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/6775 | IndexError: Invalid key: 0 is out of bounds for size 0 | [
"Same problem.",
"Hi! You should be able to fix this by passing `remove_unused_columns=False` to the `transformers` `TrainingArguments` as explained in https://github.com/huggingface/peft/issues/1299.\r\n\r\n(I'm not familiar with Vertex AI, but I'd assume `remove_unused_columns` can be passed as a flag to the docker container) ",
"I had the same problem, but I spent a whole day trying different combination with my own dataset with the example data set and found the reason: the example data is multi-turn conversation between human and assistant, so # Humman or # Assistant appear at least twice. If your own custom data only has single turn conversation, it might end up with the same error. What you can do is repeat your single turn conversation twice in your training data (keep the key 'text' the same) and maybe it works. I guess the reason is the specific way processing the data requires and counts multi-turn only (single turn will be discarded so it ends up with no training data), but since I am using Google Vertex AI, I don't have direct access to the underlying code so that was just my guess. ",
"> Hi! You should be able to fix this by passing `remove_unused_columns=False` to the `transformers` `TrainingArguments` as explained in [huggingface/peft#1299](https://github.com/huggingface/peft/issues/1299).\r\n> \r\n> (I'm not familiar with Vertex AI, but I'd assume `remove_unused_columns` can be passed as a flag to the docker container)\r\n\r\n@mariosasko Thanks for the response and suggestion. \r\nWhen I set `remove_unused_columns` as `False` , I end up getting different error (will post the error soon). \r\nEither the Vertex-AI does not support `remove_unused_columns` or my dataset is completely wrong. \r\n\r\nThank you, \r\nKK",
"> I had the same problem, but I spent a whole day trying different combination with my own dataset with the example data set and found the reason: the example data is multi-turn conversation between human and assistant, so # Humman or # Assistant appear at least twice. If your own custom data only has single turn conversation, it might end up with the same error. What you can do is repeat your single turn conversation twice in your training data (keep the key 'text' the same) and maybe it works. I guess the reason is the specific way processing the data requires and counts multi-turn only (single turn will be discarded so it ends up with no training data), but since I am using Google Vertex AI, I don't have direct access to the underlying code so that was just my guess.\r\n\r\n@cyberyu Thanks for your suggestions. \r\nI have tried the approach you suggested, copied the same conversation in each jsonl element so every jsonl item has 2 `HUMAN` and `ASSISTANT`. \r\nHowever in my case, the issue persists. I am gonna give few more tries, and post the results here. \r\nYou can find my dataset [here](https://huggingface.co/datasets/kk2491/test/tree/main) \r\n\r\nThank you, \r\nKK ",
"> > I had the same problem, but I spent a whole day trying different combination with my own dataset with the example data set and found the reason: the example data is multi-turn conversation between human and assistant, so # Humman or # Assistant appear at least twice. If your own custom data only has single turn conversation, it might end up with the same error. What you can do is repeat your single turn conversation twice in your training data (keep the key 'text' the same) and maybe it works. I guess the reason is the specific way processing the data requires and counts multi-turn only (single turn will be discarded so it ends up with no training data), but since I am using Google Vertex AI, I don't have direct access to the underlying code so that was just my guess.\r\n> \r\n> @cyberyu Thanks for your suggestions. I have tried the approach you suggested, copied the same conversation in each jsonl element so every jsonl item has 2 `HUMAN` and `ASSISTANT`. However in my case, the issue persists. I am gonna give few more tries, and post the results here. You can find my dataset [here](https://huggingface.co/datasets/kk2491/test/tree/main)\r\n> \r\n> Thank you, KK\r\n\r\nI think another reason is your training sample length is too short. I saw a relevant report (/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Findexerror-invalid-key-16-is-out-of-bounds-for-size-0%2F14298%2F16) stating that the processing code might have a bug discarding sequence length short than max_seq_length, which is 512. Not sure the Vertex AI backend code has fixed that bug or not. So I tried to add some garbage content in your data, and extended the length longer than 512 for a single turn, and repeated twice. You can copy the following line as 5 repeated lines as your training data jsonl file of five samples (no eval or test needed, for speed up, set evaluation step to 5 and training step to 10,), and it will pass.\r\n\r\n{\"text\":\"### Human: You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You will handle customers queries and provide effective help message. Please provide response to 'Can Interplai software optimize routes for minimizing package handling and transfer times in distribution centers'? ### Assistant: Yes, Interplai software can optimize routes for distribution centers by streamlining package handling processes, minimizing transfer times between loading docks and storage areas, and optimizing warehouse layouts for efficient order fulfillment. ### Human: You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You are a helpful AI Assistant familiar with customer service. You will handle customers queries and provide effective help message. Please provide response to 'Can Interplai software optimize routes for minimizing package handling and transfer times in distribution centers'? ### Assistant: Yes, Interplai software can optimize routes for distribution centers by streamlining package handling processes, minimizing transfer times between loading docks and storage areas, and optimizing warehouse layouts for efficient order fulfillment.\"}\r\n",
"@cyberyu **Thank you so much, You saved my day (+ so many days)**. \r\nI tried the example you provided above, and the training is successfully completed in Vertex-AI (through GUI). \r\nI never thought there would be constraints on the length of the samples and also on the number of turns. \r\nI will update my complete dataset and see update here once the training is completed. \r\n\r\nThank you, \r\nKK "
] | ### Describe the bug
I am trying to fine-tune llama2-7b model in GCP. The notebook I am using for this can be found [here](https://github.com/GoogleCloudPlatform/vertex-ai-samples/blob/main/notebooks/community/model_garden/model_garden_pytorch_llama2_peft_finetuning.ipynb).
When I use the dataset given in the example, the training gets successfully completed (example dataset can be found [here](https://huggingface.co/datasets/timdettmers/openassistant-guanaco)).
However when I use my own dataset which is in the same format as the example dataset, I get the below error (my dataset can be found [here](https://huggingface.co/datasets/kk2491/finetune_dataset_002)).
I see the files are being read correctly from the logs:
### Steps to reproduce the bug
1. Clone the [vertex-ai-samples](https://github.com/GoogleCloudPlatform/vertex-ai-samples) repository.
2. Run the [llama2-7b peft fine-tuning](https://github.com/GoogleCloudPlatform/vertex-ai-samples/blob/main/notebooks/community/model_garden/model_garden_pytorch_llama2_peft_finetuning.ipynb).
3. Change the dataset `kk2491/finetune_dataset_002`
### Expected behavior
The training should complete successfully, and model gets deployed to an endpoint.
### Environment info
Python version : Python 3.10.12
Dataset : https://huggingface.co/datasets/kk2491/finetune_dataset_002
| 6,775 |
https://github.com/huggingface/datasets/issues/6774 | Generating split is very slow when Image format is PNG | [
"I think this is due to the speed of reading a `png` image using pillow compared to a `jpg` image.\r\nNotably the same is true with `tiff`, it is even faster than `jpg` in my case."
] | ### Describe the bug
When I create a dataset, it gets stuck while generating cached data.
The image format is PNG, and it will not get stuck when the image format is jpeg.
After debugging, I know that it is because of the `pa.array` operation in [arrow_writer](https://github.com/huggingface/datasets/blob/2.13.0/src/datasets/arrow_writer.py#L553), but i don't why.
### Steps to reproduce the bug
```
from datasets import Dataset
def generator(lines):
for line in lines:
img = Image.open(open(line["url"], "rb"))
# print(img.format) # "PNG"
yield {
"image": img,
}
lines = open(dataset_path, "r")
dataset = Dataset.from_generator(
generator,
gen_kwargs={"lines": lines}
)
```
### Expected behavior
Generating split done.
### Environment info
datasets 2.13.0 | 6,774 |
https://github.com/huggingface/datasets/issues/6773 | Dataset on Hub re-downloads every time? | [
"The caching works as expected when I try to reproduce this locally or on Colab...",
"hi @mariosasko , Thank you for checking. I also tried running this again just now, and it seems like the `load_dataset()` caches properly (though I'll double check later).\r\n\r\nI think the issue might be in the caching of the function output for `territories.map(lambda row: {'Claimants': row['Claimants'].split(';')})`. My current run re-ran this, even though I have run this many times before, and as demonstrated by loading from cache, the loaded dataset is the same.\r\n\r\nI wonder if the issue stems from using CSV output. Do you recommend changing to Parquet, and if so, is there an easy way to take the already uploaded data on the Hub and reformat?",
"This issue seems similar to https://github.com/huggingface/datasets/issues/6184 (`dill` serializes objects defined outside the `__main__` module by reference). You should be able to work around this limitation by defining the lambdas outside of `load_borderlines_hf` (as module variables) and then setting their `__module__` attribute's value to `None` to force serializing them by value, e.g., like this: \r\n```python\r\nsplit_Claimants_row = lambda row: {'Claimants': row['Claimants'].split(';')}\r\nsplit_Claimants_row.__module__ = None\r\n```",
"Thank you, I'll give this a try. Your fix makes sense to me, so this issue can be closed for now.\r\n\r\nUnrelated comment -- for \"Downloads last month\" on the hub page, I'm assuming for this project that each downloaded CSV is 1 download? The dataset consists of 51 CSVs, so I'm trying to see why it's incrementing so quickly (1125 2 days ago, 1246 right now).",
"This doc explains how we count \"Downloads last month\": https://huggingface.co/docs/hub/datasets-download-stats"
] | ### Describe the bug
Hi, I have a dataset on the hub [here](https://huggingface.co/datasets/manestay/borderlines). It has 1k+ downloads, which I sure is mostly just me and my colleagues working with it. It should have far fewer, since I'm using the same machine with a properly set up HF_HOME variable. However, whenever I run the below function `load_borderlines_hf`, it downloads the entire dataset from the hub and then does the other logic:
https://github.com/manestay/borderlines/blob/4e161f444661e2ebfe643f3fe149d9258d63a57d/run_gpt/lib.py#L80
Let me know what I'm doing wrong here, or if it's a bug with the `datasets` library itself. On the hub I have my data stored in CSVs, but several columns are lists, so that's why I have the code to map splitting on `;`. I looked into dataset loading scripts, but it seemed difficult to set up. I have verified that other `datasets` and `models` on my system are using the cache properly (e.g. I have a 13B parameter model and large datasets, but those are cached and don't redownload).
__EDIT: __ as pointed out in the discussion below, it may be the `map()` calls that aren't being cached properly. Supposing the `load_dataset()` retrieve from the cache, then it should be the case that the `map()` calls also retrieve from the cached output. But the `map()` commands re-execute sometimes.
### Steps to reproduce the bug
1. Copy and paste the function from [here](https://github.com/manestay/borderlines/blob/4e161f444661e2ebfe643f3fe149d9258d63a57d/run_gpt/lib.py#L80) (lines 80-100)
2. Run it in Python `load_borderlines_hf(None)`
3. It completes successfully, downloading from HF hub, then doing the mapping logic etc.
4. If you run it again after some time, it will re-download, ignoring the cache
### Expected behavior
Re-running the code, which calls `datasets.load_dataset('manestay/borderlines', 'territories')`, should use the cached version
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-5.14.21-150500.55.7-default-x86_64-with-glibc2.31
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 1.5.3
- `fsspec` version: 2023.10.0 | 6,773 |
https://github.com/huggingface/datasets/issues/6771 | Datasets FileNotFoundError when trying to generate examples. | [
"Hi! I've opened a PR in the repo to fix this issue: https://huggingface.co/datasets/RitchieP/VerbaLex_voice/discussions/6",
"@mariosasko Thanks for the PR and help! Guess I could close the issue for now. Appreciate the help!"
] | ### Discussed in https://github.com/huggingface/datasets/discussions/6768
<div type='discussions-op-text'>
<sup>Originally posted by **RitchieP** April 1, 2024</sup>
Currently, I have a dataset hosted on Huggingface with a custom script [here](https://huggingface.co/datasets/RitchieP/VerbaLex_voice).
I'm loading my dataset as below.
```py
from datasets import load_dataset, IterableDatasetDict
dataset = IterableDatasetDict()
dataset["train"] = load_dataset("RitchieP/VerbaLex_voice", "ar", split="train", use_auth_token=True, streaming=True)
dataset["test"] = load_dataset("RitchieP/VerbaLex_voice", "ar", split="test", use_auth_token=True, streaming=True)
```
And when I try to see the data I have loaded with
```py
list(dataset["train"].take(1))
```
And it gives me this stack trace
```
---------------------------------------------------------------------------
FileNotFoundError Traceback (most recent call last)
Cell In[2], line 1
----> 1 list(dataset["train"].take(1))
File /opt/conda/lib/python3.10/site-packages/datasets/iterable_dataset.py:1388, in IterableDataset.__iter__(self)
1385 yield formatter.format_row(pa_table)
1386 return
-> 1388 for key, example in ex_iterable:
1389 if self.features:
1390 # `IterableDataset` automatically fills missing columns with None.
1391 # This is done with `_apply_feature_types_on_example`.
1392 example = _apply_feature_types_on_example(
1393 example, self.features, token_per_repo_id=self._token_per_repo_id
1394 )
File /opt/conda/lib/python3.10/site-packages/datasets/iterable_dataset.py:1044, in TakeExamplesIterable.__iter__(self)
1043 def __iter__(self):
-> 1044 yield from islice(self.ex_iterable, self.n)
File /opt/conda/lib/python3.10/site-packages/datasets/iterable_dataset.py:234, in ExamplesIterable.__iter__(self)
233 def __iter__(self):
--> 234 yield from self.generate_examples_fn(**self.kwargs)
File ~/.cache/huggingface/modules/datasets_modules/datasets/RitchieP--VerbaLex_voice/9465eaee58383cf9d7c3e14111d7abaea56398185a641b646897d6df4e4732f7/VerbaLex_voice.py:127, in VerbaLexVoiceDataset._generate_examples(self, local_extracted_archive_paths, archives, meta_path)
125 for i, audio_archive in enumerate(archives):
126 print(audio_archive)
--> 127 for path, file in audio_archive:
128 _, filename = os.path.split(path)
129 if filename in metadata:
File /opt/conda/lib/python3.10/site-packages/datasets/download/streaming_download_manager.py:869, in _IterableFromGenerator.__iter__(self)
868 def __iter__(self):
--> 869 yield from self.generator(*self.args, **self.kwargs)
File /opt/conda/lib/python3.10/site-packages/datasets/download/streaming_download_manager.py:919, in ArchiveIterable._iter_from_urlpath(cls, urlpath, download_config)
915 @classmethod
916 def _iter_from_urlpath(
917 cls, urlpath: str, download_config: Optional[DownloadConfig] = None
918 ) -> Generator[Tuple, None, None]:
--> 919 compression = _get_extraction_protocol(urlpath, download_config=download_config)
920 # Set block_size=0 to get faster streaming
921 # (e.g. for hf:// and https:// it uses streaming Requests file-like instances)
922 with xopen(urlpath, "rb", download_config=download_config, block_size=0) as f:
File /opt/conda/lib/python3.10/site-packages/datasets/download/streaming_download_manager.py:400, in _get_extraction_protocol(urlpath, download_config)
398 urlpath, storage_options = _prepare_path_and_storage_options(urlpath, download_config=download_config)
399 try:
--> 400 with fsspec.open(urlpath, **(storage_options or {})) as f:
401 return _get_extraction_protocol_with_magic_number(f)
402 except FileNotFoundError:
File /opt/conda/lib/python3.10/site-packages/fsspec/core.py:100, in OpenFile.__enter__(self)
97 def __enter__(self):
98 mode = self.mode.replace("t", "").replace("b", "") + "b"
--> 100 f = self.fs.open(self.path, mode=mode)
102 self.fobjects = [f]
104 if self.compression is not None:
File /opt/conda/lib/python3.10/site-packages/fsspec/spec.py:1307, in AbstractFileSystem.open(self, path, mode, block_size, cache_options, compression, **kwargs)
1305 else:
1306 ac = kwargs.pop("autocommit", not self._intrans)
-> 1307 f = self._open(
1308 path,
1309 mode=mode,
1310 block_size=block_size,
1311 autocommit=ac,
1312 cache_options=cache_options,
1313 **kwargs,
1314 )
1315 if compression is not None:
1316 from fsspec.compression import compr
File /opt/conda/lib/python3.10/site-packages/fsspec/implementations/local.py:180, in LocalFileSystem._open(self, path, mode, block_size, **kwargs)
178 if self.auto_mkdir and "w" in mode:
179 self.makedirs(self._parent(path), exist_ok=True)
--> 180 return LocalFileOpener(path, mode, fs=self, **kwargs)
File /opt/conda/lib/python3.10/site-packages/fsspec/implementations/local.py:302, in LocalFileOpener.__init__(self, path, mode, autocommit, fs, compression, **kwargs)
300 self.compression = get_compression(path, compression)
301 self.blocksize = io.DEFAULT_BUFFER_SIZE
--> 302 self._open()
File /opt/conda/lib/python3.10/site-packages/fsspec/implementations/local.py:307, in LocalFileOpener._open(self)
305 if self.f is None or self.f.closed:
306 if self.autocommit or "w" not in self.mode:
--> 307 self.f = open(self.path, mode=self.mode)
308 if self.compression:
309 compress = compr[self.compression]
FileNotFoundError: [Errno 2] No such file or directory: '/kaggle/working/h'
```
After looking into the stack trace, and referring to the source codes, it looks like its trying to access a directory in the notebook's environment and I don't understand why.
