r\n```\r\n\r\nI found type can be ['numpy', 'torch', 'tensorflow', 'pandas'] only, how can I deal with the string type?",
"You need to tokenize the string inputs to convert them in integers before you can feed them to a pytorch dataloader.\r\n\r\nYou can read the quicktour of the datasets or the transformers libraries to know more about that:\r\n- transformers: https://huggingface.co/transformers/quicktour.html\r\n- dataset: https://huggingface.co/docs/datasets/quicktour.html",
"Hey @chiyuzhang94, I was also having trouble in loading a large text file (11GB).\r\nBut finally got it working. This is what I did after looking into the documentation.\r\n\r\n1. split the whole dataset file into smaller files\r\n```bash\r\nmkdir ./shards\r\nsplit -a 4 -l 256000 -d full_raw_corpus.txt ./shards/shard_\r\n````\r\n2. Pass paths of small data files to `load_dataset`\r\n```python\r\nfiles = glob.glob('shards/*')\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('text', data_files=files, split='train')\r\n```\r\n(On passing the whole dataset file (11GB) directly to `load_dataset` was resulting into RAM issue)\r\n\r\n3. Tokenization\r\n```python\r\ndef encode(examples):\r\n return tokenizer(examples['text'], truncation=True, padding='max_length')\r\ndataset = dataset.map(encode, batched=True)\r\ndataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\n```\r\n Now you can pass `dataset` to `Trainer` or `pytorch DataLoader`\r\n```python\r\ndataloader = torch.utils.data.DataLoader(dataset, batch_size=4)\r\nnext(iter(dataloader))\r\n```\r\nHope this helps\r\n",
"Thanks, @thomwolf and @sipah00 ,\r\n\r\nI tried to implement your suggestions in my scripts. \r\nNow, I am facing some connection time-out error. I am using my local file, I have no idea why the module request s3 database.\r\n\r\nThe log is:\r\n```\r\nTraceback (most recent call last):\r\n File \"/home/.local/lib/python3.6/site-packages/requests/adapters.py\", line 449, in send\r\n raise err\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/util/connection.py\", line 74, in create_connection\r\n timeout=timeout\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connectionpool.py\", line 720, in urlopen\r\n sock.connect(sa)\r\nTimeoutError: [Errno 110] Connection timed out\r\n\r\nTraceback (most recent call last):\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connectionpool.py\", line 672, in urlopen\r\n method, url, error=e, _pool=self, _stacktrace=sys.exc_info()[2]\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/util/retry.py\", line 436, in increment\r\n chunked=chunked,\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connectionpool.py\", line 376, in _make_request\r\n raise MaxRetryError(_pool, url, error or ResponseError(cause))\r\nurllib3.exceptions.MaxRetryError: HTTPSConnectionPool(host='s3.amazonaws.com', port=443): Max retries exceeded with url: /datasets.huggingface.co/datasets/datasets/text/text.py (Caused by NewConnectionError('<urllib3.connection.VerifiedHTTPSConnection obj\r\nect at 0x7fff401e0e48>: Failed to establish a new connection: [Errno 110] Connection timed out',))\r\n\r\nTraceback (most recent call last):\r\n File \"/scratch/roberta_emohash/run_language_modeling.py\", line 1019, in <module>\r\n main()\r\n File \"/scratch/roberta_emohash/run_language_modeling.py\", line 962, in main\r\n train_dataset = load_and_cache_examples(args, tokenizer, evaluate=False)\r\n File \"/scratch/roberta_emohash/run_language_modeling.py\", line 177, in load_and_cache_examples\r\n return HG_Datasets(tokenizer, file_path, args)\r\n File \"/scratch/roberta_emohash/run_language_modeling.py\", line 117, in HG_Datasets\r\n dataset = load_dataset('text', data_files=files, cache_dir = args.data_cache_dir, split=\"train\")\r\n File \"/arc/project/evn_py36/datasets/datasets/src/datasets/load.py\", line 590, in load_dataset\r\n self._validate_conn(conn)\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connectionpool.py\", line 994, in _validate_conn\r\n conn.connect()\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connection.py\", line 300, in connect\r\n conn = self._new_conn()\r\n File \"/home/.local/lib/python3.6/site-packages/urllib3/connection.py\", line 169, in _new_conn\r\n self, \"Failed to establish a new connection: %s\" % e\r\nurllib3.exceptions.NewConnectionError: <urllib3.connection.VerifiedHTTPSConnection object at 0x7fff401e0da0>: Failed to establish a new connection: [Errno 110] Connection timed out\r\n\r\n``` \r\n\r\nDo you have any experience on this issue?",
"No, I didn't encounter this problem, it seems to me a network problem",
"I noticed this is because I use a cloud server where does not provide for connections from our standard compute nodes to outside resources. \r\n\r\nFor the `datasets` package, it seems that if the loading script is not already cached in the library it will attempt to connect to an AWS resource to download the dataset loading script. \r\n\r\nI am wondering why the package works in this way. Do you have any suggestions to solve this issue? ",
"I solved the above issue by downloading text.py manually and passing the path to the `load_dataset` function. \r\n\r\nNow, I have a new issue with the Read-only file system.\r\n\r\nThe error is: \r\n```\r\nI0916 22:14:38.453380 140737353971520 filelock.py:274] Lock 140734268996072 acquired on /scratch/chiyuzh/roberta/text.py.lock\r\nFound main folder for dataset /scratch/chiyuzh/roberta/text.py at /home/chiyuzh/.cache/huggingface/modules/datasets_modules/datasets/text\r\nCreating specific version folder for dataset /scratch/chiyuzh/roberta/text.py at /home/chiyuzh/.cache/huggingface/modules/datasets_modules/datasets/text/512f465342e4f4cd07a8791428a629c043bb89d55ad7817cbf7fcc649178b014\r\nI0916 22:14:38.530371 140737353971520 filelock.py:318] Lock 140734268996072 released on /scratch/chiyuzh/roberta/text.py.lock\r\nTraceback (most recent call last):\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling_hg.py\", line 1019, in <module>\r\n main()\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling_hg.py\", line 962, in main\r\n train_dataset = load_and_cache_examples(args, tokenizer, evaluate=False)\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling_hg.py\", line 177, in load_and_cache_examples\r\n return HG_Datasets(tokenizer, file_path, args)\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling_hg.py\", line 117, in HG_Datasets\r\n dataset = load_dataset('/scratch/chiyuzh/roberta/text.py', data_files=files, cache_dir = args.data_cache_dir, split=\"train\")\r\n File \"/arc/project/chiyuzh/evn_py36/datasets/src/datasets/load.py\", line 590, in load_dataset\r\n path, script_version=script_version, download_config=download_config, download_mode=download_mode, dataset=True\r\n File \"/arc/project/chiyuzh/evn_py36/datasets/src/datasets/load.py\", line 385, in prepare_module\r\n os.makedirs(hash_folder_path)\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/os.py\", line 220, in makedirs\r\n mkdir(name, mode)\r\nOSError: [Errno 30] Read-only file system: '/home/chiyuzh/.cache/huggingface/modules/datasets_modules/datasets/text/512f465342e4f4cd07a8791428a629c043bb89d55ad7817cbf7fcc649178b014'\r\n\r\n```\r\n\r\nI installed datasets at /project/chiyuzh/evn_py36/datasets/src where is a writable directory.\r\nI also tried change the environment variables to the writable directory:\r\n`export HF_MODULES_PATH=/project/chiyuzh/evn_py36/datasets/cache_dir/`\r\n`export HF_DATASETS_CACHE=/project/chiyuzh/evn_py36/datasets/cache_dir/`\r\n \r\nIn my scripts, I also changed to:\r\n`dataset = load_dataset('/scratch/chiyuzh/roberta/text.py', data_files=files, cache_dir = args.data_cache_dir, split=\"train\")`\r\n`data_cache_dir = $TMPDIR/data/` that also a writable directory.\r\n \r\nBut it still try to make directory at /home/chiyuzh/.cache/huggingface/modules/.\r\nDo you have any idea about this issue? @thomwolf \r\n",
"> Hey @chiyuzhang94, I was also having trouble in loading a large text file (11GB).\r\n> But finally got it working. This is what I did after looking into the documentation.\r\n> \r\n> 1. split the whole dataset file into smaller files\r\n> \r\n> ```shell\r\n> mkdir ./shards\r\n> split -a 4 -l 256000 -d full_raw_corpus.txt ./shards/shard_\r\n> ```\r\n> \r\n> 1. Pass paths of small data files to `load_dataset`\r\n> \r\n> ```python\r\n> files = glob.glob('shards/*')\r\n> from datasets import load_dataset\r\n> dataset = load_dataset('text', data_files=files, split='train')\r\n> ```\r\n> \r\n> (On passing the whole dataset file (11GB) directly to `load_dataset` was resulting into RAM issue)\r\n> \r\n> 1. Tokenization\r\n> \r\n> ```python\r\n> def encode(examples):\r\n> return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> dataset = dataset.map(encode, batched=True)\r\n> dataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\n> ```\r\n> \r\n> Now you can pass `dataset` to `Trainer` or `pytorch DataLoader`\r\n> \r\n> ```python\r\n> dataloader = torch.utils.data.DataLoader(dataset, batch_size=4)\r\n> next(iter(dataloader))\r\n> ```\r\n> \r\n> Hope this helps\r\n\r\nWhen I run 'dataset = dataset.map(encode, batched=True)',\r\nI encountered a problem like this:\r\n\r\n> Testing the mapped function outputs\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/dataset_dict.py\", line 300, in map\r\n for k, dataset in self.items()\r\n File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/dataset_dict.py\", line 300, in <dictcomp>\r\n for k, dataset in self.items()\r\n File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1224, in map\r\n update_data = does_function_return_dict(test_inputs, test_indices)\r\n File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1195, in does_function_return_dict\r\n function(*fn_args, indices, **fn_kwargs) if with_indices else function(*fn_args, **fn_kwargs)\r\n File \"<stdin>\", line 3, in encode\r\nTypeError: __init__() takes 1 positional argument but 2 were given",
"> > Hey @chiyuzhang94, I was also having trouble in loading a large text file (11GB).\r\n> > But finally got it working. This is what I did after looking into the documentation.\r\n> > \r\n> > 1. split the whole dataset file into smaller files\r\n> > \r\n> > ```shell\r\n> > mkdir ./shards\r\n> > split -a 4 -l 256000 -d full_raw_corpus.txt ./shards/shard_\r\n> > ```\r\n> > \r\n> > \r\n> > \r\n> > 1. Pass paths of small data files to `load_dataset`\r\n> > \r\n> > ```python\r\n> > files = glob.glob('shards/*')\r\n> > from datasets import load_dataset\r\n> > dataset = load_dataset('text', data_files=files, split='train')\r\n> > ```\r\n> > \r\n> > \r\n> > (On passing the whole dataset file (11GB) directly to `load_dataset` was resulting into RAM issue)\r\n> > \r\n> > 1. Tokenization\r\n> > \r\n> > ```python\r\n> > def encode(examples):\r\n> > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> > dataset = dataset.map(encode, batched=True)\r\n> > dataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\n> > ```\r\n> > \r\n> > \r\n> > Now you can pass `dataset` to `Trainer` or `pytorch DataLoader`\r\n> > ```python\r\n> > dataloader = torch.utils.data.DataLoader(dataset, batch_size=4)\r\n> > next(iter(dataloader))\r\n> > ```\r\n> > \r\n> > \r\n> > Hope this helps\r\n> \r\n> When I run 'dataset = dataset.map(encode, batched=True)',\r\n> I encountered a problem like this:\r\n> \r\n> > Testing the mapped function outputs\r\n> > Traceback (most recent call last):\r\n> > File \"\", line 1, in \r\n> > File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/dataset_dict.py\", line 300, in map\r\n> > for k, dataset in self.items()\r\n> > File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/dataset_dict.py\", line 300, in \r\n> > for k, dataset in self.items()\r\n> > File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1224, in map\r\n> > update_data = does_function_return_dict(test_inputs, test_indices)\r\n> > File \"/anaconda3/envs/torch-xla-1.6/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1195, in does_function_return_dict\r\n> > function(*fn_args, indices, **fn_kwargs) if with_indices else function(*fn_args, **fn_kwargs)\r\n> > File \"\", line 3, in encode\r\n> > TypeError: **init**() takes 1 positional argument but 2 were given\r\n\r\nWhat is your encoder function?",
"> ```python\r\n> def encode(examples):\r\n> return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> ```\r\n\r\nIt is the same as suggested:\r\n\r\n> def encode(examples):\r\n return tokenizer(examples['text'], truncation=True, padding='max_length')",
"> > ```python\r\n> > def encode(examples):\r\n> > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> > ```\r\n> \r\n> It is the same as suggested:\r\n> \r\n> > def encode(examples):\r\n> > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n\r\nDo you use this function in a `class` object? \r\n\r\ninit() takes 1 positional argument but 2 were given. I guess the additional argument is self?",
"> > > ```python\r\n> > > def encode(examples):\r\n> > > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> > > ```\r\n> > \r\n> > \r\n> > It is the same as suggested:\r\n> > > def encode(examples):\r\n> > > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> \r\n> Do you use this function in a `class` object?\r\n> \r\n> init() takes 1 positional argument but 2 were given. I guess the additional argument is self?\r\n\r\nThanks for your reply.\r\nCould you provide some simple example here?\r\nCurrently, I do not use this function in a class object. \r\nI think you are right and I was wondering how to construct this class.\r\nI try to modify it based on transformers' LineByLineTextDataset. Am I correct?\r\n\r\n> class LineByLineTextDataset(Dataset):\r\n \"\"\"\r\n This will be superseded by a framework-agnostic approach\r\n soon.\r\n \"\"\"\r\n\r\n def __init__(self, tokenizer: PreTrainedTokenizer, file_path: str, block_size: int):\r\n assert os.path.isfile(file_path), f\"Input file path {file_path} not found\"\r\n # Here, we do not cache the features, operating under the assumption\r\n # that we will soon use fast multithreaded tokenizers from the\r\n # `tokenizers` repo everywhere =)\r\n #logger.info(\"Creating features from dataset file at %s\", file_path)\r\n #with open(file_path, encoding=\"utf-8\") as f:\r\n # lines = [line for line in f.read().splitlines() if (len(line) > 0 and not line.isspace())]\r\n #batch_encoding = tokenizer(lines, add_special_tokens=True, truncation=True, max_length=block_size)\r\n\r\n\timport glob\r\n\tfiles = glob.glob('/home/mtzhang111/fairseq/cs_doc/shards/shard_003*')\r\n\tfrom datasets import load_dataset\r\n\tdataset = load_dataset('text', data_files=files)\r\n batch_encoding= dataset.map(encode, batched=True)\r\n self.examples = batch_encoding[\"input_ids\"]\r\n\t\r\n\r\n def encode(examples):\r\n return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n\r\n def __len__(self):\r\n return len(self.examples)\r\n\r\n def __getitem__(self, i) -> torch.Tensor:\r\n return torch.tensor(self.examples[i], dtype=torch.long)\r\n",
"> > > > ```python\r\n> > > > def encode(examples):\r\n> > > > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> > > > ```\r\n> > > \r\n> > > \r\n> > > It is the same as suggested:\r\n> > > > def encode(examples):\r\n> > > > return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> > \r\n> > \r\n> > Do you use this function in a `class` object?\r\n> > init() takes 1 positional argument but 2 were given. I guess the additional argument is self?\r\n> \r\n> Thanks for your reply.\r\n> Could you provide some simple example here?\r\n> Currently, I do not use this function in a class object.\r\n> I think you are right and I was wondering how to construct this class.\r\n> I try to modify it based on transformers' LineByLineTextDataset. Am I correct?\r\n> \r\n> > class LineByLineTextDataset(Dataset):\r\n> > \"\"\"\r\n> > This will be superseded by a framework-agnostic approach\r\n> > soon.\r\n> > \"\"\"\r\n> \r\n> ```\r\n> def __init__(self, tokenizer: PreTrainedTokenizer, file_path: str, block_size: int):\r\n> assert os.path.isfile(file_path), f\"Input file path {file_path} not found\"\r\n> # Here, we do not cache the features, operating under the assumption\r\n> # that we will soon use fast multithreaded tokenizers from the\r\n> # `tokenizers` repo everywhere =)\r\n> #logger.info(\"Creating features from dataset file at %s\", file_path)\r\n> #with open(file_path, encoding=\"utf-8\") as f:\r\n> # lines = [line for line in f.read().splitlines() if (len(line) > 0 and not line.isspace())]\r\n> #batch_encoding = tokenizer(lines, add_special_tokens=True, truncation=True, max_length=block_size)\r\n> \r\n> import glob\r\n> files = glob.glob('/home/mtzhang111/fairseq/cs_doc/shards/shard_003*')\r\n> from datasets import load_dataset\r\n> dataset = load_dataset('text', data_files=files)\r\n> batch_encoding= dataset.map(encode, batched=True)\r\n> self.examples = batch_encoding[\"input_ids\"]\r\n> \r\n> \r\n> def encode(examples):\r\n> return tokenizer(examples['text'], truncation=True, padding='max_length')\r\n> \r\n> def __len__(self):\r\n> return len(self.examples)\r\n> \r\n> def __getitem__(self, i) -> torch.Tensor:\r\n> return torch.tensor(self.examples[i], dtype=torch.long)\r\n> ```\r\n\r\nI am also struggling with this adaptation. \r\nI am not sure whether I am right.\r\n\r\nI think you don't need to construct `class LazyLineByLineTextDataset(Dataset)` at all. \r\ntorch.utils.data.Dataset is a generator. \r\n\r\nNow, we use `dataset = dataset.map(encode, batched=True)` as a generator. So we just pass dataset to torch.utils.data.DataLoader. ",
"@chiyuzhang94 Thanks for your reply. After some changes, currently, I managed to make the data loading process running.\r\nI published it in case you might want to take a look. Thanks for your help!\r\nhttps://github.com/shizhediao/Transformers_TPU",
"Hi @shizhediao ,\r\n\r\nThanks! It looks great!\r\n\r\nBut my problem still is the cache directory is a read-only file system. \r\n[As I mentioned](https://github.com/huggingface/datasets/issues/610#issuecomment-693912285), I tried to change the cache directory but it didn't work. \r\n\r\nDo you have any suggestions?\r\n\r\n",
"> I installed datasets at /project/chiyuzh/evn_py36/datasets/src where is a writable directory.\r\n> I also tried change the environment variables to the writable directory:\r\n> `export HF_MODULES_PATH=/project/chiyuzh/evn_py36/datasets/cache_dir/`\r\n\r\nI think it is `HF_MODULES_CACHE` and not `HF_MODULES_PATH` @chiyuzhang94 .\r\nCould you try again and let me know if it fixes your issue ?\r\n",
"We should probably add a section in the doc on the caching system with the env variables in particular.",
"Hi @thomwolf , @lhoestq ,\r\n\r\nThanks for your suggestions. With the latest version of this package, I can load text data without Internet. \r\n\r\nBut I found the speed of dataset loading is very slow. \r\n\r\nMy scrips like this: \r\n```\r\n def token_encode(examples):\r\n tokenizer_out = tokenizer(examples['text'], truncation=True, padding=\"max_length\", add_special_tokens=True, max_length=args.block_size)\r\n return tokenizer_out\r\n \r\n path = Path(file_path)\r\n files = sorted(path.glob('*'))\r\n dataset = load_dataset('./text.py', data_files=files, cache_dir = args.data_cache_dir, split=\"train\")\r\n dataset = dataset.map(token_encode, batched=True)\r\n\r\n dataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\n```\r\n\r\nI have 1,123,870,657 lines in my input directory. \r\nI can find the processing speed as following. It is very slow. \r\n```\r\n| 13/1123871 [00:02<62:37:39, 4.98ba/s]^M 0%| \r\n| 14/1123871 [00:03<61:27:31, 5.08ba/s]^M 0%| \r\n| 15/1123871 [00:03<66:34:19, 4.69ba/s]^M 0%| \r\n| 16/1123871 [00:03<68:25:01, 4.56ba/s]^M 0%| \r\n| 17/1123871 [00:03<72:00:03, 4.34ba/s]^M 0%| \r\n```\r\nDo you have any suggestions to accelerate this loading process?",
"You can use multiprocessing by specifying `num_proc=` in `.map()`\r\n\r\nAlso it looks like you have `1123871` batches of 1000 elements (default batch size), i.e. 1,123,871,000 lines in total.\r\nAm I right ?",
"> You can use multiprocessing by specifying `num_proc=` in `.map()`\r\n> \r\n> Also it looks like you have `1123871` batches of 1000 elements (default batch size), i.e. 1,123,871,000 lines in total.\r\n> Am I right ?\r\n\r\nHi @lhoestq ,\r\n\r\nThanks. I will try it.\r\n\r\nYou are right. I have 1,123,870,657 lines totally in the path. I split the large file into 440 small files. Each file has 2,560,000 lines.\r\n\r\nI have another question. Because I am using a cloud server where only allows running a job up to 7 days. Hence, I need to resume my model every week. If the script needs to load and process the dataset every time. It is very low efficient based on the current processing speed. Is it possible that I process the dataset once and use the process cache to in the future work? \r\n",
"Hi @lhoestq ,\r\n\r\nI tried to use multi-processor, but I got errors as follow: \r\nBecause I am using python distributed training, it seems some conflicts with the distributed job. \r\n\r\nDo you have any suggestions?\r\n```\r\nI0925 10:19:35.603023 140737353971520 filelock.py:318] Lock 140737229443368 released on /tmp/pbs.1120510.pbsha.ib.sockeye/cache/_tmp_pbs.1120510.pbsha.ib.sockeye_cache_text_default-7fb934ed6fac5d01_0.0.0_512f465342e4f4cd07a8791428a629c043bb89d55ad7817cbf7\r\nfcc649178b014.lock\r\nTraceback (most recent call last):\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling.py\", line 1024, in <module>\r\n main()\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling.py\", line 967, in main\r\n train_dataset = load_and_cache_examples(args, tokenizer, evaluate=False)\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling.py\", line 180, in load_and_cache_examples\r\n return HG_Datasets(tokenizer, file_path, args)\r\n File \"/scratch/chiyuzh/roberta/run_language_modeling.py\", line 119, in HG_Datasets\r\n dataset = dataset.map(token_encode, batched=True, batch_size = 10000, num_proc = 16)\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1287, in map\r\n transformed_shards = [r.get() for r in results]\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/site-packages/datasets/arrow_dataset.py\", line 1287, in <listcomp>\r\n transformed_shards = [r.get() for r in results]\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/multiprocessing/pool.py\", line 644, in get\r\n raise self._value\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/multiprocessing/pool.py\", line 424, in _handle_tasks\r\n put(task)\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/multiprocessing/connection.py\", line 206, in send\r\n self._send_bytes(_ForkingPickler.dumps(obj))\r\n File \"/project/chiyuzh/evn_py36/lib/python3.6/multiprocessing/reduction.py\", line 51, in dumps\r\n cls(buf, protocol).dump(obj)\r\nAttributeError: Can't pickle local object 'HG_Datasets.<locals>.token_encode'\r\n```",