Not sure if its a bug in Datasets library, so I'm opening a discussions first. Feel free to ask for more information if needed. Appreciate any help in advance!</div>
Hi, referring to the discussion title above, after further digging, I think it's an issue within the datasets library. But not quite sure where it is.
If you require any more info or actions from me, please let me know. Appreciate any help in advance! | 6,771 |
https://github.com/huggingface/datasets/issues/6770 | [Bug Report] `datasets==2.18.0` is not compatible with `fsspec==2023.12.2` | [
"You should be able to fix this by updating `huggingface_hub` with `pip install -U huggingface_hub`. We use this package under the hood to resolve the Hub's files."
] | ### Describe the bug
`Datasets==2.18.0` is not compatible with `fsspec==2023.12.2`.
I have to downgrade fsspec to `fsspec==2023.10.0` to make `Datasets==2.18.0` work properly.
### Steps to reproduce the bug
To reproduce the bug:
1. Make sure that `Datasets==2.18.0` and `fsspec==2023.12.2`.
2. Run the following code:
```
from datasets import load_dataset
dataset = load_dataset("trec")
```
3. Then one will get the following error message:
```
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/opt/conda/lib/python3.10/site-packages/datasets/load.py", line 2556, in load_dataset
builder_instance = load_dataset_builder(
File "/opt/conda/lib/python3.10/site-packages/datasets/load.py", line 2265, in load_dataset_builder
builder_instance: DatasetBuilder = builder_cls(
File "/opt/conda/lib/python3.10/site-packages/datasets/builder.py", line 371, in __init__
self.config, self.config_id = self._create_builder_config(
File "/opt/conda/lib/python3.10/site-packages/datasets/builder.py", line 620, in _create_builder_config
builder_config._resolve_data_files(
File "/opt/conda/lib/python3.10/site-packages/datasets/builder.py", line 211, in _resolve_data_files
self.data_files = self.data_files.resolve(base_path, download_config)
File "/opt/conda/lib/python3.10/site-packages/datasets/data_files.py", line 799, in resolve
out[key] = data_files_patterns_list.resolve(base_path, download_config)
File "/opt/conda/lib/python3.10/site-packages/datasets/data_files.py", line 752, in resolve
resolve_pattern(
File "/opt/conda/lib/python3.10/site-packages/datasets/data_files.py", line 393, in resolve_pattern
raise FileNotFoundError(error_msg)
FileNotFoundError: Unable to find 'hf://datasets/trec@65752bf53af25bc935a0dce92fb5b6c930728450/default/train/0000.parquet' with any supported extension ['.csv', '.tsv', '.json', '.jsonl', '.parquet', '.geoparquet', '.gpq', '.arrow', '.txt', '.tar', '.blp', '.bmp', '.dib', '.bufr', '.cur', '.pcx', '.dcx', '.dds', '.ps', '.eps', '.fit', '.fits', '.fli', '.flc', '.ftc', '.ftu', '.gbr', '.gif', '.grib', '.h5', '.hdf', '.png', '.apng', '.jp2', '.j2k', '.jpc', '.jpf', '.jpx', '.j2c', '.icns', '.ico', '.im', '.iim', '.tif', '.tiff', '.jfif', '.jpe', '.jpg', '.jpeg', '.mpg', '.mpeg', '.msp', '.pcd', '.pxr', '.pbm', '.pgm', '.ppm', '.pnm', '.psd', '.bw', '.rgb', '.rgba', '.sgi', '.ras', '.tga', '.icb', '.vda', '.vst', '.webp', '.wmf', '.emf', '.xbm', '.xpm', '.BLP', '.BMP', '.DIB', '.BUFR', '.CUR', '.PCX', '.DCX', '.DDS', '.PS', '.EPS', '.FIT', '.FITS', '.FLI', '.FLC', '.FTC', '.FTU', '.GBR', '.GIF', '.GRIB', '.H5', '.HDF', '.PNG', '.APNG', '.JP2', '.J2K', '.JPC', '.JPF', '.JPX', '.J2C', '.ICNS', '.ICO', '.IM', '.IIM', '.TIF', '.TIFF', '.JFIF', '.JPE', '.JPG', '.JPEG', '.MPG', '.MPEG', '.MSP', '.PCD', '.PXR', '.PBM', '.PGM', '.PPM', '.PNM', '.PSD', '.BW', '.RGB', '.RGBA', '.SGI', '.RAS', '.TGA', '.ICB', '.VDA', '.VST', '.WEBP', '.WMF', '.EMF', '.XBM', '.XPM', '.aiff', '.au', '.avr', '.caf', '.flac', '.htk', '.svx', '.mat4', '.mat5', '.mpc2k', '.ogg', '.paf', '.pvf', '.raw', '.rf64', '.sd2', '.sds', '.ircam', '.voc', '.w64', '.wav', '.nist', '.wavex', '.wve', '.xi', '.mp3', '.opus', '.AIFF', '.AU', '.AVR', '.CAF', '.FLAC', '.HTK', '.SVX', '.MAT4', '.MAT5', '.MPC2K', '.OGG', '.PAF', '.PVF', '.RAW', '.RF64', '.SD2', '.SDS', '.IRCAM', '.VOC', '.W64', '.WAV', '.NIST', '.WAVEX', '.WVE', '.XI', '.MP3', '.OPUS', '.zip']
```
4. Similar issue also found for the following code:
```
dataset = load_dataset("sst", "default")
```
### Expected behavior
If the dataset is loaded correctly, one will have:
```
>>> print(dataset)
DatasetDict({
train: Dataset({
features: ['text', 'coarse_label', 'fine_label'],
num_rows: 5452
})
test: Dataset({
features: ['text', 'coarse_label', 'fine_label'],
num_rows: 500
})
})
>>>
```
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-6.2.0-35-generic-x86_64-with-glibc2.31
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.1
- Pandas version: 2.2.1
- `fsspec` version: 2023.12.2 | 6,770 |
https://github.com/huggingface/datasets/issues/6769 | (Willing to PR) Datasets with custom python objects | [] | ### Feature request
Hi thanks for the library! I would like to have a huggingface Dataset, and one of its column is custom (non-serializable) Python objects. For example, a minimal code:
```
class MyClass:
pass
dataset = datasets.Dataset.from_list([
dict(a=MyClass(), b='hello'),
])
```
It gives error:
```
ArrowInvalid: Could not convert <__main__.MyClass object at 0x7a852830d050> with type MyClass: did not recognize Python value type when inferring an Arrow data type
```
I guess it is because Dataset forces to convert everything into arrow format. However, is there any ways to make the scenario work? Thanks!
### Motivation
(see above)
### Your contribution
Yes, I am happy to PR!
Cross-posted: /static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdatasets-with-custom-python-objects%2F79050%3Fu%3Dfzyzcjy
EDIT: possibly related https://github.com/huggingface/datasets/issues/5766 | 6,769 |
https://github.com/huggingface/datasets/issues/6765 | Compatibility issue between s3fs, fsspec, and datasets | [
"Hi! Instead of running `pip install` separately for each package, you should pass all the packages to a single `pip install` call (in this case, `pip install datasets s3fs`) to let `pip` properly resolve their versions.",
"> Hi! Instead of running `pip install` separately for each package, you should pass all the packages to a single `pip install` call (in this case, `pip install datasets s3fs`) to let `pip` properly resolve their versions.\r\n\r\nThanks so much! My inexperience with pip is showing π π ",
"> Hi! Instead of running `pip install` separately for each package, you should pass all the packages to a single `pip install` call (in this case, `pip install datasets s3fs`) to let `pip` properly resolve their versions.\r\n\r\nyou are awesome bro"
] | ### Describe the bug
Here is the full error stack when installing:
```
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
datasets 2.18.0 requires fsspec[http]<=2024.2.0,>=2023.1.0, but you have fsspec 2024.3.1 which is incompatible.
Successfully installed aiobotocore-2.12.1 aioitertools-0.11.0 botocore-1.34.51 fsspec-2024.3.1 jmespath-1.0.1 s3fs-2024.3.1 urllib3-2.0.7 wrapt-1.16.0
```
When I install with pip, pip allows this error to exist while still installing s3fs, but this error breaks poetry, since poetry will refuse to install s3fs because of the dependency conflict.
Maybe I'm missing something so maybe it's not a bug but some mistake on my end? Any input would be helpful. Thanks!
### Steps to reproduce the bug
1. conda create -n tmp python=3.10 -y
2. conda activate tmp
3. pip install datasets
4. pip install s3fs
### Expected behavior
I would expect there to be no error.
### Environment info
MacOS (ARM), Python3.10, conda 23.11.0. | 6,765 |
https://github.com/huggingface/datasets/issues/6764 | load_dataset can't work with symbolic links | [] | ### Feature request
Enable the `load_dataset` function to load local datasets with symbolic links.
E.g, this dataset can be loaded:
βββ example_dataset/
β βββ data/
β β βββ train/
β β β βββ file0
β β β βββ file1
β β βββ dev/
β β β βββ file2
β β β βββ file3
β βββ metadata.csv
while this dataset can't:
βββ example_dataset_symlink/
β βββ data/
β β βββ train/
β β β βββ sym0 -> file0
β β β βββ sym1 -> file1
β β βββ dev/
β β β βββ sym2 -> file2
β β β βββ sym3 -> file3
β βββ metadata.csv
I have created an example dataset in order to reproduce the problem:
1. Unzip `example_dataset.zip`.
2. Run `no_symlink.sh`. Training should start without issues.
3. Run `symlink.sh`. You will see that all four examples will be in train split, instead of having two examples in train and two examples in dev. The script won't load the correct audio files.
[example_dataset.zip](https://github.com/huggingface/datasets/files/14807053/example_dataset.zip)
### Motivation
I have a very large dataset locally. Instead of initiating training on the entire dataset, I need to start training on smaller subsets of the data. Due to the purpose of the experiments I am running, I will need to create many smaller datasets with overlapping data. Instead of copying the all the files for each subset, I would prefer copying symbolic links of the data. This way, the memory usage would not significantly increase beyond the initial dataset size.
Advantages of this approach:
- It would leave a smaller memory footprint on the hard drive
- Creating smaller datasets would be much faster
### Your contribution
I would gladly contribute, if this is something useful to the community. It seems like a simple change of code, something like `file_path = os.path.realpath(file_path)` should be added before loading the files. If anyone has insights on how to incorporate this functionality, I would greatly appreciate your knowledge and input. | 6,764 |
https://github.com/huggingface/datasets/issues/6760 | Load codeparrot/apps raising UnicodeDecodeError in datasets-2.18.0 | [
"The same error with mteb datasets.",
"Unfortunately, I'm unable to reproduce this error locally or on Colab.",
"Here is the requirements.txt from a clean virtual environment (managed by conda) where I only install `datasets` by \r\n`pip install datasets`. \r\nThe pip list:\r\n```\r\naiohttp==3.9.3\r\naiosignal==1.3.1\r\nattrs==23.2.0\r\ncertifi==2024.2.2\r\ncharset-normalizer==3.3.2\r\ndatasets==2.18.0\r\ndill==0.3.8\r\nfilelock==3.13.3\r\nfrozenlist==1.4.1\r\nfsspec==2024.2.0\r\nhuggingface-hub==0.22.2\r\nidna==3.6\r\nmultidict==6.0.5\r\nmultiprocess==0.70.16\r\nnumpy==1.26.4\r\npackaging==24.0\r\npandas==2.2.1\r\npyarrow==15.0.2\r\npyarrow-hotfix==0.6\r\npython-dateutil==2.9.0.post0\r\npytz==2024.1\r\nPyYAML==6.0.1\r\nrequests==2.31.0\r\nsix==1.16.0\r\ntqdm==4.66.2\r\ntyping_extensions==4.11.0\r\ntzdata==2024.1\r\nurllib3==2.2.1\r\nxxhash==3.4.1\r\nyarl==1.9.4\r\n```\r\nAnd the error can be reproduced.\r\n\r\nDowngrading to datasets==2.14.6 changes some packages' versions:\r\n\r\n```\r\nSuccessfully installed datasets-2.14.6 dill-0.3.7 fsspec-2023.10.0 multiprocess-0.70.15\r\n```\r\nand the dataset can be downloaded and loaded. \r\n\r\nThen I upgrade the version to 2.18.0 again; now the dataset can be loaded with such a line:\r\n```Using the latest cached version of the module from /home/xxx/.cache/huggingface/modules/datasets_modules/datasets/codeparrot--apps/04ac807715d07d6e5cc580f59cdc8213cd7dc4529d0bb819cca72c9f8e8c1aa5 (last modified on Sun Apr 7 09:06:43 2024) since it couldn't be found locally at codeparrot/apps, or remotely on the Hugging Face Hub. ```\r\n\r\nSo the latest version works wrong when requesting the dataset info. \r\n\r\n**But if you cannot reproduce this, I may ignore some detailed information: I use `HF_ENDPOINT=https://hf-mirror.com` for some reason (if not use this I cannot connect to huggingface resources) and the error occurs when requesting the dataset's info card.** \r\nMaybe the error is caused by this environment variable.\r\nI'll open an issue in the author's repo now.",
"> Here is the requirements.txt from a clean virtual environment (managed by conda) where I only install `datasets` by `pip install datasets`. The pip list:\r\n> \r\n> ```\r\n> aiohttp==3.9.3\r\n> aiosignal==1.3.1\r\n> attrs==23.2.0\r\n> certifi==2024.2.2\r\n> charset-normalizer==3.3.2\r\n> datasets==2.18.0\r\n> dill==0.3.8\r\n> filelock==3.13.3\r\n> frozenlist==1.4.1\r\n> fsspec==2024.2.0\r\n> huggingface-hub==0.22.2\r\n> idna==3.6\r\n> multidict==6.0.5\r\n> multiprocess==0.70.16\r\n> numpy==1.26.4\r\n> packaging==24.0\r\n> pandas==2.2.1\r\n> pyarrow==15.0.2\r\n> pyarrow-hotfix==0.6\r\n> python-dateutil==2.9.0.post0\r\n> pytz==2024.1\r\n> PyYAML==6.0.1\r\n> requests==2.31.0\r\n> six==1.16.0\r\n> tqdm==4.66.2\r\n> typing_extensions==4.11.0\r\n> tzdata==2024.1\r\n> urllib3==2.2.1\r\n> xxhash==3.4.1\r\n> yarl==1.9.4\r\n> ```\r\n> \r\n> And the error can be reproduced.\r\n> \r\n> Downgrading to datasets==2.14.6 changes some packages' versions:\r\n> \r\n> ```\r\n> Successfully installed datasets-2.14.6 dill-0.3.7 fsspec-2023.10.0 multiprocess-0.70.15\r\n> ```\r\n> \r\n> and the dataset can be downloaded and loaded.\r\n> \r\n> Then I upgrade the version to 2.18.0 again; now the dataset can be loaded with such a line: `Using the latest cached version of the module from /home/xxx/.cache/huggingface/modules/datasets_modules/datasets/codeparrot--apps/04ac807715d07d6e5cc580f59cdc8213cd7dc4529d0bb819cca72c9f8e8c1aa5 (last modified on Sun Apr 7 09:06:43 2024) since it couldn't be found locally at codeparrot/apps, or remotely on the Hugging Face Hub. `\r\n> \r\n> So the latest version works wrong when requesting the dataset info.\r\n> \r\n> **But if you cannot reproduce this, I may ignore some detailed information: I use `HF_ENDPOINT=https://hf-mirror.com` for some reason (if not use this I cannot connect to huggingface resources) and the error occurs when requesting the dataset's info card.** Maybe the error is caused by this environment variable. I'll open an issue in the author's repo now.\r\n\r\nThis is useful and my same error is settled!!!"