"For multiprocessing, the function given to `map` must be picklable.\r\nMaybe you could try to define `token_encode` outside `HG_Datasets` ?\r\n\r\nAlso maybe #656 could make functions defined locally picklable for multiprocessing, once it's merged.",
"> I have another question. Because I am using a cloud server where only allows running a job up to 7 days. Hence, I need to resume my model every week. If the script needs to load and process the dataset every time. It is very low efficient based on the current processing speed. Is it possible that I process the dataset once and use the process cache to in the future work?\r\n\r\nFeel free to save your processed dataset using `dataset.save_to_disk(\"path/to/save/directory\")`.\r\n\r\nThen you'll be able to reload it again using\r\n```python\r\nfrom datasets import load_from_disk\r\n\r\ndataset = load_from_disk(\"path/to/save/directory\")\r\n```",
"Hi @lhoestq ,\r\n\r\nThanks for your suggestion. \r\nI tried to process the dataset and save it to disk. \r\nI have 1.12B samples in the raw dataset. I used 16 processors.\r\nI run this process job for 7 days. But it didn't finish. I don't why the processing is such slow. \r\n\r\nThe log shows that some processors (\\#12, \\#14, \\#15) are very slow. The different processor has a different speed. These slow processors look like a bottleneck. \r\n\r\nCould you please give me any suggestion to improve the processing speed? \r\n\r\nThanks. \r\nChiyu\r\n\r\nHere is my code:\r\n```\r\ndef token_encode(examples):\r\n tokenizer_out = tokenizer(examples['text'], truncation=True, padding=\"max_length\", add_special_tokens=True, max_length=args.block_size)\r\n return tokenizer_out\r\n\r\npath = Path(file_path)\r\nfiles = sorted(path.glob('*'))\r\ndataset = load_dataset('./text.py', data_files=files, cache_dir = args.data_cache_dir, split=\"train\")\r\ndataset = dataset.map(token_encode, batched=True, batch_size = 16384, num_proc = 16)\r\ndataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\ndataset.save_to_disk(output_dir)\r\n```\r\nHere is the log. \r\n```\r\n^M#6: 1%|β | 59/4288 [55:10<66:11:58, 56.35s/ba]\r\n^M#1: 8%|β | 356/4288 [55:39<10:40:02, 9.77s/ba]\r\n^M#2: 5%|β | 210/4288 [55:33<17:47:19, 15.70s/ba]\r\n^M#0: 19%|ββ | 836/4288 [55:53<4:08:56, 4.33s/ba]\r\n^M#0: 20%|ββ | 837/4288 [55:57<4:01:52, 4.21s/ba]\r\n^M#1: 8%|β | 357/4288 [55:48<10:38:09, 9.74s/ba]\r\n^M#0: 20%|ββ | 838/4288 [56:01<4:02:56, 4.23s/ba]\r\n^M#3: 4%|β | 155/4288 [55:43<24:41:20, 21.51s/ba]\r\n^M#0: 20%|ββ | 839/4288 [56:05<4:04:48, 4.26s/ba]\r\n^M#12: 1%| | 29/4288 [54:50<133:20:53, 112.72s/ba]\r\n^M#2: 5%|β | 211/4288 [55:48<17:40:33, 15.61s/ba]\r\n^M#14: 0%| | 2/4288 [04:24<157:17:50, 132.12s/ba]\r\n^M#15: 0%| | 1/4288 [02:24<172:11:37, 144.60s/ba]\r\n```",
"Hi !\r\n\r\nAs far as I can tell, there could be several reasons for your processes to have different speeds:\r\n- some parts of your dataset have short passages while some have longer passages, that take more time to be processed\r\n- OR there are other processes running that prevent some of them to run at full speed\r\n- OR the value of `num_proc` is higher than the number of actual processes that you can run in parallel at full speed.\r\n\r\nSo I'd suggest you to check that you have nothing else running in parallel to your processing job, and also maybe take a look at the slow parts of the datasets.\r\nWhen doing multiprocessing, the dataset is sharded in `num_proc` contiguous parts that are processed individually in each process. If you want to take a look at the dataset processed in the 12th shard of 16 for example, you can do:\r\n\r\n```python\r\nmy_shard = dataset.shard(num_shards=16, index=12, contiguous=True)\r\n```\r\n\r\nHope this helps, let me know if you find what is causing this slow down.",
"Do you use a fast or a slow tokenizer from the `transformers` library @chiyuzhang94?",
"> Do you use a fast or a slow tokenizer from the `transformers` library @chiyuzhang94?\r\n\r\nHi @thomwolf ,\r\n I use this: \r\n```\r\nfrom transformers import\r\nAutoTokenizer.from_pretrained(args.model_name_or_path, cache_dir=args.cache_dir)\r\n```\r\n\r\nI guess this is a slow one, let me explore the fast tokenizer. ",
"> Hi !\r\n> \r\n> As far as I can tell, there could be several reasons for your processes to have different speeds:\r\n> \r\n> * some parts of your dataset have short passages while some have longer passages, that take more time to be processed\r\n> * OR there are other processes running that prevent some of them to run at full speed\r\n> * OR the value of `num_proc` is higher than the number of actual processes that you can run in parallel at full speed.\r\n> \r\n> So I'd suggest you to check that you have nothing else running in parallel to your processing job, and also maybe take a look at the slow parts of the datasets.\r\n> When doing multiprocessing, the dataset is sharded in `num_proc` contiguous parts that are processed individually in each process. If you want to take a look at the dataset processed in the 12th shard of 16 for example, you can do:\r\n> \r\n> ```python\r\n> my_shard = dataset.shard(num_shards=16, index=12, contiguous=True)\r\n> ```\r\n> \r\n> Hope this helps, let me know if you find what is causing this slow down.\r\n\r\nHi @lhoestq ,\r\n\r\nThanks for your suggestions. \r\nI don't think my problem is due to any one of these seasons. \r\n\r\n1. I have 1,123,870,657 lines totally in the path. I split the large file into 440 small files. Each file has 2,560,000 lines. The last file is smaller a little bit. But they are similar. I randomly shuffled all the 1,123,870,657 lines. Hence, the sequences should also be similar across all the files. \r\n\r\n2. I run this script on the entire node. I requested all the resources on the nodes (40 CPUs, 384GB memory). Hence, these were not any other processes. \r\n\r\n 3. As I say, the node has 40 CPUs, but I set num_proc = 16. This should not be a problem.",
"Hi @thomwolf \r\nI am using `RobertaTokenizerFast` now. \r\n\r\nBut the speed is still imbalanced, some processors are still slow. \r\nHere is the part of the log. #0 is always much fast than lower rank processors. \r\n\r\n```\r\n#15: 3%|β | 115/3513 [3:18:36<98:01:33, 103.85s/ba]\r\n#2: 24%|βββ | 847/3513 [3:20:43<11:06:49, 15.01s/ba]\r\n#1: 37%|ββββ | 1287/3513 [3:20:52<6:19:02, 10.22s/ba]\r\n#0: 72%|ββββββββ | 2546/3513 [3:20:52<1:51:03, 6.89s/ba]\r\n#3: 18%|ββ | 617/3513 [3:20:36<15:50:30, 19.69s/ba]\r\n#0: 73%|ββββββββ | 2547/3513 [3:20:59<1:50:25, 6.86s/ba]\r\n#1: 37%|ββββ | 1288/3513 [3:21:02<6:21:13, 10.28s/ba]\r\n#7: 7%|β | 252/3513 [3:20:09<44:09:03, 48.74s/ba]\r\n#12: 4%|β | 144/3513 [3:19:19<78:00:54, 83.36s/ba]\r\n#4: 14%|ββ | 494/3513 [3:20:37<20:46:06, 24.77s/ba]\r\n#0: 73%|ββββββββ | 2548/3513 [3:21:06<1:49:26, 6.80s/ba]\r\n#2: 24%|βββ | 848/3513 [3:20:58<11:06:17, 15.00s/ba]\r\n```\r\nHere is my script related to the datasets processing, \r\n\r\n```\r\ntokenizer = RobertaTokenizerFast.from_pretrained(args.model_name_or_path, cache_dir=args.cache_dir)\r\n\r\ndef token_encode(examples):\r\n tokenizer_out = tokenizer(examples['text'], truncation=True, padding=\"max_length\", add_special_tokens=True, max_length=128)\r\n return tokenizer_out\r\n\r\ndef HG_Datasets(tokenizer, file_path, args):\r\n \r\n path = Path(file_path)\r\n files = sorted(path.glob('*'))\r\n dataset = load_dataset('./text.py', data_files=files, cache_dir = \"\"./, split=\"train\")\r\n dataset = dataset.map(token_encode, batched=True, batch_size = 20000, num_proc = 16)\r\n\r\n dataset.set_format(type='torch', columns=['input_ids', 'attention_mask'])\r\n return dataset\r\n\r\n```\r\nI have 1,123,870,657 lines totally in the path. I split the large file into 440 small files. Each file has 2,560,000 lines.\r\n\r\nCould you please give any suggestion? Thanks very much!!"
] | I migrate my question from https://github.com/huggingface/transformers/pull/4009#issuecomment-690039444
I tried to train a Roberta from scratch using transformers. But I got OOM issues with loading a large text file.
According to the suggestion from @thomwolf , I tried to implement `datasets` to load my text file. This test.txt is a simple sample where each line is a sentence.
```
from datasets import load_dataset
dataset = load_dataset('text', data_files='test.txt',cache_dir="./")
dataset.set_format(type='torch',columns=["text"])
dataloader = torch.utils.data.DataLoader(dataset, batch_size=8)
next(iter(dataloader))
```
But dataload cannot yield sample and error is:
```
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
<ipython-input-12-388aca337e2f> in <module>
----> 1 next(iter(dataloader))
/Library/Python/3.7/site-packages/torch/utils/data/dataloader.py in __next__(self)
361
362 def __next__(self):
--> 363 data = self._next_data()
364 self._num_yielded += 1
365 if self._dataset_kind == _DatasetKind.Iterable and \
/Library/Python/3.7/site-packages/torch/utils/data/dataloader.py in _next_data(self)
401 def _next_data(self):
402 index = self._next_index() # may raise StopIteration
--> 403 data = self._dataset_fetcher.fetch(index) # may raise StopIteration
404 if self._pin_memory:
405 data = _utils.pin_memory.pin_memory(data)
/Library/Python/3.7/site-packages/torch/utils/data/_utils/fetch.py in fetch(self, possibly_batched_index)
42 def fetch(self, possibly_batched_index):
43 if self.auto_collation:
---> 44 data = [self.dataset[idx] for idx in possibly_batched_index]
45 else:
46 data = self.dataset[possibly_batched_index]
/Library/Python/3.7/site-packages/torch/utils/data/_utils/fetch.py in <listcomp>(.0)
42 def fetch(self, possibly_batched_index):
43 if self.auto_collation:
---> 44 data = [self.dataset[idx] for idx in possibly_batched_index]
45 else:
46 data = self.dataset[possibly_batched_index]
KeyError: 0
```
`dataset.set_format(type='torch',columns=["text"])` returns a log says:
```
Set __getitem__(key) output type to torch for ['text'] columns (when key is int or slice) and don't output other (un-formatted) columns.
```
I noticed the dataset is `DatasetDict({'train': Dataset(features: {'text': Value(dtype='string', id=None)}, num_rows: 44)})`.
Each sample can be accessed by `dataset["train"]["text"]` instead of `dataset["text"]`.
Could you please give me any suggestions on how to modify this code to load the text file?
Versions:
Python version 3.7.3
PyTorch version 1.6.0
TensorFlow version 2.3.0
datasets version: 1.0.1 | 610 |
https://github.com/huggingface/datasets/issues/608 | Don't use the old NYU GLUE dataset URLs | [
"Feel free to open the PR ;)\r\nThanks for updating the dataset_info.json file !"
] | NYU is switching dataset hosting from Google to FB. Initial changes to `datasets` are in https://github.com/jeswan/nlp/commit/b7d4a071d432592ded971e30ef73330529de25ce. What tests do you suggest I run before opening a PR?
See: https://github.com/jiant-dev/jiant/issues/161 and https://github.com/nyu-mll/jiant/pull/1112 | 608 |
https://github.com/huggingface/datasets/issues/600 | Pickling error when loading dataset | [
"When I change from python3.6 to python3.8, it works! ",
"Does it work when you install `nlp` from source on python 3.6?",
"No, still the pickling error.",
"I wasn't able to reproduce on google colab (python 3.6.9 as well) with \r\n\r\npickle==4.0\r\ndill=0.3.2\r\ntransformers==3.1.0\r\ndatasets=1.0.1 (also tried nlp 0.4.0)\r\n\r\nIf I try\r\n\r\n```python\r\nfrom datasets import load_dataset # or from nlp\r\nfrom transformers import BertTokenizer\r\n\r\ntokenizer = BertTokenizer.from_pretrained(\"bert-base-uncased\")\r\ndataset = load_dataset(\"text\", data_files=file_path, split=\"train\")\r\ndataset = dataset.map(lambda ex: tokenizer(ex[\"text\"], add_special_tokens=True,\r\n truncation=True, max_length=512), batched=True)\r\ndataset.set_format(type='torch', columns=['input_ids'])\r\n```\r\nIt runs without error",
"Closing since it looks like it's working on >= 3.6.9\r\nFeel free to re-open if you have other questions :)"
] | Hi,
I modified line 136 in the original [run_language_modeling.py](https://github.com/huggingface/transformers/blob/master/examples/language-modeling/run_language_modeling.py) as:
```
# line 136: return LineByLineTextDataset(tokenizer=tokenizer, file_path=file_path, block_size=args.block_size)
dataset = load_dataset("text", data_files=file_path, split="train")
dataset = dataset.map(lambda ex: tokenizer(ex["text"], add_special_tokens=True,
truncation=True, max_length=args.block_size), batched=True)
dataset.set_format(type='torch', columns=['input_ids'])
return dataset
```
When I run this with transformers (3.1.0) and nlp (0.4.0), I get the following error:
```
Traceback (most recent call last):
File "src/run_language_modeling.py", line 319, in <module>
main()
File "src/run_language_modeling.py", line 248, in main
get_dataset(data_args, tokenizer=tokenizer, cache_dir=model_args.cache_dir) if training_args.do_train else None
File "src/run_language_modeling.py", line 139, in get_dataset
dataset = dataset.map(lambda ex: tokenizer(ex["text"], add_special_tokens=True, truncation=True, max_length=args.block_size), batched=True)
File "/data/nlp/src/nlp/arrow_dataset.py", line 1136, in map
new_fingerprint=new_fingerprint,
File "/data/nlp/src/nlp/fingerprint.py", line 158, in wrapper
self._fingerprint, transform, kwargs_for_fingerprint
File "/data/nlp/src/nlp/fingerprint.py", line 105, in update_fingerprint
hasher.update(transform_args[key])
File "/data/nlp/src/nlp/fingerprint.py", line 57, in update
self.m.update(self.hash(value).encode("utf-8"))
File "/data/nlp/src/nlp/fingerprint.py", line 53, in hash
return cls.hash_default(value)
File "/data/nlp/src/nlp/fingerprint.py", line 46, in hash_default
return cls.hash_bytes(dumps(value))
File "/data/nlp/src/nlp/utils/py_utils.py", line 362, in dumps
dump(obj, file)
File "/data/nlp/src/nlp/utils/py_utils.py", line 339, in dump
Pickler(file, recurse=True).dump(obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 446, in dump
StockPickler.dump(self, obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 409, in dump
self.save(obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 1438, in save_function
obj.__dict__, fkwdefaults), obj=obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 610, in save_reduce
save(args)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 751, in save_tuple
save(element)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 736, in save_tuple
save(element)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 1170, in save_cell
pickler.save_reduce(_create_cell, (f,), obj=obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 610, in save_reduce
save(args)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 736, in save_tuple
save(element)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 521, in save
self.save_reduce(obj=obj, *rv)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 605, in save_reduce
save(cls)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 1365, in save_type
obj.__bases__, _dict), obj=obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 610, in save_reduce
save(args)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 751, in save_tuple
save(element)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 933, in save_module_dict
StockPickler.save_dict(pickler, obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 821, in save_dict
self._batch_setitems(obj.items())
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 847, in _batch_setitems
save(v)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 476, in save
f(self, obj) # Call unbound method with explicit self
File "/root/miniconda3/envs/py3.6/lib/python3.6/site-packages/dill/_dill.py", line 933, in save_module_dict
StockPickler.save_dict(pickler, obj)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 821, in save_dict
self._batch_setitems(obj.items())
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 847, in _batch_setitems
save(v)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 507, in save
self.save_global(obj, rv)
File "/root/miniconda3/envs/py3.6/lib/python3.6/pickle.py", line 927, in save_global
(obj, module_name, name))
_pickle.PicklingError: Can't pickle typing.Union[str, NoneType]: it's not the same object as typing.Union
``` | 600 |
https://github.com/huggingface/datasets/issues/598 | The current version of the package on github has an error when loading dataset | [
"Thanks for reporting !\r\nWhich version of transformers are you using ?\r\nIt looks like it doesn't have the PreTrainedTokenizerBase class",
"I was using transformer 2.9. And I switch to the latest transformer package. Everything works just fine!!\r\n\r\nThanks for helping! I should look more carefully next time. Didn't realize loading the data part requires using tokenizer.\r\n",
"Yes it shouldnβt fail with older version of transformers since this is only a special feature to make caching more efficient when using transformers for tokenization.\r\nWeβll update this."
] | Instead of downloading the package from pip, downloading the version from source will result in an error when loading dataset (the pip version is completely fine):
To recreate the error:
First, installing nlp directly from source:
```
git clone https://github.com/huggingface/nlp.git
cd nlp
pip install -e .
```
Then run:
```
from nlp import load_dataset
dataset = load_dataset('wikitext', 'wikitext-2-v1',split = 'train')
```
will give error:
```
>>> dataset = load_dataset('wikitext', 'wikitext-2-v1',split = 'train')
Checking /home/zeyuy/.cache/huggingface/datasets/84a754b488511b109e2904672d809c041008416ae74e38f9ee0c80a8dffa1383.2e21f48d63b5572d19c97e441fbb802257cf6a4c03fbc5ed8fae3d2c2273f59e.py for additional imports.
Found main folder for dataset https://raw.githubusercontent.com/huggingface/nlp/0.4.0/datasets/wikitext/wikitext.py at /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext
Found specific version folder for dataset https://raw.githubusercontent.com/huggingface/nlp/0.4.0/datasets/wikitext/wikitext.py at /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d
Found script file from https://raw.githubusercontent.com/huggingface/nlp/0.4.0/datasets/wikitext/wikitext.py to /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d/wikitext.py
Found dataset infos file from https://raw.githubusercontent.com/huggingface/nlp/0.4.0/datasets/wikitext/dataset_infos.json to /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d/dataset_infos.json
Found metadata file for dataset https://raw.githubusercontent.com/huggingface/nlp/0.4.0/datasets/wikitext/wikitext.py at /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d/wikitext.json
Loading Dataset Infos from /home/zeyuy/.cache/huggingface/modules/nlp_modules/datasets/wikitext/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d
Overwrite dataset info from restored data version.