] | ### Describe the bug
This happens with datasets-2.18.0; I downgraded the version to 2.14.6 fixing this temporarily.
```
Traceback (most recent call last):
File "/home/xxx/miniconda3/envs/py310/lib/python3.10/site-packages/datasets/load.py", line 2556, in load_dataset
builder_instance = load_dataset_builder(
File "/home/xxx/miniconda3/envs/py310/lib/python3.10/site-packages/datasets/load.py", line 2228, in load_dataset_builder
dataset_module = dataset_module_factory(
File "/home/xxx/miniconda3/envs/py310/lib/python3.10/site-packages/datasets/load.py", line 1879, in dataset_module_factory
raise e1 from None
File "/home/xxx/miniconda3/envs/py310/lib/python3.10/site-packages/datasets/load.py", line 1831, in dataset_module_factory
can_load_config_from_parquet_export = "DEFAULT_CONFIG_NAME" not in f.read()
File "/home/xxx/miniconda3/envs/py310/lib/python3.10/codecs.py", line 322, in decode
(result, consumed) = self._buffer_decode(data, self.errors, final)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position 1: invalid start byte
```
### Steps to reproduce the bug
1. Using Python3.10/3.11
2. Install datasets-2.18.0
3. test with
```
from datasets import load_dataset
dataset = load_dataset("codeparrot/apps")
```
### Expected behavior
Normally it should manage to download and load the dataset without such error.
### Environment info
Ubuntu, Python3.10/3.11 | 6,760 |
https://github.com/huggingface/datasets/issues/6759 | Persistent multi-process Pool | [] | ### Feature request
Running .map and filter functions with `num_procs` consecutively instantiates several multiprocessing pools iteratively.
As instantiating a Pool is very resource intensive it can be a bottleneck to performing iteratively filtering.
My ideas:
1. There should be an option to declare `persistent_workers` similar to pytorch DataLoader. Downside would be that would be complex to determine the correct resource allocation and deallocation of the pool. i.e. the dataset can outlive the utility of the pool.
2. Provide a pool as an argument. Downside would be the expertise required by the user. Upside, is that there is better resource management.
### Motivation
Is really slow to iteratively perform map and filter operations on a dataset.
### Your contribution
If approved I could integrate it. I would need to know what method would be most suitable to implement from the two options above. | 6,759 |
https://github.com/huggingface/datasets/issues/6758 | Passing `sample_by` to `load_dataset` when loading text data does not work | [
"Thanks for reporting! We are working on a fix."
] | ### Describe the bug
I have a dataset that consists of a bunch of text files, each representing an example. There is an undocumented `sample_by` argument for the `TextConfig` class that is used by `Text` to decide whether to split files into lines, paragraphs or take them whole. Passing `sample_by=βdocumentβ` to `load_dataset` results in files getting split into lines regardless. I have edited `src/datasets/packaged_modules/text/text.py` for myself to switch the default and it works fine.
As a side note, the `if-else` for `sample_by` will silently load an empty dataset if someone makes a typo in the argument, which is not ideal.
### Steps to reproduce the bug
1. Prepare data as a bunch of files in a directory.
2. Load that data via `load_dataset(βtextβ, data_files=<data_dir>/<files_glob>, β¦, sample_by=βdocumentβ)`.
3. Inspect the resultant dataset β every item should have the form of `{βtextβ: <a line from a file>}`.
### Expected behavior
`load_dataset(βtextβ, data_files=<data_dir>/<files_glob>, β¦, sample_by=βdocumentβ)` should result in a dataset with items of the form `{βtextβ: <one document>}`.
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-5.15.0-1046-nvidia-x86_64-with-glibc2.35
- Python version: 3.11.8
- `huggingface_hub` version: 0.21.4
- PyArrow version: 15.0.2
- Pandas version: 2.2.1
- `fsspec` version: 2024.2.0 | 6,758 |
https://github.com/huggingface/datasets/issues/6756 | Support SQLite files? | [
"You can use `Dataset.from_sql(path_to_sql_file)` already. Though we haven't added the Sql dataset builder to the `_PACKAGED_DATASETS_MODULES` list or in `_EXTENSION_TO_MODULE` to map `.sqlite` to the Sql dataset builder\r\n\r\nThis would allow to load a dataset repository with a `.sqlite` file using `load_dataset` and enable the Dataset Viewer",
"Considering `Dataset.from_sql`'s (extremely) low usage, I don't think many users are interested in using this format for their datasets. Also, SQLite files are hard/impossible to stream efficiently and require custom logic to define splits/subsets, so IMO we shouldn't encourage people to use SQLite on the Hub.\r\n\r\n@severo Do you have some real-world examples of datasets published in this format?",
"No. Indeed, it seems better to explicitly not support sqlite"
] | ### Feature request
Support loading a dataset from a SQLite file
https://huggingface.co/datasets/severo/test_iris_sqlite/tree/main
### Motivation
SQLite is a popular file format.
### Your contribution
See discussion on slack: https://huggingface.slack.com/archives/C04L6P8KNQ5/p1702481859117909 (internal)
In particular: a SQLite file can contain multiple tables, which might be matched to multiple configs. Maybe the detail of splits and configs should be defined in the README YAML, or use the same format as for ZIP files: `Iris.sqlite::Iris`.
See dataset here: https://huggingface.co/datasets/severo/test_iris_sqlite
Note: should we also support DuckDB files? | 6,756 |
https://github.com/huggingface/datasets/issues/6755 | Small typo on the documentation | [
"Thanks for reporting @fostiropoulos! I've edited your comment to fix the link to the problematic line.\r\n",
"@mariosasko can i take this up?",
"#self-assign"
] | ### Describe the bug
There is a small typo on https://github.com/huggingface/datasets/blob/d5468836fe94e8be1ae093397dd43d4a2503b926/src/datasets/dataset_dict.py#L938
It should be `caching is enabled`.
### Steps to reproduce the bug
Please visit
https://github.com/huggingface/datasets/blob/d5468836fe94e8be1ae093397dd43d4a2503b926/src/datasets/dataset_dict.py#L938
### Expected behavior
`caching is enabled`
### Environment info
- `datasets` version: 2.17.1
- Platform: Linux-5.15.0-101-generic-x86_64-with-glibc2.35
- Python version: 3.11.7
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.1
- `fsspec` version: 2023.10.0 | 6,755 |
https://github.com/huggingface/datasets/issues/6753 | Type error when importing datasets on Kaggle | [
"I have the same problem \r\nIt seems that it only appears when you are using GPU \r\nIt seems to work fine with the 2.17 version though",
"Same here.",
"> I have the same problem\r\n> It seems that it only appears when you are using GPU\r\n> It seems to work fine with the 2.17 version though\r\n\r\nI downgraded from 2.18 to 2.17, and it works with CPU/GPU .. except now pyarrow complains\r\n\r\n```\r\n...\r\nFile /opt/conda/lib/python3.10/site-packages/pyarrow/array.pxi:830, in pyarrow.lib._PandasConvertible.to_pandas()\r\n\r\nFile /opt/conda/lib/python3.10/site-packages/pyarrow/table.pxi:3989, in pyarrow.lib.Table._to_pandas()\r\n\r\nImportError: cannot import name table_to_blockmanager\r\n```\r\n\r\nsee also https://www.kaggle.com/competitions/pii-detection-removal-from-educational-data/discussion/487474#2722594",
"Solved for me by downgrading `!pip install -U datasets==2.16.0` Works with gpu aswell",
"I think you should remain open this issue. It works at the previous version but not the latter versions. It is possible as a bug that the maintainer could take note for.",
"> Solved for me by downgrading `!pip install -U datasets==2.16.0` Works with gpu as well\r\n\r\nVerified it's working w/ GPU if I make these 3 updates.\r\n\r\n```\r\ndatasets==2.16.0\r\nfsspec==2023.10.0\r\ngcsfs==2023.10.0\r\n```\r\n\r\nbut the issue shouldn't be closed, this is just a workaround until they get the issue with 2.18.0 resolved.\r\n\r\nSee also: https://www.kaggle.com/competitions/pii-detection-removal-from-educational-data/discussion/487474",
"> > Solved for me by downgrading `!pip install -U datasets==2.16.0` Works with gpu as well\r\n> \r\n> Verified it's working w/ GPU if I make these 3 updates.\r\n> \r\n> ```\r\n> datasets==2.16.0\r\n> fsspec==2023.10.0\r\n> gcsfs==2023.10.0\r\n> ```\r\n> \r\n> but the issue shouldn't be closed, this is just a workaround until they get the issue with 2.18.0 resolved.\r\n> \r\n> See also: https://www.kaggle.com/competitions/pii-detection-removal-from-educational-data/discussion/487474\r\n\r\nThis also works for me, thanks"
] | ### Describe the bug
When trying to run
```
import datasets
print(datasets.__version__)
```
It generates the following error
```
TypeError: expected string or bytes-like object
```
It looks like It cannot find the valid versions of `fsspec`
though fsspec version is fine when I checked Via command
```
import fsspec
print(fsspec.__version__)
β
# output: 2024.3.1
```
Detailed crash report
```
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[1], line 1
----> 1 import datasets
2 print(datasets.__version__)
File /opt/conda/lib/python3.10/site-packages/datasets/__init__.py:18
1 # ruff: noqa
2 # Copyright 2020 The HuggingFace Datasets Authors and the TensorFlow Datasets Authors.
3 #
(...)
13 # See the License for the specific language governing permissions and
14 # limitations under the License.
16 __version__ = "2.18.0"
---> 18 from .arrow_dataset import Dataset
19 from .arrow_reader import ReadInstruction
20 from .builder import ArrowBasedBuilder, BeamBasedBuilder, BuilderConfig, DatasetBuilder, GeneratorBasedBuilder
File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:66
63 from multiprocess import Pool
64 from tqdm.contrib.concurrent import thread_map
---> 66 from . import config
67 from .arrow_reader import ArrowReader
68 from .arrow_writer import ArrowWriter, OptimizedTypedSequence
File /opt/conda/lib/python3.10/site-packages/datasets/config.py:41
39 # Imports
40 DILL_VERSION = version.parse(importlib.metadata.version("dill"))
---> 41 FSSPEC_VERSION = version.parse(importlib.metadata.version("fsspec"))
42 PANDAS_VERSION = version.parse(importlib.metadata.version("pandas"))
43 PYARROW_VERSION = version.parse(importlib.metadata.version("pyarrow"))
File /opt/conda/lib/python3.10/site-packages/packaging/version.py:49, in parse(version)
43 """
44 Parse the given version string and return either a :class:`Version` object
45 or a :class:`LegacyVersion` object depending on if the given version is
46 a valid PEP 440 version or a legacy version.
47 """
48 try:
---> 49 return Version(version)
50 except InvalidVersion:
51 return LegacyVersion(version)
File /opt/conda/lib/python3.10/site-packages/packaging/version.py:264, in Version.__init__(self, version)
261 def __init__(self, version: str) -> None:
262
263 # Validate the version and parse it into pieces
--> 264 match = self._regex.search(version)
265 if not match:
266 raise InvalidVersion(f"Invalid version: '{version}'")
TypeError: expected string or bytes-like object
```
### Steps to reproduce the bug
1. run `!pip install -U datasets` on kaggle
2. check datasets is installed via
```
import datasets
print(datasets.__version__)
```
### Expected behavior
Expected to print datasets version, like `2.18.0`
### Environment info
Running on Kaggle, latest enviornment , here is the notebook https://www.kaggle.com/code/jtv199/mistrial-7b-part2 | 6,753 |
https://github.com/huggingface/datasets/issues/6752 | Precision being changed from float16 to float32 unexpectedly | [
"This is because of the formatter (`torch` in this case).\r\nIt defaults to `float32`.\r\n\r\nYou can load it in `float16` using `dataset.set_format(\"torch\", dtype=torch.float16)`."
] | ### Describe the bug
I'm loading a HuggingFace Dataset for images.
I'm running a preprocessing (map operation) step that runs a few operations, one of them being conversion to float16. The Dataset features also say that the 'img' is of type float16. Whenever I take an image from that HuggingFace Dataset instance, the type turns out to be float32.
### Steps to reproduce the bug
```python
import torchvision.transforms.v2 as transforms
from datasets import load_dataset
dataset = load_dataset('cifar10', split='test')
dataset = dataset.with_format("torch")
data_transform = transforms.Compose([transforms.Resize((32, 32)),
transforms.ToDtype(torch.float16, scale=True),
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]),
])
def _preprocess(examples):
# Permutes from (BS x H x W x C) to (BS x C x H x W)
images = torch.permute(examples['img'], (0, 3, 2, 1))
examples['img'] = data_transform(images)
return examples
dataset = dataset.map(_preprocess, batched=True, batch_size=8)
```
Now at this point the dataset.features are showing float16 which is great because that's what I want.