Loading Dataset info from /home/zeyuy/.cache/huggingface/datasets/wikitext/wikitext-2-v1/1.0.0/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d
Reusing dataset wikitext (/home/zeyuy/.cache/huggingface/datasets/wikitext/wikitext-2-v1/1.0.0/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d)
Constructing Dataset for split train, from /home/zeyuy/.cache/huggingface/datasets/wikitext/wikitext-2-v1/1.0.0/5de6e79516446f747fcccc09aa2614fa159053b75909594d28d262395f72d89d
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/load.py", line 600, in load_dataset
ds = builder_instance.as_dataset(split=split, ignore_verifications=ignore_verifications)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/builder.py", line 611, in as_dataset
datasets = utils.map_nested(
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/utils/py_utils.py", line 216, in map_nested
return function(data_struct)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/builder.py", line 631, in _build_single_dataset
ds = self._as_dataset(
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/builder.py", line 704, in _as_dataset
return Dataset(**dataset_kwargs)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/arrow_dataset.py", line 188, in __init__
self._fingerprint = generate_fingerprint(self)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/fingerprint.py", line 91, in generate_fingerprint
hasher.update(key)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/fingerprint.py", line 57, in update
self.m.update(self.hash(value).encode("utf-8"))
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/fingerprint.py", line 53, in hash
return cls.hash_default(value)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/fingerprint.py", line 46, in hash_default
return cls.hash_bytes(dumps(value))
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/utils/py_utils.py", line 361, in dumps
with _no_cache_fields(obj):
File "/home/zeyuy/miniconda3/lib/python3.8/contextlib.py", line 113, in __enter__
return next(self.gen)
File "/home/zeyuy/transformers/examples/language-modeling/nlp/src/nlp/utils/py_utils.py", line 348, in _no_cache_fields
if isinstance(obj, tr.PreTrainedTokenizerBase) and hasattr(obj, "cache") and isinstance(obj.cache, dict):
AttributeError: module 'transformers' has no attribute 'PreTrainedTokenizerBase'
```
| 598 |
https://github.com/huggingface/datasets/issues/597 | Indices incorrect with multiprocessing | [
"I fixed a bug that could cause this issue earlier today. Could you pull the latest version and try again ?",
"Still the case on master.\r\nI guess we should have an offset in the multi-procs indeed (hopefully it's enough).\r\n\r\nAlso, side note is that we should add some logging before the \"test\" to say we are testing the function otherwise its confusing for the user to see two outputs I think. Proposal (see the \"Testing the mapped function outputs:\" lines):\r\n```\r\n>>> d.select(range(10)).map(fn, with_indices=True, batched=True, num_proc=2)\r\nDone writing 10 indices in 80 bytes .\r\nDone writing 5 indices in 41 bytes .\r\nDone writing 5 indices in 41 bytes .\r\nSpawning 2 processes\r\nTesting the mapped function outputs:\r\ninds: [0, 1]\r\ninds: [0, 1]\r\nTesting finished, running the mapped function on the dataset:\r\n#0: 0%| | 0/1 [00:00<?, ?ba/s]\r\ninds: [0, 1, 2, 3, 4] inds: [0, 1, 2, 3, 4] | 0/1 [00:00<?, ?ba/s]\r\n#0: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 1321.04ba/s]\r\n#1: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [00:00<00:00, 1841.22ba/s]\r\nConcatenating 2 shards from multiprocessing\r\nDataset(features: {'text': Value(dtype='string', id=None), 'label': ClassLabel(num_classes=2, names=['neg', 'pos'], names_file=None, id=None)}, num_rows: 10)\r\n```"
] | When `num_proc` > 1, the indices argument passed to the map function is incorrect:
```python
d = load_dataset('imdb', split='test[:1%]')
def fn(x, inds):
print(inds)
return x
d.select(range(10)).map(fn, with_indices=True, batched=True)
# [0, 1]
# [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
d.select(range(10)).map(fn, with_indices=True, batched=True, num_proc=2)
# [0, 1]
# [0, 1]
# [0, 1, 2, 3, 4]
# [0, 1, 2, 3, 4]
```
As you can see, the subset passed to each thread is indexed from 0 to N which doesn't reflect their positions in `d`. | 597 |
https://github.com/huggingface/datasets/issues/595 | `Dataset`/`DatasetDict` has no attribute 'save_to_disk' | [
"`pip install git+https://github.com/huggingface/nlp.git` should have done the job.\r\n\r\nDid you uninstall `nlp` before installing from github ?",
"> Did you uninstall `nlp` before installing from github ?\r\n\r\nI did not. I created a new environment and installed `nlp` directly from `github` and it worked!\r\n\r\nThanks.\r\n"
] | Hi,
As the title indicates, both `Dataset` and `DatasetDict` classes don't seem to have the `save_to_disk` method. While the file [`arrow_dataset.py`](https://github.com/huggingface/nlp/blob/34bf0b03bfe03e7f77b8fec1cd48f5452c4fc7c1/src/nlp/arrow_dataset.py) in the repo here has the method, the file `arrow_dataset.py` which is saved after `pip install nlp -U` in my `conda` environment DOES NOT contain the `save_to_disk` method. I even tried `pip install git+https://github.com/huggingface/nlp.git ` and still no luck. Do I need to install the library in another way? | 595 |
https://github.com/huggingface/datasets/issues/590 | The process cannot access the file because it is being used by another process (windows) | [
"Hi, which version of `nlp` are you using?\r\n\r\nBy the way we'll be releasing today a significant update fixing many issues (but also comprising a few breaking changes).\r\nYou can see more informations here #545 and try it by installing from source from the master branch.",
"I'm using version 0.4.0.\r\n\r\n",
"Ok, it's probably fixed on master. Otherwise if you can give me a fully self-contained exemple to reproduce the error, I can try to investigate.",
"I get the same behavior, on Windows, when `map`ping a function to a loaded dataset. \r\nThe error doesn't occur if I re-run the cell a second time though! \r\nI'm on version 1.0.1.",
"This is going to be fixed by #644 ",
"@saareliad I got the same issue that troubled me quite a while. Unfortunately, there are no good answers to this issue online, I tried it on Linux and that's absolutely fine. After hacking the source code, I solved this problem as follows.\r\n\r\nIn the source code file: arrow_dataset.py -> _map_single(...)\r\n\r\nchange\r\n```python\r\nif update_data and tmp_file is not None:\r\n shutil.move(tmp_file.name, cache_file_name)\r\n```\r\nto\r\n```python\r\ntmp_file.close()\r\nif update_data and tmp_file is not None:\r\n shutil.move(tmp_file.name, cache_file_name)\r\n```\r\n\r\nThen it works without needing multiple times runs to avoid the permission error.\r\nI know this solution is unusual since it changes the source code. Hopefully, the lib's contributors can have better solutions in the future.\r\n",
"@wangcongcong123 thanks for sharing.\n(BTW I also solved it locally on windows by putting the problematic line under try except and not using cache... On windows I just needed 1% of the dataset anyway)"
] | Hi, I consistently get the following error when developing in my PC (windows 10):
```
train_dataset = train_dataset.map(convert_to_features, batched=True)
File "C:\Users\saareliad\AppData\Local\Continuum\miniconda3\envs\py38\lib\site-packages\nlp\arrow_dataset.py", line 970, in map
shutil.move(tmp_file.name, cache_file_name)
File "C:\Users\saareliad\AppData\Local\Continuum\miniconda3\envs\py38\lib\shutil.py", line 803, in move
os.unlink(src)
PermissionError: [WinError 32] The process cannot access the file because it is being used by another process: 'C:\\Users\\saareliad\\.cache\\huggingface\\datasets\\squad\\plain_text\\1.0.0\\408a8fa46a1e2805445b793f1022e743428ca739a34809fce872f0c7f17b44ab\\tmpsau1bep1'
``` | 590 |
https://github.com/huggingface/datasets/issues/589 | Cannot use nlp.load_dataset text, AttributeError: module 'nlp.utils' has no attribute 'logging' | [] |
```
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/root/anaconda3/envs/pytorch/lib/python3.7/site-packages/nlp/load.py", line 533, in load_dataset
builder_cls = import_main_class(module_path, dataset=True)
File "/root/anaconda3/envs/pytorch/lib/python3.7/site-packages/nlp/load.py", line 61, in import_main_class
module = importlib.import_module(module_path)
File "/root/anaconda3/envs/pytorch/lib/python3.7/importlib/__init__.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "<frozen importlib._bootstrap>", line 1006, in _gcd_import
File "<frozen importlib._bootstrap>", line 983, in _find_and_load
File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 677, in _load_unlocked
File "<frozen importlib._bootstrap_external>", line 728, in exec_module
File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
File "/root/anaconda3/envs/pytorch/lib/python3.7/site-packages/nlp/datasets/text/5dc629379536c4037d9c2063e1caa829a1676cf795f8e030cd90a537eba20c08/text.py", line 9, in <module>
logger = nlp.utils.logging.get_logger(__name__)
AttributeError: module 'nlp.utils' has no attribute 'logging'
```
Occurs on the following code, or any code including the load_dataset('text'):
```
dataset = load_dataset("text", data_files=file_path, split="train")
dataset = dataset.map(lambda ex: tokenizer(ex["text"], add_special_tokens=True,
truncation=True, max_length=args.block_size), batched=True)
dataset.set_format(type='torch', columns=['input_ids'])
return dataset
``` | 589 |
https://github.com/huggingface/datasets/issues/583 | ArrowIndexError on Dataset.select | [] | If the indices table consists in several chunks, then `dataset.select` results in an `ArrowIndexError` error for pyarrow < 1.0.0
Example:
```python
from nlp import load_dataset
mnli = load_dataset("glue", "mnli", split="train")
shuffled = mnli.shuffle(seed=42)
mnli.select(list(range(len(mnli))))
```
raises:
```python
---------------------------------------------------------------------------
ArrowIndexError Traceback (most recent call last)
<ipython-input-64-006a5d38d418> in <module>
----> 1 mnli.shuffle(seed=42).select(list(range(len(mnli))))
~/Desktop/hf/nlp/src/nlp/fingerprint.py in wrapper(*args, **kwargs)
161 # Call actual function
162
--> 163 out = func(self, *args, **kwargs)
164
165 # Update fingerprint of in-place transforms + update in-place history of transforms
~/Desktop/hf/nlp/src/nlp/arrow_dataset.py in select(self, indices, keep_in_memory, indices_cache_file_name, writer_batch_size, new_fingerprint)
1653 if self._indices is not None:
1654 if PYARROW_V0:
-> 1655 indices_array = self._indices.column(0).chunk(0).take(indices_array)
1656 else:
1657 indices_array = self._indices.column(0).take(indices_array)
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/array.pxi in pyarrow.lib.Array.take()
~/.virtualenvs/hf-datasets/lib/python3.7/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowIndexError: take index out of bounds
```
This is because the `take` method is only done on the first chunk which only contains 1000 elements by default (mnli has ~400 000 elements).
Shall we change that to use
```python
pa.concat_tables(self._indices._indices.slice(i, 1) for i in indices_array)
```
instead of `take` ? @thomwolf | 583 |
https://github.com/huggingface/datasets/issues/582 | Allow for PathLike objects | [] | Using PathLike objects as input for `load_dataset` does not seem to work. The following will throw an error.
```python
files = list(Path(r"D:\corpora\yourcorpus").glob("*.txt"))
dataset = load_dataset("text", data_files=files)
```
Traceback:
```
Traceback (most recent call last):
File "C:/dev/python/dutch-simplification/main.py", line 7, in <module>
dataset = load_dataset("text", data_files=files)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\load.py", line 548, in load_dataset
builder_instance.download_and_prepare(
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 470, in download_and_prepare
self._save_info()
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 564, in _save_info
self.info.write_to_directory(self._cache_dir)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\info.py", line 149, in write_to_directory
self._dump_info(f)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\info.py", line 156, in _dump_info
file.write(json.dumps(asdict(self)).encode("utf-8"))
File "c:\users\bramv\appdata\local\programs\python\python38\lib\json\__init__.py", line 231, in dumps
return _default_encoder.encode(obj)
File "c:\users\bramv\appdata\local\programs\python\python38\lib\json\encoder.py", line 199, in encode
chunks = self.iterencode(o, _one_shot=True)
File "c:\users\bramv\appdata\local\programs\python\python38\lib\json\encoder.py", line 257, in iterencode
return _iterencode(o, 0)
TypeError: keys must be str, int, float, bool or None, not WindowsPath
```
We have to cast to a string explicitly to make this work. It would be nicer if we could actually use PathLike objects.
```python
files = [str(f) for f in Path(r"D:\corpora\wablieft").glob("*.txt")]
```
| 582 |
https://github.com/huggingface/datasets/issues/581 | Better error message when input file does not exist | [] | In the following scenario, when `data_files` is an empty list, the stack trace and error message could be improved. This can probably be solved by checking for each file whether it actually exists and/or whether the argument is not false-y.
```python
dataset = load_dataset("text", data_files=[])
```
Example error trace.
```
Using custom data configuration default
Downloading and preparing dataset text/default-d18f9b6611eb8e16 (download: Unknown size, generated: Unknown size, post-processed: Unknown sizetotal: Unknown size) to C:\Users\bramv\.cache\huggingface\datasets\text\default-d18f9b6611eb8e16\0.0.0\3a79870d85f1982d6a2af884fde86a71c771747b4b161fd302d28ad22adf985b...
Traceback (most recent call last):
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 424, in incomplete_dir
yield tmp_dir
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 462, in download_and_prepare
self._download_and_prepare(
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 537, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 813, in _prepare_split
num_examples, num_bytes = writer.finalize()
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\arrow_writer.py", line 217, in finalize
self.pa_writer.close()
AttributeError: 'NoneType' object has no attribute 'close'
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "C:/dev/python/dutch-simplification/main.py", line 7, in <module>
dataset = load_dataset("text", data_files=files)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\load.py", line 548, in load_dataset
builder_instance.download_and_prepare(
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 470, in download_and_prepare
self._save_info()
File "c:\users\bramv\appdata\local\programs\python\python38\lib\contextlib.py", line 131, in __exit__
self.gen.throw(type, value, traceback)
File "C:\Users\bramv\.virtualenvs\dutch-simplification-nbNdqK9u\lib\site-packages\nlp\builder.py", line 430, in incomplete_dir
shutil.rmtree(tmp_dir)
File "c:\users\bramv\appdata\local\programs\python\python38\lib\shutil.py", line 737, in rmtree
return _rmtree_unsafe(path, onerror)
File "c:\users\bramv\appdata\local\programs\python\python38\lib\shutil.py", line 615, in _rmtree_unsafe
onerror(os.unlink, fullname, sys.exc_info())
File "c:\users\bramv\appdata\local\programs\python\python38\lib\shutil.py", line 613, in _rmtree_unsafe
os.unlink(fullname)
PermissionError: [WinError 32] The process cannot access the file because it is being used by another process: 'C:\\Users\\bramv\\.cache\\huggingface\\datasets\\text\\default-d18f9b6611eb8e16\\0.0.0\\3a79870d85f1982d6a2af884fde86a71c771747b4b161fd302d28ad22adf985b.incomplete\\text-train.arrow'
``` | 581 |
https://github.com/huggingface/datasets/issues/580 | nlp re-creates already-there caches when using a script, but not within a shell | [
"Couln't reproduce on my side :/ \r\nlet me know if you manage to reproduce on another env (colab for example)",
"Fixed with a clean re-install!"
] | `nlp` keeps creating new caches for the same file when launching `filter` from a script, and behaves correctly from within the shell.
Example: try running
```
import nlp
hans_easy_data = nlp.load_dataset('hans', split="validation").filter(lambda x: x['label'] == 0)
hans_hard_data = nlp.load_dataset('hans', split="validation").filter(lambda x: x['label'] == 1)
```
twice. If launched from a `file.py` script, the cache will be re-created the second time. If launched as 3 shell/`ipython` commands, `nlp` will correctly re-use the cache.
As observed with @lhoestq. | 580 |
https://github.com/huggingface/datasets/issues/577 | Some languages in wikipedia dataset are not loading | [
"Some wikipedia languages have already been processed by us and are hosted on our google storage. This is the case for \"fr\" and \"en\" for example.\r\n\r\nFor other smaller languages (in terms of bytes), they are directly downloaded and parsed from the wikipedia dump site.\r\nParsing can take some time for languages with hundreds of MB of xml.\r\n\r\nLet me know if you encounter an error or if you feel that is is taking too long for you.\r\nWe could process those that really take too much time",
"Ok, thanks for clarifying, that makes sense. I will time those examples later today and post back here.\r\n\r\nAlso, it seems that not all dumps should use the same date. For instance, I was checking the Spanish dump doing the following:\r\n```\r\ndata = nlp.load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner', split='train')\r\n```\r\n\r\nI got the error below because this URL does not exist: https://dumps.wikimedia.org/eswiki/20200501/dumpstatus.json. So I checked the actual available dates here https://dumps.wikimedia.org/eswiki/ and there is no 20200501. If one tries for a date available in the link, then the nlp library does not allow such a request because is not in the list of expected datasets.\r\n\r\n```\r\nDownloading and preparing dataset wikipedia/20200501.es (download: Unknown size, generated: Unknown size, post-processed: Unknown sizetotal: Unknown size) to /home/gaguilar/.cache/huggingface/datasets/wikipedia/20200501.es/1.0.0/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50...\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/load.py\", line 548, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/builder.py\", line 462, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/builder.py\", line 965, in _download_and_prepare\r\n super(BeamBasedBuilder, self)._download_and_prepare(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/builder.py\", line 518, in _download_and_prepare\r\n split_generators = self._split_generators(dl_manager, **split_generators_kwargs)\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/datasets/wikipedia/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50/wikipedia.py\", line 422, in _split_generators\r\n downloaded_files = dl_manager.download_and_extract({\"info\": info_url})\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/download_manager.py\", line 220, in download_and_extract\r\n return self.extract(self.download(url_or_urls))\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/download_manager.py\", line 155, in download\r\n downloaded_path_or_paths = map_nested(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/py_utils.py\", line 163, in map_nested\r\n return {\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/py_utils.py\", line 164, in <dictcomp>\r\n k: map_nested(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/py_utils.py\", line 191, in map_nested\r\n return function(data_struct)\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/download_manager.py\", line 156, in <lambda>\r\n lambda url: cached_path(url, download_config=self._download_config,), url_or_urls,\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/file_utils.py\", line 191, in cached_path\r\n output_path = get_from_cache(\r\n File \"/home/gaguilar/.conda/envs/pytorch/lib/python3.8/site-packages/nlp/utils/file_utils.py\", line 356, in get_from_cache\r\n raise ConnectionError(\"Couldn't reach {}\".format(url))\r\nConnectionError: Couldn't reach https://dumps.wikimedia.org/eswiki/20200501/dumpstatus.json\r\n```",
"Thanks ! This will be very helpful.\r\n\r\nAbout the date issue, I think it's possible to use another date with\r\n\r\n```python\r\nload_dataset(\"wikipedia\", language=\"es\", date=\"...\", beam_runner=\"...\")\r\n```\r\n\r\nHowever we've not processed wikipedia dumps for other dates than 20200501 (yet ?)\r\n\r\nOne more thing that is specific to 20200501.es: it was available once but the `mwparserfromhell` was not able to parse it for some reason, so we didn't manage to get a processed version of 20200501.es (see #321 )",
"Cool! Thanks for the trick regarding different dates!\r\n\r\nI checked the download/processing time for retrieving the Arabic Wikipedia dump, and it took about 3.2 hours. I think that this may be a bit impractical when it comes to working with multiple languages (although I understand that storing those datasets in your Google storage may not be very appealing either). \r\n\r\nFor the record, here's what I did:\r\n```python\r\nimport nlp\r\nimport time\r\n\r\ndef timeit(filename):\r\n elapsed = time.time()\r\n data = nlp.load_dataset('wikipedia', filename, beam_runner='DirectRunner', split='train')\r\n elapsed = time.time() - elapsed\r\n print(f\"Loading the '{filename}' data took {elapsed:,.1f} seconds...\")\r\n return data\r\n\r\ndata = timeit('20200501.ar')\r\n```\r\n\r\nHere's the output:\r\n```\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 13.0k/13.0k [00:00<00:00, 8.34MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 28.7k/28.7k [00:00<00:00, 954kB/s]\r\nDownloading and preparing dataset wikipedia/20200501.ar (download: Unknown size, generated: Unknown size, post-processed: Unknown sizetotal: Unknown size) to /home/gaguil20/.cache/huggingface/datasets/wikipedia/20200501.ar/1.0.0/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50...\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 47.4k/47.4k [00:00<00:00, 1.40MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 79.8M/79.8M [00:15<00:00, 5.13MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 171M/171M [00:33<00:00, 5.13MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 103M/103M [00:20<00:00, 5.14MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 227M/227M [00:44<00:00, 5.06MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 140M/140M [00:28<00:00, 4.96MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 160M/160M [00:30<00:00, 5.20MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 97.5M/97.5M [00:19<00:00, 5.06MB/s]\r\nDownloading: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 222M/222M [00:42<00:00, 5.21MB/s]\r\n100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1/1 [03:16<00:00, 196.39s/sources]\r\nDataset wikipedia downloaded and prepared to /home/gaguil20/.cache/huggingface/datasets/wikipedia/20200501.ar/1.0.0/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50. Subsequent calls will reuse this data.\r\nLoading the '20200501.ar' data took 11,582.7 seconds...\r\n````",
"> About the date issue, I think it's possible to use another date with\r\n> ```python\r\n> load_dataset(\"wikipedia\", language=\"es\", date=\"...\", beam_runner=\"...\")\r\n> ```\r\n\r\nI tried your suggestion about the date and the function does not accept the language and date keywords. I tried both on `nlp` v0.4 and the new `datasets` library (v1.0.2):\r\n```\r\nload_dataset(\"wikipedia\", language=\"es\", date=\"20200601\", beam_runner='DirectRunner', split='train')\r\n```\r\nFor now, my quick workaround to keep things moving was to simply change the date inside the library at this line: [https://github.com/huggingface/datasets/blob/master/datasets/wikipedia/wikipedia.py#L403](https://github.com/huggingface/datasets/blob/master/datasets/wikipedia/wikipedia.py#L403)\r\n\r\nNote that the date and languages are valid: [https://dumps.wikimedia.org/eswiki/20200601/dumpstatus.json](https://dumps.wikimedia.org/eswiki/20200601/dumpstatus.json)\r\n\r\nAny suggestion is welcome :) @lhoestq \r\n\r\n\r\n## **[UPDATE]**\r\n\r\nThe workaround I mentioned fetched the data, but then I faced another issue (even the log says to report this as bug):\r\n```\r\nERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.\r\n```\r\n\r\nHere's the full stack (which says that there is a key error caused by this key: `KeyError: '000nbsp'`):\r\n\r\n```Downloading and preparing dataset wikipedia/20200601.es (download: Unknown size, generated: Unknown size, post-processed: Unknown sizetotal: Unknown size) to /home/gustavoag/.cache/huggingface/datasets/wikipedia/20200601.es/1.0.0/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50...\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 74.7k/74.7k [00:00<00:00, 1.53MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 232M/232M [00:48<00:00, 4.75MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 442M/442M [01:39<00:00, 4.44MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 173M/173M [00:33<00:00, 5.12MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 344M/344M [01:14<00:00, 4.59MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 541M/541M [01:59<00:00, 4.52MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 476M/476M [01:31<00:00, 5.18MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 545M/545M [02:02<00:00, 4.46MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 299M/299M [01:01<00:00, 4.89MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 9.60M/9.60M [00:01<00:00, 4.84MB/s]\r\nDownloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 423M/423M [01:36<00:00, 4.38MB/s]\r\nWARNING:apache_beam.options.pipeline_options:Discarding unparseable args: ['--lang', 'es', '--date', '20200601', '--tokenizer', 'bert-base-multilingual-cased', '--cache', 'train', 'valid', '--max_dataset_length', '200000', '10000']\r\n\r\nERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.\r\nERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.\r\nERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.\r\nERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.\r\nTraceback (most recent call last):\r\n File \"apache_beam/runners/common.py\", line 961, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 553, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1095, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/datasets/wikipedia/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50/wikipedia.py\", line 500, in _clean_content\r\n text = _parse_and_clean_wikicode(raw_content, parser=mwparserfromhell)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/datasets/wikipedia/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50/wikipedia.py\", line 556, in _parse_and_clean_wikicode\r\n section_text.append(section.strip_code().strip())\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/wikicode.py\", line 643, in strip_code\r\n stripped = node.