```python
print(data_loader.features['img'])
Sequence(feature=Sequence(feature=Sequence(feature=Value(dtype='float16', id=None), length=-1, id=None), length=-1, id=None), length=-1, id=None)
```
But when I try to sample an image from this dataloader; I'm getting a float32 image, when I'm expecting float16:
```python
print(next(iter(data_loader))['img'].dtype)
torch.float32
```
### Expected behavior
I'm expecting the images loaded after the transformation to stay in float16.
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-5.15.146.1-microsoft-standard-WSL2-x86_64-with-glibc2.31
- Python version: 3.10.9
- `huggingface_hub` version: 0.21.4
- PyArrow version: 14.0.2
- Pandas version: 2.0.3
- `fsspec` version: 2023.10.0 | 6,752 |
https://github.com/huggingface/datasets/issues/6750 | `load_dataset` requires a network connection for local download? | [
"Are you using `HF_DATASETS_OFFLINE=1` ?",
"> Are you using `HF_DATASETS_OFFLINE=1` ?\r\n\r\nThis doesn't work for me. `datasets=2.18.0`\r\n\r\n`test.py`:\r\n```\r\nimport datasets\r\n\r\ndatasets.utils.logging.set_verbosity_info()\r\n\r\nds = datasets.load_dataset('C-MTEB/AFQMC', revision='b44c3b011063adb25877c13823db83bb193913c4')\r\n\r\nprint(ds)\r\n```\r\n\r\nrun `python test.py`\r\n```\r\nGenerating dataset afqmc (/home/data/.cache/huggingface/datasets/C-MTEB___afqmc/default/0.0.0/b44c3b011063adb25877c13823db83bb193913c4)\r\nDownloading and preparing dataset afqmc/default to /home/data/.cache/huggingface/datasets/C-MTEB___afqmc/default/0.0.0/b44c3b011063adb25877c13823db83bb193913c4...\r\nDataset not on Hf google storage. Downloading and preparing it from source\r\nhf://datasets/C-MTEB/AFQMC@b44c3b011063adb25877c13823db83bb193913c4/data/validation-00000-of-00001-b8fc393b5ddedac7.parquet not found in cache or force_download set to True, downloading to /home/data/.cache/huggingface/datasets/downloads/78949f93104662359f4f3d5a2f7ec1ae37af5a5af44420a51212ea08c0be966b.incomplete\r\nDownloading data: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 240k/240k [00:01<00:00, 178kB/s]\r\nstoring hf://datasets/C-MTEB/AFQMC@b44c3b011063adb25877c13823db83bb193913c4/data/validation-00000-of-00001-b8fc393b5ddedac7.parquet in cache at /home/data/.cache/huggingface/datasets/downloads/78949f93104662359f4f3d5a2f7ec1ae37af5a5af44420a51212ea08c0be966b\r\ncreating metadata file for /home/data/.cache/huggingface/datasets/downloads/78949f93104662359f4f3d5a2f7ec1ae37af5a5af44420a51212ea08c0be966b\r\nDownloading took 0.0 min\r\nChecksum Computation took 0.0 min\r\nGenerating test split\r\nGenerating test split: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 3861/3861 [00:00<00:00, 3972.00 examples/s]\r\nGenerating train split\r\nGenerating train split: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 34334/34334 [00:00<00:00, 34355.50 examples/s]\r\nGenerating validation split\r\nGenerating validation split: 100%|ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 4316/4316 [00:00<00:00, 4477.00 examples/s]\r\nAll the splits matched successfully.\r\nDataset afqmc downloaded and prepared to /home/data/.cache/huggingface/datasets/C-MTEB___afqmc/default/0.0.0/b44c3b011063adb25877c13823db83bb193913c4. Subsequent calls will reuse this data.\r\nDatasetDict({\r\n test: Dataset({\r\n features: ['sentence1', 'sentence2', 'score', 'idx'],\r\n num_rows: 3861\r\n })\r\n train: Dataset({\r\n features: ['sentence1', 'sentence2', 'score', 'idx'],\r\n num_rows: 34334\r\n })\r\n validation: Dataset({\r\n features: ['sentence1', 'sentence2', 'score', 'idx'],\r\n num_rows: 4316\r\n })\r\n})\r\n```\r\n\r\nThen run `HF_DATASETS_OFFLINE=1 python test.py`\r\n```\r\nTraceback (most recent call last):\r\n File \"test.py\", line 9, in <module>\r\n ds = datasets.load_dataset('C-MTEB/AFQMC', revision='b44c3b011063adb25877c13823db83bb193913c4')\r\n File \"/dev/shm/tmp_env/lib/python3.10/site-packages/datasets/load.py\", line 2556, in load_dataset\r\n builder_instance = load_dataset_builder(\r\n File \"/dev/shm/tmp_env/lib/python3.10/site-packages/datasets/load.py\", line 2228, in load_dataset_builder\r\n dataset_module = dataset_module_factory(\r\n File \"/dev/shm/tmp_env/lib/python3.10/site-packages/datasets/load.py\", line 1871, in dataset_module_factory\r\n raise ConnectionError(f\"Couldn't reach the Hugging Face Hub for dataset '{path}': {e1}\") from None\r\nConnectionError: Couldn't reach the Hugging Face Hub for dataset 'C-MTEB/AFQMC': Offline mode is enabled.\r\n```\r\n\r\n",
"I was having similar inexplicable issues.\r\n\r\nDoing this I *think* helped, but, `datasets` still *clearly* does not want to respect the cache:\r\n\r\n```python\r\npip install --upgrade datasets # now it is 2.18.0\r\nHF_DATASETS_OFFLINE=\"1\" python blah.py\r\n```\r\n\r\nOr similarly, I must spacify that env var to resuse the cache, IE, no arg to `load_dataset` helps it reuse the cache:\r\n\r\n```python\r\n\r\nimport os\r\nos.environ[\"HF_DATASETS_OFFLINE\"] = \"1\"\r\n\r\nimport logging\r\nlogging.basicConfig(level=logging.DEBUG)\r\n\r\nimport datasets\r\n# >>> datasets.__version__\r\n# '2.18.0'\r\n\r\ndatasets.utils.logging.set_verbosity_info()\r\ndata = datasets.load_dataset(\"c-s-ale/dolly-15k-instruction-alpaca-format\")\r\n```"
] | ### Describe the bug
Hi all - I see that in the past a network dependency has been mistakenly introduced into `load_dataset` even for local loads. Is it possible this has happened again?
### Steps to reproduce the bug
```
>>> import datasets
>>> datasets.load_dataset("hh-rlhf")
Repo card metadata block was not found. Setting CardData to empty.
*hangs bc i'm firewalled*
````
stack trace from ctrl-c:
```
^CTraceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/load.py", line 2582, in load_dataset
builder_instance.download_and_prepare(
output_path = get_from_cache( [0/122]
File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 532, in get_from_cache
response = http_head(
File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 419, in http_head
response = _request_with_retry(
File "/home/jobuser/.local/lib/python3.10/site-packages/datasets/utils/file_utils.py", line 304, in _request_with_retry
response = requests.request(method=method.upper(), url=url, timeout=timeout, **params)
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/api.py", line 59, in request
return session.request(method=method, url=url, **kwargs)
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/sessions.py", line 587, in request
resp = self.send(prep, **send_kwargs)
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/sessions.py", line 701, in send
r = adapter.send(request, **kwargs)
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/requests/adapters.py", line 487, in send
resp = conn.urlopen(
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 703, in urlopen
httplib_response = self._make_request(
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 386, in _make_request
self._validate_conn(conn)
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connectionpool.py", line 1042, in _validate_conn
conn.connect()
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connection.py", line 363, in connect
self.sock = conn = self._new_conn()
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/connection.py", line 174, in _new_conn
conn = connection.create_connection(
File "/home/jobuser/build/lipy-flytekit-image/environments/satellites/python/lib/python3.10/site-packages/urllib3/util/connection.py", line 85, in create_connection
sock.connect(sa)
KeyboardInterrupt
```
### Expected behavior
loads the dataset
### Environment info
```
> pip show datasets
Name: datasets
Version: 2.18.0
```
Python 3.10.2 | 6,750 |
https://github.com/huggingface/datasets/issues/6748 | Strange slicing behavior | [
"As explained in the [docs](https://huggingface.co/docs/datasets/v2.18.0/en/access#slicing), slicing a `Dataset` returns a dictionary that maps its column names to their values. So, `len(dataset[:300])=2` is expected, assuming your dataset has 2 columns (the returned dict has 2 keys, but each value in the dict has 300 items).\r\n` "
] | ### Describe the bug
I have loaded a dataset, and then slice first 300 samples using `:` ops, however, the resulting dataset is not expected, as the output below:
```bash
len(dataset)=1050324
len(dataset[:300])=2
len(dataset[0:300])=2
len(dataset.select(range(300)))=300
```
### Steps to reproduce the bug
load a dataset then:
```bash
dataset = load_from_disk(args.train_data_dir)
print(f"{len(dataset)=}", flush=True)
print(f"{len(dataset[:300])=}", flush=True)
print(f"{len(dataset[0:300])=}", flush=True)
print(f"{len(dataset.select(range(300)))=}", flush=True)
```
### Expected behavior
```bash
len(dataset)=1050324
len(dataset[:300])=300
len(dataset[0:300])=300
len(dataset.select(range(300)))=300
```
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-5.15.0-60-generic-x86_64-with-glibc2.35
- Python version: 3.10.11
- `huggingface_hub` version: 0.20.2
- PyArrow version: 10.0.1
- Pandas version: 1.5.3
- `fsspec` version: 2023.10.0 | 6,748 |
https://github.com/huggingface/datasets/issues/6746 | ExpectedMoreSplits error when loading C4 dataset | [
"Hi ! We updated the `allenai/c4` repository to allow people to specify which language to load easily (the the [c4 dataset page](https://huggingface.co/datasets/allenai/c4))\r\n\r\nTo fix this issue **you can update** `datasets` and remove the mention of the legacy configuration name \"allenai--c4\":\r\n\r\n```python\r\ntraindata = load_dataset('allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train')\r\nvaldata = load_dataset('allenai/c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation')\r\n```",
"Did you solve this problemοΌI have the same bug.It is no use to delete \"allenai--c4\".",
"Did you solve it? I met this problem too.",
"But after I romove allenai--c4,it still fails",
"For me it works this way. I'm using datasets version 2.17.0",
"First, pip install --upgrade datasets.\r\nSecond, Update the following two lines of code in data.py (in lib)\r\ntraindata = load_dataset('allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train')\r\nvaldata = load_dataset('allenai/c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation')"
] | ### Describe the bug
I encounter bug when running the example command line
```python
python main.py \
--model decapoda-research/llama-7b-hf \
--prune_method wanda \
--sparsity_ratio 0.5 \
--sparsity_type unstructured \
--save out/llama_7b/unstructured/wanda/
```
The bug occurred at these lines of code (when loading c4 dataset)
```python
traindata = load_dataset('allenai/c4', 'allenai--c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train')
valdata = load_dataset('allenai/c4', 'allenai--c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation')
```
The error message states:
```
raise ExpectedMoreSplits(str(set(expected_splits) - set(recorded_splits)))
datasets.utils.info_utils.ExpectedMoreSplits: {'validation'}
```
### Steps to reproduce the bug
1. I encounter bug when running the example command line
### Expected behavior
The error message states:
```
raise ExpectedMoreSplits(str(set(expected_splits) - set(recorded_splits)))
datasets.utils.info_utils.ExpectedMoreSplits: {'validation'}
```
### Environment info
I'm using cuda 12.4, so I use ```pip install pytorch``` instead of conda provided in install.md
Also, I've tried another environment using the same commands in install.md, but the same bug occured | 6,746 |
https://github.com/huggingface/datasets/issues/6745 | Scraping the whole of github including private repos is bad; kindly stop | [
"It's not twitter here"
] | ### Feature request
https://github.com/bigcode-project/opt-out-v2 - opt out is not consent. kindly quit this ridiculous nonsense.
### Motivation
[EDITED: insults not tolerated]
### Your contribution
[EDITED: insults not tolerated] | 6,745 |
https://github.com/huggingface/datasets/issues/6744 | Option to disable file locking | [] | ### Feature request
Commands such as `load_dataset` creates file locks with `filelock.FileLock`. It would be good if there was a way to disable this.
### Motivation
File locking doesn't work on all file-systems (in my case NFS mounted Weka). If the `cache_dir` only had small files then it would be possible to point to local disk and the problem would be solved. However, as cache_dir is both where the small info files are written and the processed datasets are put this isn't a feasible solution.
Considering https://github.com/huggingface/datasets/issues/6395 I still do think this is something that belongs in HuggingFace. The possibility to control packages separately is valuable. It might be that a user has their dataset on a file-system that doesn't support file-locking while they are using file locking on local disk to control some other type of access.
### Your contribution
My suggested solution:
```
diff --git a/src/datasets/utils/_filelock.py b/src/datasets/utils/_filelock.py
index 19620e6e..58f41a02 100644
--- a/src/datasets/utils/_filelock.py
+++ b/src/datasets/utils/_filelock.py
@@ -18,11 +18,15 @@
import os
from filelock import FileLock as FileLock_
-from filelock import UnixFileLock
+from filelock import SoftFileLock, UnixFileLock
from filelock import __version__ as _filelock_version
from packaging import version
+if os.getenv('HF_USE_SOFTFILELOCK', 'false').lower() in ('true', '1'):
+ FileLock_ = SoftFileLock
+
+
class FileLock(FileLock_):
"""
A `filelock.FileLock` initializer that handles long paths.
```
| 6,744 |
https://github.com/huggingface/datasets/issues/6740 | Support for loading geotiff files as a part of the ImageFolder | [] | ### Feature request
Request for adding rasterio support to load geotiff as a part of ImageFolder, instead of using PIL
### Motivation
As of now, there are many datasets in HuggingFace Hub which are predominantly focussed towards RemoteSensing or are from RemoteSensing. The current ImageFolder (if I have understood correctly) uses PIL. This is not really optimized because mostly these datasets have images with many channels and additional metadata. Using PIL makes one loose it unless we provide a custom script. Hence, maybe an API could be added to have this in common?
### Your contribution
If the issue is accepted - i can contribute the code, because I would like to have it automated and generalised. | 6,740 |
https://github.com/huggingface/datasets/issues/6738 | Dict feature is non-nullable while nested dict feature is | [
"It looks like a bug, by default every feature should be nullable.",
"I've linked a PR with a fix :)",
"@mariosasko awesome thank you!"
] | When i try to create a `Dataset` object with None values inside a dict column, like this:
```python
from datasets import Dataset, Features, Value
Dataset.from_dict(
{
"dict": [{"a": 0, "b": 0}, None],
}, features=Features(
{"dict": {"a": Value("int16"), "b": Value("int16")}}
)
)
```
i get `ValueError: Got None but expected a dictionary instead`.
At the same time, having None in _nested_ dict feature works, for example, this doesn't throw any errors:
```python
from datasets import Dataset, Features, Value, Sequence
dataset = Dataset.from_dict(
{
"list_dict": [[{"a": 0, "b": 0}], None],
"sequence_dict": [[{"a": 0, "b": 0}], None],
}, features=Features({
"list_dict": [{"a": Value("int16"), "b": Value("int16")}],
"sequence_dict": Sequence({"a": Value("int16"), "b": Value("int16")}),
})
)
```
Other types of features also seem to be nullable (but I haven't checked all of them).
Version of `datasets` is the latest atm (2.18.0)
Is this an expected behavior or a bug? | 6,738 |
https://github.com/huggingface/datasets/issues/6737 | Invalid pattern: '**' can only be an entire path component | [
"I couldn't reproduce the issue on my side on MacOS, I guess the issue comes from the recent `fsspec` on Windows.\r\n\r\nCan you try downgrading to `fsspec==2023.9.2` for now ? It would also be great to investigate this and see if we need a fix in `datasets` or `fsspec`",
"I had the same issue! \r\nDowngrading to fsspec from 2023.10.0 to 2023.9.2 solved it for me.\r\n\r\n(env: python 3.11.7, datasets version: 2.15.0, Windows 10 22H2, Build 19045.4170)\r\n\r\nThanks a lot!",
"Ubuntu 20.04 had the same issue\r\npython 3.9 \r\n\r\nFile \"/home/delight-gpu/Workspace2/azuryl/FLAP/main.py\", line 112, in <module>\r\n main()\r\n File \"/home/delight-gpu/Workspace2/azuryl/FLAP/main.py\", line 85, in main\r\n prune_flap(args, model, tokenizer, device)\r\n File \"/home/delight-gpu/Workspace2/azuryl/FLAP/lib/prune.py\", line 294, in prune_flap\r\n dataloader, _ = get_loaders(\"wikitext2\", nsamples=args.nsamples,seed=args.seed,seqlen=model.seqlen,tokenizer=tokenizer)\r\n File \"/home/delight-gpu/Workspace2/azuryl/FLAP/lib/data.py\", line 159, in get_loaders\r\n return get_wikitext2(nsamples, seed, seqlen, tokenizer)\r\n File \"/home/delight-gpu/Workspace2/azuryl/FLAP/lib/data.py\", line 79, in get_wikitext2\r\n traindata = load_dataset('wikitext', 'wikitext-2-raw-v1', split='train')\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/load.py\", line 1767, in load_dataset\r\n builder_instance = load_dataset_builder(\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/load.py\", line 1498, in load_dataset_builder\r\n dataset_module = dataset_module_factory(\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/load.py\", line 1215, in dataset_module_factory\r\n raise e1 from None\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/load.py\", line 1192, in dataset_module_factory\r\n return HubDatasetModuleFactoryWithoutScript(\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/load.py\", line 765, in get_module\r\n else get_data_patterns_in_dataset_repository(hfh_dataset_info, self.data_dir)\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/data_files.py\", line 675, in get_data_patterns_in_dataset_repository\r\n return _get_data_files_patterns(resolver)\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/data_files.py\", line 236, in _get_data_files_patterns\r\n data_files = pattern_resolver(pattern)\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/datasets/data_files.py\", line 486, in _resolve_single_pattern_in_dataset_repository\r\n glob_iter = [PurePath(filepath) for filepath in fs.glob(PurePath(pattern).as_posix()) if fs.isfile(filepath)]\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/fsspec/spec.py\", line 606, in glob\r\n pattern = glob_translate(path + (\"/\" if ends_with_sep else \"\"))\r\n File \"/home/azuryl/anaconda3/envs/flap/lib/python3.9/site-packages/fsspec/utils.py\", line 734, in glob_translate\r\n raise ValueError(\r\nValueError: Invalid pattern: '**' can only be an entire path component",
"on ubuntu you just need to have the latest `datasets` and `fsspec`\r\n\r\n```\r\npip install -U datasets fsspec\r\n```",
"The issue was caused by an incompatibility between the versions of `datasets`, `huggingface-hub` and `fsspec`.\r\n\r\nThe issue was fixed in:\r\n- huggingface-hub-0.21.2: https://github.com/huggingface/huggingface_hub/pull/2056\r\n- and datasets-2.18.0: https://github.com/huggingface/datasets/pull/6687\r\n - datasets-2.19.1 fixed the minimum requirement huggingface-hub >= 0.21.2: https://github.com/huggingface/datasets/pull/6713",
"@albertvillanova, thank you for this solution. I encountered the same issue and had to use:\r\n```\r\nconda install -c conda-forge huggingface_hub=0.21.2 datasets=2.19.1\r\n```\r\n\r\nCheers",
"CheersοΌ"
] | ### Describe the bug
ValueError: Invalid pattern: '**' can only be an entire path component
when loading any dataset
### Steps to reproduce the bug
import datasets
ds = datasets.load_dataset("TokenBender/code_instructions_122k_alpaca_style")
### Expected behavior
loading the dataset successfully
### Environment info
- `datasets` version: 2.18.0
- Platform: Windows-10-10.0.22631-SP0
- Python version: 3.11.7
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.1
- `fsspec` version: 2023.12.2 | 6,737 |
https://github.com/huggingface/datasets/issues/6736 | Mosaic Streaming (MDS) Support | [
"Hi ! that would be great :) Though note that `datasets` doesn't implement format-specific resuming when streaming, so in general I think it's better if users can use the mosaic-streaming library to read their MDS datasets. I wonder if they support `hf://` paths though...\r\n\r\nAnyway for those interested, the code for WebDataset is a single file here: https://github.com/huggingface/datasets/blob/main/src/datasets/packaged_modules/webdataset/webdataset.py.\r\n\r\nIt implements `_split_generators` that downloads files and returns the lists of splits (train/validation/test) and `_split_generators` to generate examples (dicts) from the downloaded files. Streaming is automatically supported by making download steps lazy and by extending `open()` to work with remote URLs."
] | ### Feature request
I'm a huge fan of the current HF Datasets `webdataset` integration (especially the built-in streaming support). However, I'd love to upload some robotics and multimodal datasets I've processed for use with [Mosaic Streaming](https://docs.mosaicml.com/projects/streaming/en/stable/), specifically their [MDS Format](https://docs.mosaicml.com/projects/streaming/en/stable/fundamentals/dataset_format.html#mds).
Because the shard files have similar semantics to WebDataset, I'm hoping that adding such support won't be too much trouble?
### Motivation
One of the downsides with WebDataset is a lack of out-of-the-box determinism (especially for large-scale training and reproducibility), easy job resumption, and the ability to quickly debug / visualize individual examples.
Mosaic Streaming provides a [great interface for this out of the box](https://docs.mosaicml.com/projects/streaming/en/stable/#key-features), so I'd love to see it supported in HF Datasets.
### Your contribution
Happy to help test things / provide example data. Can potentially submit a PR if maintainers could point me to the necessary WebDataset logic / steps for adding a new streaming format! | 6,736 |
https://github.com/huggingface/datasets/issues/6734 | Tokenization slows towards end of dataset | [
"Hi ! First note that if the dataset is not heterogeneous / shuffled, there might be places in the data with shorter texts that are faster to tokenize.\r\n\r\nMoreover, the way `num_proc` works is by slicing the dataset and passing each slice to a process to run the `map()` function. So at the very end of `map()`, some processes might have finished transforming their slice of data while others are still running, causing the throughput to become lower.",
"I did see some comments about how num_proc=None could help and outputting numpy arrays can also help in the docs, but this seems quite odd now dropping down to 1it/s\r\n\r\n```bash\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46048888/46390354 [12:33:30<4:20:32, 21.84 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46049888/46390354 [12:36:11<8:37:59, 10.95 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46050888/46390354 [12:46:35<24:56:56, 3.78 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46051888/46390354 [12:56:43<35:08:10, 2.68 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46052888/46390354 [13:06:58<42:05:41, 2.23 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46053888/46390354 [13:16:01<44:40:18, 2.09 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46054888/46390354 [13:25:11<46:35:28, 2.00 examples/s]\r\nRunning tokenizer on dataset (num_proc=48): 99%|ββββββββββ| 46055888/46390354 [13:34:23<47:55:34, 1.94 examples/s]\r\n```\r\n\r\n",
"@ethansmith2000 Hi, did you solve this problem? I'm strugging with the same problem now."
] | ### Describe the bug
Mapped tokenization slows down substantially towards end of dataset.
train set started off very slow, caught up to 20k then tapered off til the end.
what's particularly strange is that the tokenization crashed a few times before due to errors with invalid tokens somewhere or corrupted downloads, and the speed ups/downs consistently happened the same times
```bash
Running tokenizer on dataset (num_proc=48): 0%| | 847000/881416735 [12:18<252:45:45, 967.72 examples/s]
Running tokenizer on dataset (num_proc=48): 0%| | 848000/881416735 [12:19<224:16:10, 1090.66 examples/s]
Running tokenizer on dataset (num_proc=48): 10%|β | 84964000/881416735 [3:48:00<11:21:34, 19476.01 examples/s]
Running tokenizer on dataset (num_proc=48): 10%|β | 84967000/881416735 [3:48:00<12:04:01, 18333.79 examples/s]
Running tokenizer on dataset (num_proc=48): 61%|ββββββ | 538631977/881416735 [13:46:40<27:50:04, 3420.84 examples/s]
Running tokenizer on dataset (num_proc=48): 61%|ββββββ | 538632977/881416735 [13:46:40<23:48:20, 3999.77 examples/s]
Running tokenizer on dataset (num_proc=48): 100%|ββββββββββ| 881365886/881416735 [38:30:19<04:34, 185.10 examples/s]
Running tokenizer on dataset (num_proc=48): 100%|ββββββββββ| 881366886/881416735 [38:30:25<04:36, 180.57 examples/s]
```
and validation set as well
```bash
Running tokenizer on dataset (num_proc=48): 90%|βββββββββ | 41544000/46390354 [28:44<02:37, 30798.76 examples/s]
Running tokenizer on dataset (num_proc=48): 90%|βββββββββ | 41550000/46390354 [28:44<02:08, 37698.08 examples/s]
Running tokenizer on dataset (num_proc=48): 96%|ββββββββββ| 44747422/46390354 [2:15:48<12:22:44, 36.87 examples/s]
Running tokenizer on dataset (num_proc=48): 96%|ββββββββββ| 44747422/46390354 [2:16:00<12:22:44, 36.87 examples/s]
```
### Steps to reproduce the bug
using the following kwargs
```python
with accelerator.main_process_first():
lm_datasets = tokenized_datasets.map(
group_texts,
batched=True,
num_proc=48
load_from_cache_file=True,
desc=f"Grouping texts in chunks of {block_size}",
)
```
running through slurm script
```bash
#SBATCH --partition=gpu-nvidia-a100
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --gpus-per-task=8
#SBATCH --cpus-per-task=96
```
using this dataset https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T
### Expected behavior
Constant speed throughout
### Environment info
- `datasets` version: 2.15.0
- Platform: Linux-5.15.0-1049-aws-x86_64-with-glibc2.10
- Python version: 3.8.18
- `huggingface_hub` version: 0.19.4
- PyArrow version: 14.0.1
- Pandas version: 2.0.3
- `fsspec` version: 2023.10.0 | 6,734 |
https://github.com/huggingface/datasets/issues/6733 | EmptyDatasetError when loading dataset downloaded with HuggingFace cli | [
"Hi! `datasets` is not compatible with `huggingface_hub`'s cache structure, hence the error.\r\n\r\nYou can track https://github.com/huggingface/datasets/issues/5080 to get notified when this is implemented."
] | ### Describe the bug
I am using a cluster that does not have access to the internet when given a job. I tried downloading the dataset using the huggingface-cli command and then loading it with load_dataset but I get an error:
```raise EmptyDatasetError(f"The directory at {base_path} doesn't contain any data files") from None```
The dataset I'm using is "lmsys/chatbot_arena_conversations". The folder structure is
- README.md
- data
- train-00000-of-00001-cced8514c7ed782a.parquet
### Steps to reproduce the bug
1. Download dataset using HuggingFace CLI: ```huggingface-cli download lmsys/chatbot_arena_conversations --local-dir ./lmsys/chatbot_arena_conversations```
2. In Python
```
from datasets import load_dataset
load_dataset("lmsys/chatbot_arena_conversations")
```
### Expected behavior
Should return a Dataset Dict in the form of
```
DatasetDict({
train: Dataset({
features: [...],
num_rows: 33,000
})
})
```
### Environment info
Python 3.11.5
Datasets 2.18.0
Transformers 4.38.2
Pytorch 2.2.0
Pyarrow 15.0.1
Rocky Linux release 8.9 (Green Obsidian)
| 6,733 |
https://github.com/huggingface/datasets/issues/6731 | Unexpected behavior when using load_dataset with streaming=True in a for loop | [
"This is normal behavior in python when using `lambda`: the `i` defined in your `lambda` refers to the global variable `i` in your loop, and `i` equals to `1` when you run your `for e in res[0]` line.\r\n\r\nYou should pass `fn_kwargs` that will be passed to your `lambda` instead of using the global variable:\r\n\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nres=[]\r\nfor i in [0,1]:\r\n di = load_dataset(\r\n \"json\", \r\n data_files='path_to.json', \r\n split='train',\r\n streaming=True, \r\n ).map(lambda x, source: {\"source\": source}, fn_kwargs={\"source\": i})\r\n\r\n res.append(di)\r\n\r\nfor e in res[0]:\r\n print(e)\r\n```\r\n\r\nThis doesn't happen in non-streaming since in that case `map` is executed while the variable `i` has the right value. In streaming mode, `map` is executed on-the-fly when you iterate on the dataset.",
"Thank you very much for your answer. I think this issue can be closed now."
] | ### Describe the bug
### My Code
```
from datasets import load_dataset
res=[]
for i in [0,1]:
di=load_dataset(
"json",
data_files='path_to.json',
split='train',
streaming=True,
).map(lambda x: {"source": i})
res.append(di)
for e in res[0]:
print(e)
```
### Unexpected Behavior
Data in `res[0]` has `source=1`. However the expected value is 0.
### FYI
I further switch `streaming` to `False`. And the output value is as expected (0). So there may exist bugs in setting `streaming=True` in a for loop.
### Environment
Python 3.8.0
datasets==2.18.0
transformers==4.28.1
### Steps to reproduce the bug
1. Create a Json file with any content.
2. Run the provided code.
3. Switch `streaming` to `False` and run again to see the expected behavior.
### Expected behavior
The expected behavior is the data are mapped with its corresponding value in the for loop.