__strip__(**kwargs)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/nodes/html_entity.py\", line 63, in __strip__\r\n return self.normalize()\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/nodes/html_entity.py\", line 178, in normalize\r\n return chrfunc(htmlentities.name2codepoint[self.value])\r\nKeyError: '000nbsp'\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/runpy.py\", line 194, in _run_module_as_main\r\n return _run_code(code, main_globals, None,\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/runpy.py\", line 87, in _run_code\r\n exec(code, run_globals)\r\n File \"/raid/data/gustavoag/projects/char2subword/research/preprocessing/split_wiki.py\", line 96, in <module>\r\n main()\r\n File \"/raid/data/gustavoag/projects/char2subword/research/preprocessing/split_wiki.py\", line 65, in main\r\n data = nlp.load_dataset('wikipedia', f'{args.date}.{args.lang}', beam_runner='DirectRunner', split='train')\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/load.py\", line 548, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/builder.py\", line 462, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/builder.py\", line 969, in _download_and_prepare\r\n pipeline_results = pipeline.run()\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/pipeline.py\", line 534, in run\r\n return self.runner.run_pipeline(self, self._options)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/direct/direct_runner.py\", line 119, in run_pipeline\r\n return runner.run_pipeline(pipeline, options)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 172, in run_pipeline\r\n self._latest_run_result = self.run_via_runner_api(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 183, in run_via_runner_api\r\n return self.run_stages(stage_context, stages)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 338, in run_stages\r\n stage_results = self._run_stage(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 512, in _run_stage\r\n last_result, deferred_inputs, fired_timers = self._run_bundle(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 556, in _run_bundle\r\n result, splits = bundle_manager.process_bundle(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 940, in process_bundle\r\n for result, split_result in executor.map(execute, zip(part_inputs, # pylint: disable=zip-builtin-not-iterating\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/concurrent/futures/_base.py\", line 611, in result_iterator\r\n yield fs.pop().result()\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/concurrent/futures/_base.py\", line 439, in result\r\n return self.__get_result()\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/concurrent/futures/_base.py\", line 388, in __get_result\r\n raise self._exception\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/utils/thread_pool_executor.py\", line 44, in run\r\n self._future.set_result(self._fn(*self._fn_args, **self._fn_kwargs))\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 932, in execute\r\n return bundle_manager.process_bundle(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 837, in process_bundle\r\n result_future = self._worker_handler.control_conn.push(process_bundle_req)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/worker_handlers.py\", line 352, in push\r\n response = self.worker.do_instruction(request)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/worker/sdk_worker.py\", line 479, in do_instruction\r\n return getattr(self, request_type)(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/worker/sdk_worker.py\", line 515, in process_bundle\r\n bundle_processor.process_bundle(instruction_id))\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/worker/bundle_processor.py\", line 977, in process_bundle\r\n input_op_by_transform_id[element.transform_id].process_encoded(\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/apache_beam/runners/worker/bundle_processor.py\", line 218, in process_encoded\r\n self.output(decoded_value)\r\n File \"apache_beam/runners/worker/operations.py\", line 330, in apache_beam.runners.worker.operations.Operation.output\r\n File \"apache_beam/runners/worker/operations.py\", line 332, in apache_beam.runners.worker.operations.Operation.output\r\n File \"apache_beam/runners/worker/operations.py\", line 195, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 670, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 671, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 963, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1030, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"apache_beam/runners/common.py\", line 961, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 553, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1122, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"apache_beam/runners/worker/operations.py\", line 195, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 670, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 671, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 963, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1030, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"apache_beam/runners/common.py\", line 961, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 553, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1122, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"apache_beam/runners/worker/operations.py\", line 195, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 670, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 671, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 963, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1045, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/future/utils/__init__.py\", line 446, in raise_with_traceback\r\n raise exc.with_traceback(traceback)\r\n File \"apache_beam/runners/common.py\", line 961, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 553, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1095, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/datasets/wikipedia/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50/wikipedia.py\", line 500, in _clean_content\r\n text = _parse_and_clean_wikicode(raw_content, parser=mwparserfromhell)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/nlp/datasets/wikipedia/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50/wikipedia.py\", line 556, in _parse_and_clean_wikicode\r\n section_text.append(section.strip_code().strip())\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/wikicode.py\", line 643, in strip_code\r\n stripped = node.__strip__(**kwargs)\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/nodes/html_entity.py\", line 63, in __strip__\r\n return self.normalize()\r\n File \"/home/gustavoag/anaconda3/envs/pytorch/lib/python3.8/site-packages/mwparserfromhell/nodes/html_entity.py\", line 178, in normalize\r\n return chrfunc(htmlentities.name2codepoint[self.value])\r\nKeyError: \"000nbsp [while running 'train/Clean content']\"```",
"@lhoestq Any updates on this? I have similar issues with the Romanian dump, tnx.",
"Hey @gaguilar ,\r\n\r\nI just found the [\"char2subword\" paper](https://arxiv.org/pdf/2010.12730.pdf) and I'm really interested in trying it out on own vocabs/datasets like for historical texts (I've already [trained some lms](https://github.com/stefan-it/europeana-bert) on newspaper articles with OCR errors).\r\n\r\nDo you plan to release the code for your paper or is it possible to get the implementation π€ Many thanks :hugs: ",
"Hi @stefan-it! Thanks for your interest in our work! We do plan to release the code, but we will make it available once the paper has been published at a conference. Sorry for the inconvenience!\r\n\r\nHi @lhoestq, do you have any insights for this issue by any chance? Thanks!",
"This is an issue on the `mwparserfromhell` side. You could try to update `mwparserfromhell` and see if it fixes the issue. If it doesn't we'll have to create an issue on their repo for them to fix it.\r\nBut first let's see if the latest version of `mwparserfromhell` does the job.",
"I think the work around as suggested in the issue [#886] is not working for several languages, such as `id`. For example, I tried all the dates to download dataset for `id` langauge from the following link: (https://github.com/huggingface/datasets/pull/886) [https://dumps.wikimedia.org/idwiki/](https://dumps.wikimedia.org/idwiki/ )\r\n\r\n> >>> dataset = load_dataset('wikipedia', language='id', date=\"20210501\", beam_runner='DirectRunner')\r\nWARNING:datasets.builder:Using custom data configuration 20210501.id-date=20210501,language=id\r\nDownloading and preparing dataset wikipedia/20210501.id (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /Users/.cache/huggingface/datasets/wikipedia/20210501.id-date=20210501,language=id/0.0.0/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1...\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/load.py\", line 745, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/builder.py\", line 574, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/builder.py\", line 1139, in _download_and_prepare\r\n super(BeamBasedBuilder, self)._download_and_prepare(\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/builder.py\", line 630, in _download_and_prepare\r\n split_generators = self._split_generators(dl_manager, **split_generators_kwargs)\r\n File \"/Users/.cache/huggingface/modules/datasets_modules/datasets/wikipedia/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1/wikipedia.py\", line 420, in _split_generators\r\n downloaded_files = dl_manager.download_and_extract({\"info\": info_url})\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/download_manager.py\", line 287, in download_and_extract\r\n return self.extract(self.download(url_or_urls))\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/download_manager.py\", line 195, in download\r\n downloaded_path_or_paths = map_nested(\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/py_utils.py\", line 203, in map_nested\r\n mapped = [\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/py_utils.py\", line 204, in <listcomp>\r\n _single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/py_utils.py\", line 142, in _single_map_nested\r\n return function(data_struct)\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/download_manager.py\", line 218, in _download\r\n return cached_path(url_or_filename, download_config=download_config)\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/file_utils.py\", line 281, in cached_path\r\n output_path = get_from_cache(\r\n File \"/Users/opt/anaconda3/envs/proj/lib/python3.9/site-packages/datasets/utils/file_utils.py\", line 623, in get_from_cache\r\n raise ConnectionError(\"Couldn't reach {}\".format(url))\r\nConnectionError: Couldn't reach https://dumps.wikimedia.org/idwiki/20210501/dumpstatus.json\r\n\r\nMoreover the downloading speed for `non-en` language is very very slow. And interestingly the download stopped after approx a couple minutes due to the read time-out. I tried numerous times and the results is same. Is there any feasible way to download non-en language using huggingface?\r\n\r\n> File \"/Users/miislamg/opt/anaconda3/envs/proj-semlm/lib/python3.9/site-packages/requests/models.py\", line 760, in generate\r\n raise ConnectionError(e)\r\nrequests.exceptions.ConnectionError: HTTPSConnectionPool(host='dumps.wikimedia.org', port=443): Read timed out.\r\nDownloading: 7%|βββββββββ | 10.2M/153M [03:35<50:07, 47.4kB/s]",
"Hi ! The link https://dumps.wikimedia.org/idwiki/20210501/dumpstatus.json seems to be working fine for me.\r\n\r\nRegarding the time outs, it must come either from an issue on the wikimedia host side, or from your internet connection.\r\nFeel free to try again several times.",
"I was trying to download dataset for `es` language, however I am getting the following error:\r\n```\r\ndataset = load_dataset('wikipedia', language='es', date=\"20210320\", beam_runner='DirectRunner') \r\n```\r\n\r\n```\r\nDownloading and preparing dataset wikipedia/20210320.es (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /scratch/user_name/datasets/wikipedia/20210320.es-date=20210320,language=es/0.0.0/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1...\r\nTraceback (most recent call last):\r\n File \"apache_beam/runners/common.py\", line 1233, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 581, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1368, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"/scratch/user_name/modules/datasets_modules/datasets/wikipedia/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1/wikipedia.py\", line 492, in _clean_content\r\n text = _parse_and_clean_wikicode(raw_content, parser=mwparserfromhell)\r\n File \"/scratch/user_name/modules/datasets_modules/datasets/wikipedia/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1/wikipedia.py\", line 548, in _parse_and_clean_wikicode\r\n section_text.append(section.strip_code().strip())\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/wikicode.py\", line 639, in strip_code\r\n stripped = node.