### Environment info
Python 3.8.0
datasets==2.18.0
transformers==4.28.1
Ubuntu 20.04 | 6,731 |
https://github.com/huggingface/datasets/issues/6729 | Support zipfiles that span multiple disks? | [
"@severo were you able to solve it?",
"No. cc @albertvillanova @lhoestq @polinaeterna for an evaluation of what it would take to support this feature.",
"The underlying issue issue is that the dataset repository has used split ZIP archive files: https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream/tree/main/data\r\n```\r\ndownstream_dataset_patches_npzip.z01\r\ndownstream_dataset_patches_npzip.z02\r\n...\r\ndownstream_dataset_patches_npzip.zip\r\n```\r\nand these are not supported by the Python standard library package `zipfile`.",
"It's a pretty bad way to share a dataset since one needs to download the full dataset to use it.\r\n\r\nWe likely won't support this format.",
"I agree it is a format we maybe should not support: streaming is not possible.",
"I opened a PR in the reported repo to disable the viewer: https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream/discussions/1"
] | See https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream
The dataset viewer gives the following error:
```
Error code: ConfigNamesError
Exception: BadZipFile
Message: zipfiles that span multiple disks are not supported
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/dataset/config_names.py", line 67, in compute_config_names_response
get_dataset_config_names(
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/inspect.py", line 347, in get_dataset_config_names
dataset_module = dataset_module_factory(
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1871, in dataset_module_factory
raise e1 from None
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1846, in dataset_module_factory
return HubDatasetModuleFactoryWithoutScript(
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 1240, in get_module
module_name, default_builder_kwargs = infer_module_for_data_files(
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 584, in infer_module_for_data_files
split_modules = {
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 585, in <dictcomp>
split: infer_module_for_data_files_list(data_files_list, download_config=download_config)
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 526, in infer_module_for_data_files_list
return infer_module_for_data_files_list_in_archives(data_files_list, download_config=download_config)
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/load.py", line 554, in infer_module_for_data_files_list_in_archives
for f in xglob(extracted, recursive=True, download_config=download_config)[
File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/download/streaming_download_manager.py", line 576, in xglob
fs, *_ = fsspec.get_fs_token_paths(urlpath, storage_options=storage_options)
File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/core.py", line 622, in get_fs_token_paths
fs = filesystem(protocol, **inkwargs)
File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/registry.py", line 290, in filesystem
return cls(**storage_options)
File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/spec.py", line 79, in __call__
obj = super().__call__(*args, **kwargs)
File "/src/services/worker/.venv/lib/python3.9/site-packages/fsspec/implementations/zip.py", line 57, in __init__
self.zip = zipfile.ZipFile(
File "/usr/local/lib/python3.9/zipfile.py", line 1266, in __init__
self._RealGetContents()
File "/usr/local/lib/python3.9/zipfile.py", line 1329, in _RealGetContents
endrec = _EndRecData(fp)
File "/usr/local/lib/python3.9/zipfile.py", line 286, in _EndRecData
return _EndRecData64(fpin, -sizeEndCentDir, endrec)
File "/usr/local/lib/python3.9/zipfile.py", line 232, in _EndRecData64
raise BadZipFile("zipfiles that span multiple disks are not supported")
zipfile.BadZipFile: zipfiles that span multiple disks are not supported
```
The files (https://huggingface.co/datasets/PhilEO-community/PhilEO-downstream/tree/main/data) are:
<img width="629" alt="Capture dβeΜcran 2024-03-11 aΜ 22 07 30" src="https://github.com/huggingface/datasets/assets/1676121/0bb15a51-d54f-4d73-8572-e427ea644b36">
| 6,729 |
https://github.com/huggingface/datasets/issues/6728 | Issue Downloading Certain Datasets After Setting Custom `HF_ENDPOINT` | [
"Through debugging, I found a potential solution is to modify the code in the error handling module of `huggingface_hub`: https://github.com/huggingface/huggingface_hub/commit/56d6c798c44e83d2a3167e74c022737d8fcbe822 ",
"@Wauplin ",
"Thanks for investigating and reporting the bug @padeoe! I've opened a PR in `huggingface_hub` with your suggested fix! :) https://github.com/huggingface/huggingface_hub/pull/2119"
] | ### Describe the bug
This bug is triggered under the following conditions:
- datasets repo ids without organization names trigger errors, such as `bookcorpus`, `gsm8k`, `wikipedia`, rather than in the form of `A/B`.
- If `HF_ENDPOINT` is set and the hostname is not in the form of `(hub-ci.)?huggingface.co`.
- This issue occurs with `datasets>2.15.0` or `huggingface-hub>0.19.4`. For example, using the latest versions: `datasets==2.18.0` and `huggingface-hub==0.21.4`,
### Steps to reproduce the bug
the issue can be reproduced with the following code:
1. install specific datasets and huggingface_hub.
```bash
pip install datasets==2.18.0
pip install huggingface_hub==0.21.4
```
2. execute python code.
```Python
import os
os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com'
from datasets import load_dataset
bookcorpus = load_dataset('bookcorpus', split='train')
```
console output:
```
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 2556, in load_dataset
builder_instance = load_dataset_builder(
File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 2228, in load_dataset_builder
dataset_module = dataset_module_factory(
File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 1879, in dataset_module_factory
raise e1 from None
File "/home/padeoe/.local/lib/python3.10/site-packages/datasets/load.py", line 1830, in dataset_module_factory
with fs.open(f"datasets/{path}/{filename}", "r", encoding="utf-8") as f:
File "/home/padeoe/.local/lib/python3.10/site-packages/fsspec/spec.py", line 1295, in open
self.open(
File "/home/padeoe/.local/lib/python3.10/site-packages/fsspec/spec.py", line 1307, in open
f = self._open(
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 228, in _open
return HfFileSystemFile(self, path, mode=mode, revision=revision, block_size=block_size, **kwargs)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 615, in __init__
self.resolved_path = fs.resolve_path(path, revision=revision)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 180, in resolve_path
repo_and_revision_exist, err = self._repo_and_revision_exist(repo_type, repo_id, revision)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_file_system.py", line 117, in _repo_and_revision_exist
self._api.repo_info(repo_id, revision=revision, repo_type=repo_type)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 118, in _inner_fn
return fn(*args, **kwargs)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2413, in repo_info
return method(
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 118, in _inner_fn
return fn(*args, **kwargs)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2286, in dataset_info
hf_raise_for_status(r)
File "/home/padeoe/.local/lib/python3.10/site-packages/huggingface_hub/utils/_errors.py", line 362, in hf_raise_for_status
raise HfHubHTTPError(str(e), response=response) from e
huggingface_hub.utils._errors.HfHubHTTPError: 401 Client Error: Unauthorized for url: https://hf-mirror.com/api/datasets/bookcorpus/bookcorpus.py (Request ID: Root=1-65ee8659-5ab10eec5960c63e71f2bb58;b00bdbea-fd6e-4a74-8fe0-bc4682ae090e)
```
### Expected behavior
The dataset was downloaded correctly without any errors.
### Environment info
datasets==2.18.0
huggingface-hub==0.21.4 | 6,728 |
https://github.com/huggingface/datasets/issues/6726 | Profiling for HF Filesystem shows there are easy performance gains to be made | [
"FWIW I debugged this while waiting for it to go",
"Oh I forgot to mention you can also cache resolve_pattern, and that seemed to also substantially improves things, if you want to load a dataset twice for whatever reason."
] | ### Describe the bug
# Let's make it faster
First, an evidence...
Figure 1: CProfile for loading 3 files from cerebras/SlimPajama-627B train split, and 3 files from test split using streaming=True. X axis is 1106 seconds long.
See? It's pretty slow.
What is resolve pattern doing?
```
resolve_pattern called with **/train/** and hf://datasets/cerebras/SlimPajama-627B@2d0accdd58c5d5511943ca1f5ff0e3eb5e293543
resolve_pattern took 20.815081119537354 seconds
```
Makes sense. How to improve it?
## Bigger project, biggest payoff
Databricks (and consequently, spark) store a compressed manifest file of the files contained in the remote filesystem.
Then, you download one tiny file, decompress it, and all the operations are local instead of this shenanigans.
It seems pretty straightforward to make dataset uploads compute a manifest and upload it alongside their data.
This would make resolution time so fast that nobody would ever think about it again.
It also means you either need to have the uploader compute it _every time_, or have a hook that computes it.
## Smaller project, immediate payoff: Be diligent in avoiding deepcopy
Revise the _ls_tree method to avoid deepcopy:
```
def _ls_tree(
self,
path: str,
recursive: bool = False,
refresh: bool = False,
revision: Optional[str] = None,
expand_info: bool = True,
):
..... omitted .....
for path_info in tree:
if isinstance(path_info, RepoFile):
cache_path_info = {
"name": root_path + "/" + path_info.path,
"size": path_info.size,
"type": "file",
"blob_id": path_info.blob_id,
"lfs": path_info.lfs,
"last_commit": path_info.last_commit,
"security": path_info.security,
}
else:
cache_path_info = {
"name": root_path + "/" + path_info.path,
"size": 0,
"type": "directory",
"tree_id": path_info.tree_id,
"last_commit": path_info.last_commit,
}
parent_path = self._parent(cache_path_info["name"])
self.dircache.setdefault(parent_path, []).append(cache_path_info)
out.append(cache_path_info)
return copy.deepcopy(out) # copy to not let users modify the dircache
```
Observe this deepcopy at the end. It is making a copy of a very simple data structure. We do not need to copy. We can simply generate the data structure twice instead. It will be much faster.
```
def _ls_tree(
self,
path: str,
recursive: bool = False,
refresh: bool = False,
revision: Optional[str] = None,
expand_info: bool = True,
):
..... omitted .....
def make_cache_path_info(path_info):
if isinstance(path_info, RepoFile):
return {
"name": root_path + "/" + path_info.path,
"size": path_info.size,
"type": "file",
"blob_id": path_info.blob_id,
"lfs": path_info.lfs,
"last_commit": path_info.last_commit,
"security": path_info.security,
}
else:
return {
"name": root_path + "/" + path_info.path,
"size": 0,
"type": "directory",
"tree_id": path_info.tree_id,
"last_commit": path_info.last_commit,
}
for path_info in tree:
cache_path_info = make_cache_path_info(path_info)
out_cache_path_info = make_cache_path_info(path_info) # copy to not let users modify the dircache
parent_path = self._parent(cache_path_info["name"])
self.dircache.setdefault(parent_path, []).append(cache_path_info)
out.append(out_cache_path_info)
return out
```
Note there is no longer a deepcopy in this method. We have replaced it with generating the output twice. This is substantially faster. For me, the entire resolution went from 1100s to 360s.
## Medium project, medium payoff
After the above change, we have this profile:
Figure 2: x-axis is 355 seconds. Note that globbing and _ls_tree deep copy is gone. No surprise there. It's much faster now, but we still spend ~187seconds in get_fs_token_paths.
Well get_fs_token_paths is part of fsspec. We don't need to fix that because we can trust their developers to write high performance code. Probably the caller has misconfigured something. Let's take a look at the storage_options being provided to the filesystem that is constructed during this call.
Ah yes, streaming_download_manager::_prepare_single_hop_path_and_storage_options. We know streaming download manager is not compatible with async right now, but we really need this specific part of the code to be async. We're spending so much time checking isDir on the remote filesystem, it's a huge waste.
We can make the call easily 20-30x faster by using async, removing this performance bottleneck almost entirely (and reducing the total time of this part of the code to <30s. There is no reason to block async isDir calls for streaming.
I'm not going to mess w/ this one myself; I didn't write the streaming impl, and I don't know how it works, but I know the isDir check can be async.
### Steps to reproduce the bug
```
with cProfile.Profile() as pr:
pr.enable()
# Begin Data
if not os.path.exists(data_cache_dir):
os.makedirs(data_cache_dir, exist_ok=True)
training_dataset = load_dataset(training_dataset_name, split=training_split, cache_dir=data_cache_dir, streaming=True).take(training_slice)
eval_dataset = load_dataset(eval_dataset_name, split=eval_split, cache_dir=data_cache_dir, streaming=True).take(eval_slice)
# End Data
pr.disable()
pr.create_stats()
if not os.path.exists(profiling_path):
os.makedirs(profiling_path, exist_ok=True)
pr.dump_stats(os.path.join(profiling_path, "cprofile.prof"))
```
run this code for "cerebras/SlimPajama-627B" and whatever other params
### Expected behavior
Something better.
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-5.15.146.1-microsoft-standard-WSL2-x86_64-with-glibc2.35
- Python version: 3.10.13
- `huggingface_hub` version: 0.21.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.1
- `fsspec` version: 2024.2.0 | 6,726 |
https://github.com/huggingface/datasets/issues/6725 | Request for a comparison of huggingface datasets compared with other data format especially webdataset | [] | ### Feature request
Request for a comparison of huggingface datasets compared with other data format especially webdataset
### Motivation
I see huggingface datasets uses Apache Arrow as its backend, it seems to be great, but I'm curious about how it is good compared with other dataset format, like webdataset, what's the pros/cons of them.
### Your contribution
More information | 6,725 |
https://github.com/huggingface/datasets/issues/6724 | Dataset with loading script does not work in renamed repos | [] | ### Describe the bug
My data repository was first called `BramVanroy/hplt-mono-v1-2` but I then renamed to use underscores instead of dashes. However, it seems that `datasets` retrieves the old repo name when it checks whether the repo contains data loading scripts in this line.
https://github.com/huggingface/datasets/blob/6fb6c834f008996c994b0a86c3808d0a33d44525/src/datasets/load.py#L1845
When I print `filename` it returns `hplt-mono-v1-2.py` but the files in the repo are of course `['.gitattributes', 'README.md', 'hplt_mono_v1_2.py']`. So the `filename` is the original reponame instead of the renamed one.
I am not sure if this is a caching issue or not or how I can resolve it.
### Steps to reproduce the bug
```
from datasets import load_dataset
ds = load_dataset(
"BramVanroy/hplt-mono-v1-2",
"ky",
trust_remote_code=True
)
```
### Expected behavior
That the most recent repo name is used when `filename` is generated.
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-5.14.0-284.25.1.el9_2.x86_64-x86_64-with-glibc2.34
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.1
- Pandas version: 2.1.3
- `fsspec` version: 2023.10.0
| 6,724 |
https://github.com/huggingface/datasets/issues/6721 | Hi,do you know how to load the dataset from local file now? | [
"\r\n@Gera001\r\n# Loading Dataset from Local Files Using π€Hugging Face.\r\n\r\nTo load a dataset from local files using the Hugging Face datasets library, you can use the `load_dataset` function.\r\n\r\n```\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('csv', data_files={'train': 'path/to/train.csv',\r\n 'test': 'path/to/test.csv'})\r\n```\r\n\r\nReference to [HF Datasets docs for loading from local](https://huggingface.co/docs/datasets/en/loading#csv). \r\n\r\n@albertvillanova\r\nthis issue can be closed here.",
"like this: from datasets import load_from_disk\r\ndataset = load_from_disk(data_path)\r\n",
"@ge00009 \r\n> like this: from datasets import load_from_disk dataset = load_from_disk(data_path)\r\n\r\nLoads a dataset that was previously saved using `save_to_disk()`.\r\n\r\nReference link:\r\nhttps://huggingface.co/docs/datasets/en/package_reference/loading_methods#datasets.load_from_disk.example"
] | Hi, if I want to load the dataset from local file, then how to specify the configuration name?
_Originally posted by @WHU-gentle in https://github.com/huggingface/datasets/issues/2976#issuecomment-1333455222_
| 6,721 |
https://github.com/huggingface/datasets/issues/6720 | TypeError: 'str' object is not callable | [
"Hi ! I opened a PR to fix an issue in the Features defined in your code\r\n\r\nBasically changing\r\n```python\r\nSequence(\"float32\")\r\n```\r\n\r\nto\r\n```python\r\nSequence(Value(\"float32\"))\r\n```\r\n\r\n\r\nhttps://huggingface.co/datasets/BramVanroy/hplt_mono_v1_2/discussions/1",
"D'oh! Was wondering why the `str() is not callable` was in there. Glad the error is my end though, and not related to zstandard (which I had not used in the past).\r\n\r\nThanks a lot!"
] | ### Describe the bug
I am trying to get the HPLT datasets on the hub. Downloading/re-uploading would be too time- and resource consuming so I wrote [a dataset loader script](https://huggingface.co/datasets/BramVanroy/hplt_mono_v1_2/blob/main/hplt_mono_v1_2.py). I think I am very close but for some reason I always get the error below. It happens during the clean-up phase where the directory cannot be removed because it is not empty.