__strip__(**kwargs)\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/nodes/html_entity.py\", line 60, in __strip__\r\n return self.normalize()\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/nodes/html_entity.py\", line 150, in normalize\r\n return chr(htmlentities.name2codepoint[self.value])\r\nKeyError: '000nbsp'\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nTraceback (most recent call last):\r\n File \"download_dataset_all.py\", line 8, in <module>\r\n dataset = load_dataset('wikipedia', language=language, date=\"20210320\", beam_runner='DirectRunner') \r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/load.py\", line 748, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/builder.py\", line 575, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/opt/conda/lib/python3.7/site-packages/datasets/builder.py\", line 1152, in _download_and_prepare\r\n pipeline_results = pipeline.run()\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/pipeline.py\", line 564, in run\r\n return self.runner.run_pipeline(self, self._options)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/direct/direct_runner.py\", line 131, in run_pipeline\r\n return runner.run_pipeline(pipeline, options)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 190, in run_pipeline\r\n pipeline.to_runner_api(default_environment=self._default_environment))\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 200, in run_via_runner_api\r\n return self.run_stages(stage_context, stages)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 366, in run_stages\r\n bundle_context_manager,\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 562, in _run_stage\r\n bundle_manager)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 602, in _run_bundle\r\n data_input, data_output, input_timers, expected_timer_output)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py\", line 903, in process_bundle\r\n result_future = self._worker_handler.control_conn.push(process_bundle_req)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/portability/fn_api_runner/worker_handlers.py\", line 378, in push\r\n response = self.worker.do_instruction(request)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py\", line 610, in do_instruction\r\n getattr(request, request_type), request.instruction_id)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py\", line 647, in process_bundle\r\n bundle_processor.process_bundle(instruction_id))\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/worker/bundle_processor.py\", line 1001, in process_bundle\r\n element.data)\r\n File \"/opt/conda/lib/python3.7/site-packages/apache_beam/runners/worker/bundle_processor.py\", line 229, in process_encoded\r\n self.output(decoded_value)\r\n File \"apache_beam/runners/worker/operations.py\", line 356, in apache_beam.runners.worker.operations.Operation.output\r\n File \"apache_beam/runners/worker/operations.py\", line 358, in apache_beam.runners.worker.operations.Operation.output\r\n File \"apache_beam/runners/worker/operations.py\", line 220, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 717, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 718, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 1235, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1300, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"apache_beam/runners/common.py\", line 1233, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 581, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1395, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"apache_beam/runners/worker/operations.py\", line 220, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 717, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 718, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 1235, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1300, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"apache_beam/runners/common.py\", line 1233, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 581, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1395, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"apache_beam/runners/worker/operations.py\", line 220, in apache_beam.runners.worker.operations.SingletonConsumerSet.receive\r\n File \"apache_beam/runners/worker/operations.py\", line 717, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/worker/operations.py\", line 718, in apache_beam.runners.worker.operations.DoOperation.process\r\n File \"apache_beam/runners/common.py\", line 1235, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 1315, in apache_beam.runners.common.DoFnRunner._reraise_augmented\r\n File \"/opt/conda/lib/python3.7/site-packages/future/utils/__init__.py\", line 446, in raise_with_traceback\r\n raise exc.with_traceback(traceback)\r\n File \"apache_beam/runners/common.py\", line 1233, in apache_beam.runners.common.DoFnRunner.process\r\n File \"apache_beam/runners/common.py\", line 581, in apache_beam.runners.common.SimpleInvoker.invoke_process\r\n File \"apache_beam/runners/common.py\", line 1368, in apache_beam.runners.common._OutputProcessor.process_outputs\r\n File \"/scratch/user_name/modules/datasets_modules/datasets/wikipedia/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1/wikipedia.py\", line 492, in _clean_content\r\n text = _parse_and_clean_wikicode(raw_content, parser=mwparserfromhell)\r\n File \"/scratch/user_name/modules/datasets_modules/datasets/wikipedia/2fe8db1405aef67dff9fcc51e133e1f9c5b0106f9d9e9638188176d278fd5ff1/wikipedia.py\", line 548, in _parse_and_clean_wikicode\r\n section_text.append(section.strip_code().strip())\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/wikicode.py\", line 639, in strip_code\r\n stripped = node.__strip__(**kwargs)\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/nodes/html_entity.py\", line 60, in __strip__\r\n return self.normalize()\r\n File \"/opt/conda/lib/python3.7/site-packages/mwparserfromhell/nodes/html_entity.py\", line 150, in normalize\r\n return chr(htmlentities.name2codepoint[self.value])\r\nKeyError: \"000nbsp [while running 'train/Clean content']\"\r\n```",
"Hi ! This looks related to this issue: https://github.com/huggingface/datasets/issues/1994\r\nBasically the parser that is used (mwparserfromhell) has some issues for some pages in `es`.\r\nWe already reported some issues for `es` on their repo at https://github.com/earwig/mwparserfromhell/issues/247 but it looks like there are still a few issues. Might be a good idea to open a new issue on the mwparserfromhell repo",
"Any updates on this so far?",
"The issue:\r\n```\r\nKeyError: \"000nbsp [while running 'train/Clean content']\"\r\n```\r\nreported in comments:\r\n- https://github.com/huggingface/datasets/issues/577#issuecomment-701890059 (by @gaguilar)\r\n- https://github.com/huggingface/datasets/issues/577#issuecomment-879513227 (by @mmiakashs)\r\n\r\nwas normally fixed in the `mwparserfromhell` library and will be accessible in their next release version `0.7`:\r\n- https://github.com/earwig/mwparserfromhell/issues/288",
"mwparserfromhell 0.7 has still not been released, but you might have luck with the dev version:\r\n`pip install git+https://github.com/earwig/mwparserfromhell.git@0f89f44`"
] | Hi,
I am working with the `wikipedia` dataset and I have a script that goes over 92 of the available languages in that dataset. So far I have detected that `ar`, `af`, `an` are not loading. Other languages like `fr` and `en` are working fine. Here's how I am loading them:
```
import nlp
langs = ['ar'. 'af', 'an']
for lang in langs:
data = nlp.load_dataset('wikipedia', f'20200501.{lang}', beam_runner='DirectRunner', split='train')
print(lang, len(data))
```
Here's what I see for 'ar' (it gets stuck there):
```
Downloading and preparing dataset wikipedia/20200501.ar (download: Unknown size, generated: Unknown size, post-processed: Unknown sizetotal: Unknown size) to /home/gaguilar/.cache/huggingface/datasets/wikipedia/20200501.ar/1.0.0/7be7f4324255faf70687be8692de57cf79197afdc33ff08d6a04ed602df32d50...
```
Note that those languages are indeed in the list of expected languages. Any suggestions on how to work around this? Thanks! | 577 |
https://github.com/huggingface/datasets/issues/575 | Couldn't reach certain URLs and for the ones that can be reached, code just blocks after downloading. | [
"Update:\r\n\r\nThe imdb download completed after a long time (about 45 mins). Ofcourse once download loading was instantaneous. Also, the loaded object was of type `arrow_dataset`. \r\n\r\nThe urls for glue still doesn't work though.",
"Thanks for the report, I'll give a look!",
"I am also seeing a similar error when running the following:\r\n\r\n```\r\nimport nlp\r\ndataset = load_dataset('cola')\r\n```\r\nError:\r\n```\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/js11133/.conda/envs/jiant/lib/python3.8/site-packages/nlp/load.py\", line 509, in load_dataset\r\n module_path = prepare_module(path, download_config=download_config, dataset=True)\r\n File \"/home/js11133/.conda/envs/jiant/lib/python3.8/site-packages/nlp/load.py\", line 248, in prepare_module\r\n local_path = cached_path(file_path, download_config=download_config)\r\n File \"/home/js11133/.conda/envs/jiant/lib/python3.8/site-packages/nlp/utils/file_utils.py\", line 191, in cached_path\r\n output_path = get_from_cache(\r\n File \"/home/js11133/.conda/envs/jiant/lib/python3.8/site-packages/nlp/utils/file_utils.py\", line 356, in get_from_cache\r\n raise ConnectionError(\"Couldn't reach {}\".format(url))\r\nConnectionError: Couldn't reach https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cola/cola.py\r\n```",
"@jeswan `\"cola\"` is not a valid dataset identifier (you can check the up-to-date list on https://huggingface.co/datasets) but you can find cola inside glue.",
"Ah right. Thanks!",
"Hi. Closing this one since #626 updated the glue urls.\r\n\r\n> 1. Why is it still blocking? Is it still downloading?\r\n\r\nAfter downloading it generates the arrow file by iterating through the examples.\r\nThe number of examples processed by second is shown during the processing (not sure why it was not the case for you)\r\n\r\n> 2. I specified split as train, so why is the test folder being populated?\r\n\r\nIt downloads every split\r\n\r\n\r\n\r\n"
] | Hi,
I'm following the [quick tour](https://huggingface.co/nlp/quicktour.html) and tried to load the glue dataset:
```
>>> from nlp import load_dataset
>>> dataset = load_dataset('glue', 'mrpc', split='train')
```
However, this ran into a `ConnectionError` saying it could not reach the URL (just pasting the last few lines):
```
/net/vaosl01/opt/NFS/su0/miniconda3/envs/hf/lib/python3.7/site-packages/nlp/utils/file_utils.py in get_from_cache(url, cache_dir, force_download, proxies, etag_timeout, resume_download, user_agent, local_files_only)
354 " to False."
355 )
--> 356 raise ConnectionError("Couldn't reach {}".format(url))
357
358 # From now on, connected is True.
ConnectionError: Couldn't reach https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2Fmrpc_dev_ids.tsv?alt=media&token=ec5c0836-31d5-48f4-b431-7480817f1adc
```
I tried glue with cola and sst2. I got the same error, just instead of mrpc in the URL, it was replaced with cola and sst2.
Since this was not working, I thought I'll try another dataset. So I tried downloading the imdb dataset:
```
ds = load_dataset('imdb', split='train')
```
This downloads the data, but it just blocks after that:
```
Downloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 4.56k/4.56k [00:00<00:00, 1.38MB/s]
Downloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 2.07k/2.07k [00:00<00:00, 1.15MB/s]
Downloading and preparing dataset imdb/plain_text (download: 80.23 MiB, generated: 127.06 MiB, post-processed: Unknown sizetotal: 207.28 MiB) to /net/vaosl01/opt/NFS/su0/huggingface/datasets/imdb/plain_text/1.0.0/76cdbd7249ea3548c928bbf304258dab44d09cd3638d9da8d42480d1d1be3743...
Downloading: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 84.1M/84.1M [00:07<00:00, 11.1MB/s]
```
I checked the folder `$HF_HOME/datasets/downloads/extracted/<id>/aclImdb`. This folder is constantly growing in size. When I navigated to the train folder within, there was no file. However, the test folder seemed to be populating. The last time I checked it was 327M. I thought the Imdb dataset was smaller than that. My questions are:
1. Why is it still blocking? Is it still downloading?
2. I specified split as train, so why is the test folder being populated?
3. I read somewhere that after downloading, `nlp` converts the text files into some sort of `arrow` files, which will also take a while. Is this also happening here?
Thanks.
| 575 |
https://github.com/huggingface/datasets/issues/568 | `metric.compute` throws `ArrowInvalid` error | [
"Hmm might be related to what we are solving in #564",
"Could you try to update to `datasets>=1.0.0` (we changed the name of the library) and try again ?\r\nIf is was related to the distributed setup settings it must be fixed.\r\nIf it was related to empty metric inputs it's going to be fixed in #654 ",
"Closing this one as it was fixed in #654 \r\nFeel free to re-open if you have other questions"
] | I get the following error with `rouge.compute`. It happens only with distributed training, and it occurs randomly I can't easily reproduce it. This is using `nlp==0.4.0`
```
File "/home/beltagy/trainer.py", line 92, in validation_step
rouge_scores = rouge.compute(predictions=generated_str, references=gold_str, rouge_types=['rouge2', 'rouge1', 'rougeL'])
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/nlp/metric.py", line 224, in compute
self.finalize(timeout=timeout)
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/nlp/metric.py", line 213, in finalize
self.data = Dataset(**reader.read_files(node_files))
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/nlp/arrow_reader.py", line 217, in read_files
dataset_kwargs = self._read_files(files=files, info=self._info, original_instructions=original_instructions)
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/nlp/arrow_reader.py", line 162, in _read_files
pa_table: pa.Table = self._get_dataset_from_filename(f_dict)
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/nlp/arrow_reader.py", line 276, in _get_dataset_from_filename
f = pa.ipc.open_stream(mmap)
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/pyarrow/ipc.py", line 173, in open_stream
return RecordBatchStreamReader(source)
File "/home/beltagy/miniconda3/envs/allennlp/lib/python3.7/site-packages/pyarrow/ipc.py", line 64, in __init__
self._open(source)
File "pyarrow/ipc.pxi", line 469, in pyarrow.lib._RecordBatchStreamReader._open
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: Tried reading schema message, was null or length 0
``` | 568 |