My only guess would be that this may have to do with zstandard
```
Traceback (most recent call last):
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1744, in _prepare_split_single
writer.write(example, key)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 492, in write
self.write_examples_on_file()
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 434, in write_examples_on_file
if self.schema
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 409, in schema
else (pa.schema(self._features.type) if self._features is not None else None)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1643, in type
return get_nested_type(self)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in get_nested_type
{key: get_nested_type(schema[key]) for key in schema}
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in <dictcomp>
{key: get_nested_type(schema[key]) for key in schema}
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1221, in get_nested_type
value_type = get_nested_type(schema.feature)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1228, in get_nested_type
return schema()
TypeError: 'str' object is not callable
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1753, in _prepare_split_single
num_examples, num_bytes = writer.finalize()
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 588, in finalize
self.write_examples_on_file()
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 434, in write_examples_on_file
if self.schema
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/arrow_writer.py", line 409, in schema
else (pa.schema(self._features.type) if self._features is not None else None)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1643, in type
return get_nested_type(self)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in get_nested_type
{key: get_nested_type(schema[key]) for key in schema}
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1209, in <dictcomp>
{key: get_nested_type(schema[key]) for key in schema}
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1221, in get_nested_type
value_type = get_nested_type(schema.feature)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/features/features.py", line 1228, in get_nested_type
return schema()
TypeError: 'str' object is not callable
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 959, in incomplete_dir
yield tmp_dir
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1005, in download_and_prepare
self._download_and_prepare(
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1767, in _download_and_prepare
super()._download_and_prepare(
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1100, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1605, in _prepare_split
for job_id, done, content in self._prepare_split_single(
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1762, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/home/pricie/vanroy/.config/JetBrains/PyCharm2023.3/scratches/scratch_5.py", line 4, in <module>
ds = load_dataset(
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/load.py", line 2549, in load_dataset
builder_instance.download_and_prepare(
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 985, in download_and_prepare
with incomplete_dir(self._output_dir) as tmp_output_dir:
File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/contextlib.py", line 153, in __exit__
self.gen.throw(typ, value, traceback)
File "/home/local/vanroy/dutch-instruction-datasets/.venv/lib/python3.10/site-packages/datasets/builder.py", line 966, in incomplete_dir
shutil.rmtree(tmp_dir)
File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/shutil.py", line 731, in rmtree
onerror(os.rmdir, path, sys.exc_info())
File "/home/pricie/vanroy/.pyenv/versions/3.10.13/lib/python3.10/shutil.py", line 729, in rmtree
os.rmdir(path)
OSError: [Errno 39] Directory not empty: '/home/pricie/vanroy/.cache/huggingface/datasets/BramVanroy___hplt_mono_v1_2/ky/1.2.0/7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete'
```
Interestingly, though, this directory _does_ appear to be empty:
```shell
> cd /home/pricie/vanroy/.cache/huggingface/datasets/BramVanroy___hplt_mono_v1_2/ky/1.2.0/7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete
> ls -lah
total 0
drwxr-xr-x. 1 vanroy vanroy 0 Mar 7 12:01 .
drwxr-xr-x. 1 vanroy vanroy 304 Mar 7 11:52 ..
> cd ..
> ls
7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47_builder.lock 7ab138629fe7e9e29fe93ce63d809d5ef9d963273b829f61ab538e012dc9cc47.incomplete
```
### Steps to reproduce the bug
```python
from datasets import load_dataset
ds = load_dataset(
"BramVanroy/hplt_mono_v1_2",
"ky",
trust_remote_code=True
)
```
### Expected behavior
No error.
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-5.14.0-284.25.1.el9_2.x86_64-x86_64-with-glibc2.34
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.1
- Pandas version: 2.1.3
- `fsspec` version: 2023.10.0
| 6,720 |
https://github.com/huggingface/datasets/issues/6719 | Is there any way to solve hanging of IterableDataset using split by node + filtering during inference | [] | ### Describe the bug
I am using an iterable dataset in a multi-node setup, trying to do training/inference while filtering the data on the fly. I usually do not use `split_dataset_by_node` but it is very slow using the IterableDatasetShard in `accelerate` and `transformers`. When I filter after applying `split_dataset_by_node`, it results in shards that are not equal sizes due to unequal samples filtered from each one.
The distributed process hangs when trying to accomplish this. Is there any way to resolve this or is it impossible to implement?
### Steps to reproduce the bug
Here is a toy example of what I am trying to do that reproduces the behavior
```
# torchrun --nproc-per-node 2 file.py
import os
import pandas as pd
import torch
from accelerate import Accelerator
from datasets import Features, Value, load_dataset
from datasets.distributed import split_dataset_by_node
from torch.utils.data import DataLoader
accelerator = Accelerator(device_placement=True, dispatch_batches=False)
if accelerator.is_main_process:
if not os.path.exists("scratch_data"):
os.mkdir("scratch_data")
n_shards = 4
for i in range(n_shards):
df = pd.DataFrame({"id": list(range(10 * i, 10 * (i + 1)))})
df.to_parquet(f"scratch_data/shard_{i}.parquet")
world_size = accelerator.num_processes
local_rank = accelerator.process_index
def collate_fn(examples):
input_ids = []
for example in examples:
input_ids.append(example["id"])
return torch.LongTensor(input_ids)
dataset = load_dataset(
"parquet", data_dir="scratch_data", split="train", streaming=True
)
dataset = (
split_dataset_by_node(dataset, rank=local_rank, world_size=world_size)
.filter(lambda x: x["id"] < 35)
.shuffle(seed=42, buffer_size=100)
)
batch_size = 2
train_dataloader = DataLoader(
dataset,
batch_size=batch_size,
collate_fn=collate_fn,
num_workers=2
)
for x in train_dataloader:
x = x.to(accelerator.device)
print({"rank": local_rank, "id": x})
y = accelerator.gather_for_metrics(x)
if accelerator.is_main_process:
print("gathered", y)
```
### Expected behavior
Is there any way to continue training/inference on the GPUs that have remaining data left without waiting for the others? Is it impossible to filter when
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-5.10.209-198.812.amzn2.x86_64-x86_64-with-glibc2.31
- Python version: 3.10.13
- `huggingface_hub` version: 0.21.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.1
- `fsspec` version: 2023.6.0 | 6,719 |
https://github.com/huggingface/datasets/issues/6717 | `remove_columns` method used with a streaming enable dataset mode produces a LibsndfileError on multichannel audio | [
"And it also works well with `dataset = dataset.select_columns([\"audio\"])`"
] | ### Describe the bug
When loading a HF dataset in streaming mode and removing some columns, it is impossible to load a sample if the audio contains more than one channel. I have the impression that the time axis and channels are swapped or concatenated.
### Steps to reproduce the bug
Minimal error code:
```python
from datasets import load_dataset
dataset_name = "zinc75/Vibravox_dummy"
config_name = "BWE_Larynx_microphone"
# if we use "ASR_Larynx_microphone" subset which is a monochannel audio, no error is thrown.
dataset = load_dataset(
path=dataset_name, name=config_name, split="train", streaming=True
)
dataset = dataset.remove_columns(["sensor_id"])
# dataset = dataset.map(lambda x:x, remove_columns=["sensor_id"])
# The commented version does not produce an error, but loses the dataset features.
sample = next(iter(dataset))
```
Error:
```
Traceback (most recent call last):
File "/home/julien/Bureau/github/vibravox/tmp.py", line 15, in <module>
sample = next(iter(dataset))
^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/iterable_dataset.py", line 1392, in __iter__
example = _apply_feature_types_on_example(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/iterable_dataset.py", line 1080, in _apply_feature_types_on_example
encoded_example = features.encode_example(example)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1889, in encode_example
return encode_nested_example(self, example)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1244, in encode_nested_example
{k: encode_nested_example(schema[k], obj.get(k), level=level + 1) for k in schema}
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1244, in <dictcomp>
{k: encode_nested_example(schema[k], obj.get(k), level=level + 1) for k in schema}
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/features.py", line 1300, in encode_nested_example
return schema.encode_example(obj) if obj is not None else None
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/datasets/features/audio.py", line 98, in encode_example
sf.write(buffer, value["array"], value["sampling_rate"], format="wav")
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 343, in write
with SoundFile(file, 'w', samplerate, channels,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 658, in __init__
self._file = self._open(file, mode_int, closefd)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/julien/.pyenv/versions/vibravox/lib/python3.11/site-packages/soundfile.py", line 1216, in _open
raise LibsndfileError(err, prefix="Error opening {0!r}: ".format(self.name))
soundfile.LibsndfileError: Error opening <_io.BytesIO object at 0x7fd795d24680>: Format not recognised.
Process finished with exit code 1
```
### Expected behavior
I would expect this code to run without error.
### Environment info
- `datasets` version: 2.18.0
- Platform: Linux-6.5.0-21-generic-x86_64-with-glibc2.35
- Python version: 3.11.0
- `huggingface_hub` version: 0.21.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.1
- `fsspec` version: 2023.10.0 | 6,717 |
https://github.com/huggingface/datasets/issues/6716 | Non-deterministic `Dataset.builder_name` value | [
"When `rotten_tomatoes` is printed out, the following warning message is also printed out:\r\n\r\n```\r\nYou can avoid this message in future by passing the argument `trust_remote_code=True`.\r\nPassing `trust_remote_code=True` will be mandatory to load this dataset from the next major release of `datasets`.\r\n```",
"Hi ! This behavior happens because the dataset was originakky created using a dataset script [rotten_tomatoes.py](https://huggingface.co/datasets/rotten_tomatoes/blob/26f40d324d7b281d8b3fb1c47f30f8b9957f206b/rotten_tomatoes.py) and because we added features recently allowing to download the dataset directly from Parquet files (parquet builder) without running the dataset script (rotten_tomatoes). The flakiness must come from the availability of the Parquet files (we automatically export them in the refs/convert/parquet branch and we recently had to move some files).\r\n\r\nAnyway the easy fix on our side is to remove the dataset script completely, let me open a PR at https://huggingface.co/datasets/rotten_tomatoes\r\n\r\nEDIT: opened https://huggingface.co/datasets/rotten_tomatoes/discussions/6, feel free to comment there if you're ok with that change",
"@lhoestq Thanks for the comment, explanation, and patch!",
"> we automatically export them in the refs/convert/parquet branch\r\n\r\nWhen this operation is in progress, the parquet files become temporarily unavailable?",
"> When this operation is in progress, the parquet files become temporarily unavailable?\r\n\r\nYes correct. I just merged the patch btw :)",
"@lhoestq Thanks for merging the PR! I think this issue can be closed."
] | ### Describe the bug
I'm not sure if this is a bug, but `print(ds.builder_name)` in the following code sometimes prints out `rotten_tomatoes` instead of `parquet`:
```python
import datasets
for _ in range(100):
ds = datasets.load_dataset("rotten_tomatoes", split="train")
print(ds.builder_name) # prints out "rotten_tomatoes" sometimes instead of "parquet"
```
Output:
```
...
parquet
parquet
parquet
rotten_tomatoes
parquet
parquet
parquet
...
```
Here's a reproduction using GitHub Actions:
https://github.com/mlflow/mlflow/actions/runs/8153247984/job/22284263613?pr=11329#step:12:241
One of our tests is flaky because `builder_name` is not deterministic.
### Steps to reproduce the bug
1. Run the code above.
### Expected behavior
Always prints out `parquet`?
### Environment info
```
Copy-and-paste the text below in your GitHub issue.
- `datasets` version: 2.18.0
- Platform: Linux-6.5.0-1015-azure-x86_64-with-glibc2.34
- Python version: 3.8.18
- `huggingface_hub` version: 0.21.3
- PyArrow version: 15.0.0
- Pandas version: 2.0.3
- `fsspec` version: 2024.2.0
``` | 6,716 |
https://github.com/huggingface/datasets/issues/6703 | Unable to load dataset that was saved with `save_to_disk` | [
"`save_to_disk` uses a special serialization that can only be read using `load_from_disk`.\r\n\r\nContrary to `load_dataset`, `load_from_disk` directly loads Arrow files and uses the dataset directory as cache.\r\n\r\nOn the other hand `load_dataset` does a conversion step to get Arrow files from the raw data files (could be in JSON, CSV, Parquet etc.) and caches them in the `datasets` cache directory (default is `~/.cache/huggingface/datasets`). We haven't implemented any logic in `load_dataset` to support datasets saved with `save_to_disk` because they don't use the same cache.\r\n\r\nEDIT: note that you can save your dataset in Parquet format locally using `.to.parquet()` (make sure to shard in multiple files your dataset if it's multiple GBs - you can use `.shard()` + `.to_parquet()` to do that) and you'll be able to reload it using `load_dataset`",
"@lhoestq, so is it correctly understood that if I run `to_parquet()` and then `save_to_disk()`, I can load it with `load_dataset`? If yes, then it would resolve this issue (and should probably be documented somewhere π)",
"Here is an example:\r\n```python\r\nds.to_parquet(\"my/local/dir/data.parquet\")\r\n\r\n# later\r\nds = load_dataset(\"my/local/dir\")\r\n```\r\n\r\nand for bigger datasets:\r\n```python\r\nnum_shards = 1024 # set number of files to save (e.g. try to have files smaller than 5GB)\r\nfor shard_idx in num_shards:\r\n shard = ds.shard(index=shard_idx, num_shards=num_shards)\r\n shard.to_parquet(f\"my/local/dir/{shard_idx:05d}.parquet\") # 00000.parquet to 01023.parquet\r\n\r\n# later\r\nds = load_dataset(\"my/local/dir\")\r\n```\r\n\r\n\r\nI hope this helps :)",
"Thanks for helping out! Does this approach work with `s3fs`? e.g. something like this:\r\n\r\n```python\r\nimport s3fs\r\ns3 = s3fs.S3FileSystem(anon=True)\r\nwith s3.open('mybucket/new-file.parquet', 'w') as f:\r\n ds.to_parquet(f)\r\n```\r\n\r\nThis is instead of `save_to_disk` to save to an S3 bucket.\r\n\r\nOtherwise, I am not sure how to make this work when saving the dataset to an S3 bucket. Would `dataset.set_format(\"arrow\")` work as a replacement?",
"`load_dataset` does't support S3 buckets unfortunately :/",
"> `load_dataset` does't support S3 buckets unfortunately :/\r\n\r\nI am aware but I have some code that downloads it to disk before using that method. The most important part is to store it in a format that load_dataset is compatible with. ",
"Feel free to use Parquet then :)",
"I ended up with this. Not ideal to save to local disk, but it works and loads via `load_datasets` after downloading from S3 with another method.\r\n\r\n```python\r\nwith tempfile.TemporaryDirectory() as dir:\r\n dataset_nbytes = ds._estimate_nbytes()\r\n max_shard_size_local = convert_file_size_to_int(max_shard_size)\r\n num_shards = int(dataset_nbytes / max_shard_size_local) + 1\r\n\r\n for shard_idx in range(num_shards):\r\n shard = ds.shard(index=shard_idx, num_shards=num_shards)\r\n shard.to_parquet(f\"{dir}/{shard_idx:05d}.parquet\")\r\n \r\n fs.upload(\r\n lpath=dir,\r\n rpath=s3_path,\r\n recursive=True,\r\n )\r\n```"
] | ### Describe the bug
I get the following error message: You are trying to load a dataset that was saved using `save_to_disk`. Please use `load_from_disk` instead.
### Steps to reproduce the bug
1. Save a dataset with `save_to_disk`
2. Try to load it with `load_datasets`
### Expected behavior
I am able to load the dataset again with `load_datasets` which most packages uses over `load_from_disk`. I want to have a workaround that allows me to create the same indexing that `push_to_hub` creates for you before using `save_to_disk` - how can that be achieved?
### Environment info
datasets 2.17.1, python 3.10 | 6,703 |
https://github.com/huggingface/datasets/issues/6702 | Push samples to dataset on hub without having the dataset locally | [
"Hi ! For now I would recommend creating a new Parquet file using `dataset_new.to_parquet()` and upload it to HF using `huggingface_hub` every time you get a new batch of data. You can name the Parquet files `0000.parquet`, `0001.parquet`, etc.\r\n\r\nThough maybe make sure to not upload one file per sample since that would be inefficient. You can buffer your data and upload when you have enough new samples for example",
"This is excellent, thanks!"
] | ### Feature request
Say I have the following code:
```
from datasets import Dataset
import pandas as pd
new_data = {
"column_1": ["value1", "value2"],
"column_2": ["value3", "value4"],
}
df_new = pd.DataFrame(new_data)
dataset_new = Dataset.from_pandas(df_new)
# add these samples to a remote dataset
```
It would be great to have a way to push dataset_new to a remote dataset that respects the same schema. This way one would not have to do the following:
```
from datasets import load_dataset
dataset = load_dataset('username/dataset_name', use_auth_token='your_hf_token_here')
updated_dataset = dataset['train'].concatenate(dataset_new)
updated_dataset.push_to_hub('username/dataset_name', use_auth_token='your_hf_token_here')
```
### Motivation
No need to download the dataset.
### Your contribution
Maybe this feature already exists, didnt see it though. I do not have the expertise to do this. | 6,702 |
https://github.com/huggingface/datasets/issues/6700 | remove_columns is not in-place but the doc shows it is in-place | [
"Good catch! I've opened a PR with a fix in the `transformers` repo.",
"@mariosasko Thanks!\r\n\r\nWill the doc of `datasets` be updated?\r\n\r\nI find some possible mistakes in doc about whether `remove_columns` is in-place.\r\n1. [You can also remove a column using map() with remove_columns but the present method is in-place (doesnβt copy the data to a new dataset) and is thus faster.](https://huggingface.co/docs/datasets/v2.17.1/en/package_reference/main_classes#datasets.Dataset.remove_columns)\r\n2. [You can also remove a column using Dataset.map() with remove_columns but the present method is in-place (doesnβt copy the data to a new dataset) and is thus faster.](https://huggingface.co/docs/datasets/v2.17.1/en/package_reference/main_classes#datasets.DatasetDict.remove_columns)\r\n3. [π€ Datasets also has a remove_columns() function which is faster because it doesnβt copy the data of the remaining columns.](https://huggingface.co/docs/datasets/v2.17.1/en/process#map)",
"I've linked a PR that will fix the usage in the `datasets` docs."
] | ### Describe the bug
The doc of `datasets` v2.17.0/v2.17.1 shows that `remove_columns` is in-place. [link](https://huggingface.co/docs/datasets/v2.17.1/en/package_reference/main_classes#datasets.DatasetDict.remove_columns)
In the text classification example of transformers v4.38.1, the columns are not removed.
https://github.com/huggingface/transformers/blob/a0857740c0e6127485c11476650314df3accc2b6/examples/pytorch/text-classification/run_classification.py#L421
### Steps to reproduce the bug
https://github.com/huggingface/transformers/blob/a0857740c0e6127485c11476650314df3accc2b6/examples/pytorch/text-classification/run_classification.py#L421
### Expected behavior
Actually remove the columns.
### Environment info
1. datasets v2.17.0
2. transformers v4.38.1 | 6,700 |
https://github.com/huggingface/datasets/issues/6699 | `Dataset` unexpected changed dict data and may cause error | [
"If `test.jsonl` contains more lines like:\r\n```\r\n{\"id\": 0, \"indexs\": {\"-1\": [0, 10]}}\r\n{\"id\": 1, \"indexs\": {\"-1\": [0, 10]}}\r\n{\"id\": 2, \"indexs\": {\"-2\": [0, 10]}}\r\n...\r\n{\"id\": n, \"indexs\": {\"-9999\": [0, 10]}}\r\n```\r\n\r\n`Dataset.from_json` will just raise an error:\r\n```\r\nAn error occurred while generating the dataset\r\nTypeError: Couldn't cast array of type\r\nstruct<-5942: list<item: int64>, -5943: list<item: int64>, -5944: list<item: int64>, -5945: list<item: int64>, -5946: list<item: int64>, -5947: list<item: int64>, -5948: list<item: int64>, -5949: list<item: int64>: ...\r\nto\r\n{... '-5312': Sequence(feature=Value(dtype='int64', id=None), length=-1, id=None), '-5313': Sequence(feature=Value(dtype='int64', id=None), length=-1, id=None)}\r\n\r\nThe above exception was the direct cause of the following exception:\r\n\r\nTraceback (most recent call last):\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/runpy.py\", line 198, in _run_module_as_main\r\n return _run_code(code, main_globals, None,\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/runpy.py\", line 88, in _run_code\r\n exec(code, run_globals)\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/__main__.py\", line 39, in <module>\r\n cli.main()\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py\", line 430, in main\r\n run()\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py\", line 284, in run_file\r\n runpy.run_path(target, run_name=\"__main__\")\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py\", line 321, in run_path\r\n return _run_module_code(code, init_globals, run_name,\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py\", line 135, in _run_module_code\r\n _run_code(code, mod_globals, init_globals,\r\n File \"/home/scruel/.vscode-server/extensions/ms-python.debugpy-2024.0.0-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py\", line 124, in _run_code\r\n exec(code, run_globals)\r\n File \"/home/scruel/Code/Python/Working/llm-memory/data_reader.py\", line 120, in <module>\r\n reader = SnippetReader(jsonl_path, npy_path)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/home/scruel/Code/Python/Working/llm-memory/data_reader.py\", line 85, in __init__\r\n self._dataset = Dataset.from_json(jsonl_path, features=)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/arrow_dataset.py\", line 1130, in from_json\r\n ).read()\r\n ^^^^^^\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/io/json.py\", line 59, in read\r\n self.builder.download_and_prepare(\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/builder.py\", line 1005, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/builder.py\", line 1100, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/builder.py\", line 1860, in _prepare_split\r\n for job_id, done, content in self._prepare_split_single(\r\n File \"/home/scruel/mambaforge/envs/vae/lib/python3.11/site-packages/datasets/builder.py\", line 2016, in _prepare_split_single\r\n raise DatasetGenerationError(\"An error occurred while generating the dataset\") from e\r\ndatasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset\r\n```",
"Hi! Our JSON parser expects all examples/rows to share the same set of columns (applies to nested columns, too), hence the error. \r\n\r\nTo read the `index` column, we would have to manually cast the input to PyArrow's `pa.map_` type, but this requires a more thorough investigation, as `pa.map_` has limited support in PyArrow."
] | ### Describe the bug
Will unexpected get keys with `None` value in the parsed json dict.
### Steps to reproduce the bug
```jsonl test.jsonl
{"id": 0, "indexs": {"-1": [0, 10]}}
{"id": 1, "indexs": {"-1": [0, 10]}}
```
```python
dataset = Dataset.from_json('.test.jsonl')
print(dataset[0])
```
Result:
```
{'id': 0, 'indexs': {'-1': [...], '-2': None, '-3': None, '-4': None, '-5': None, '-6': None, '-7': None, '-8': None, '-9': None, ...}}
```
Those keys with `None` value will unexpected appear in the dict.
### Expected behavior
Result should be
```
{'id': 0, 'indexs': {'-1': [0, 10]}}
```
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-6.5.0-14-generic-x86_64-with-glibc2.35
- Python version: 3.11.6
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.2
- Pandas version: 2.1.4
- `fsspec` version: 2023.10.0
| 6,699 |
https://github.com/huggingface/datasets/issues/6697 | Unable to Load Dataset in Kaggle | [
"FWIW, I run `load_dataset(\"llm-blender/mix-instruct\")` and it ran successfully.\r\nCan you clear your cache and try again?\r\n\r\n\r\n### Environment Info\r\n\r\n- `datasets` version: 2.17.0\r\n- Platform: Linux-6.2.6-76060206-generic-x86_64-with-glibc2.35\r\n- Python version: 3.9.13\r\n- `huggingface_hub` version: 0.20.3\r\n- PyArrow version: 15.0.0\r\n- Pandas version: 1.5.3\r\n- `fsspec` version: 2023.10.0",
"It is working on the Kaggle GPU instance but gives this same error when running on the CPU instance. Still to run it on Kaggle you require to install the latest versions of datasets and transformers.",
"This error means that `fsspec>=2023.12.0` is installed, which is incompatible with the current releases (the next `datasets` release will be the first to support it). In the meantime, downgrading `fsspec` (`pip install fsspec<=2023.12.0`) should fix the issue.",
"@mariosasko Thanks I got it to work with installing that version of fsspec."
] | ### Describe the bug
Having installed the latest versions of transformers==4.38.1 and datasets==2.17.1 Unable to load the dataset in a kaggle notebook.
Get this Error:
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[8], line 3
1 from datasets import load_dataset
----> 3 dataset = load_dataset("llm-blender/mix-instruct")
File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1664, in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, **config_kwargs)
1661 ignore_verifications = ignore_verifications or save_infos
1663 # Create a dataset builder
-> 1664 builder_instance = load_dataset_builder(
1665 path=path,
1666 name=name,
1667 data_dir=data_dir,
1668 data_files=data_files,
1669 cache_dir=cache_dir,
1670 features=features,
1671 download_config=download_config,
1672 download_mode=download_mode,
1673 revision=revision,
1674 use_auth_token=use_auth_token,
1675 **config_kwargs,
1676 )
1678 # Return iterable dataset in case of streaming
1679 if streaming:
File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1490, in load_dataset_builder(path, name, data_dir, data_files, cache_dir, features, download_config, download_mode, revision, use_auth_token, **config_kwargs)
1488 download_config = download_config.copy() if download_config else DownloadConfig()
1489 download_config.use_auth_token = use_auth_token
-> 1490 dataset_module = dataset_module_factory(
1491 path,
1492 revision=revision,
1493 download_config=download_config,
1494 download_mode=download_mode,
1495 data_dir=data_dir,
1496 data_files=data_files,
1497 )
1499 # Get dataset builder class from the processing script
1500 builder_cls = import_main_class(dataset_module.module_path)
File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1242, in dataset_module_factory(path, revision, download_config, download_mode, force_local_path, dynamic_modules_path, data_dir, data_files, **download_kwargs)
1237 if isinstance(e1, FileNotFoundError):
1238 raise FileNotFoundError(
1239 f"Couldn't find a dataset script at {relative_to_absolute_path(combined_path)} or any data file in the same directory. "
1240 f"Couldn't find '{path}' on the Hugging Face Hub either: {type(e1).__name__}: {e1}"
1241 ) from None
-> 1242 raise e1 from None
1243 else:
1244 raise FileNotFoundError(
1245 f"Couldn't find a dataset script at {relative_to_absolute_path(combined_path)} or any data file in the same directory."
1246 )
File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1230, in dataset_module_factory(path, revision, download_config, download_mode, force_local_path, dynamic_modules_path, data_dir, data_files, **download_kwargs)
1215 return HubDatasetModuleFactoryWithScript(
1216 path,
1217 revision=revision,
(...)
1220 dynamic_modules_path=dynamic_modules_path,
1221 ).get_module()
1222 else:
1223 return HubDatasetModuleFactoryWithoutScript(
1224 path,
1225 revision=revision,
1226 data_dir=data_dir,
1227 data_files=data_files,
1228 download_config=download_config,
1229 download_mode=download_mode,
-> 1230 ).get_module()
1231 except Exception as e1: # noqa: all the attempts failed, before raising the error we should check if the module is already cached.
1232 try:
File /opt/conda/lib/python3.10/site-packages/datasets/load.py:846, in HubDatasetModuleFactoryWithoutScript.get_module(self)
836 token = self.download_config.use_auth_token
837 hfh_dataset_info = HfApi(config.HF_ENDPOINT).dataset_info(
838 self.name,
839 revision=self.revision,
840 token=token,
841 timeout=100.0,
842 )
843 patterns = (
844 sanitize_patterns(self.data_files)
845 if self.data_files is not None
--> 846 else get_patterns_in_dataset_repository(hfh_dataset_info)
847 )
848 data_files = DataFilesDict.from_hf_repo(
849 patterns,
850 dataset_info=hfh_dataset_info,
851 allowed_extensions=ALL_ALLOWED_EXTENSIONS,
852 )
853 infered_module_names = {
854 key: infer_module_for_data_files(data_files_list, use_auth_token=self.download_config.use_auth_token)
855 for key, data_files_list in data_files.items()
856 }
File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:471, in get_patterns_in_dataset_repository(dataset_info)
469 resolver = partial(_resolve_single_pattern_in_dataset_repository, dataset_info)
470 try:
--> 471 return _get_data_files_patterns(resolver)
472 except FileNotFoundError:
473 raise FileNotFoundError(
474 f"The dataset repository at '{dataset_info.id}' doesn't contain any data file."
475 ) from None
File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:99, in _get_data_files_patterns(pattern_resolver)
97 try:
98 for pattern in patterns:
---> 99 data_files = pattern_resolver(pattern)
100 if len(data_files) > 0:
101 non_empty_splits.append(split)
File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:303, in _resolve_single_pattern_in_dataset_repository(dataset_info, pattern, allowed_extensions)
301 data_files_ignore = FILES_TO_IGNORE
302 fs = HfFileSystem(repo_info=dataset_info)
--> 303 glob_iter = [PurePath(filepath) for filepath in fs.glob(PurePath(pattern).as_posix()) if fs.isfile(filepath)]
304 matched_paths = [
305 filepath
306 for filepath in glob_iter
307 if filepath.name not in data_files_ignore and not filepath.name.startswith(".")
308 ]
309 if allowed_extensions is not None:
File /opt/conda/lib/python3.10/site-packages/fsspec/spec.py:606, in AbstractFileSystem.glob(self, path, maxdepth, **kwargs)
602 depth = None
604 allpaths = self.find(root, maxdepth=depth, withdirs=True, detail=True, **kwargs)
--> 606 pattern = glob_translate(path + ("/" if ends_with_sep else ""))
607 pattern = re.compile(pattern)
609 out = {
610 p: info
611 for p, info in sorted(allpaths.items())
(...)
618 )
619 }
File /opt/conda/lib/python3.10/site-packages/fsspec/utils.py:734, in glob_translate(pat)
732 continue
733 elif "**" in part:
--> 734 raise ValueError(
735 "Invalid pattern: '**' can only be an entire path component"
736 )
737 if part:
738 results.extend(_translate(part, f"{not_sep}*", not_sep))
ValueError: Invalid pattern: '**' can only be an entire path component
```
```
After loading this dataset
### Steps to reproduce the bug
```
from datasets import load_dataset
dataset = load_dataset("llm-blender/mix-instruct")
```
### Expected behavior
The dataset should load with desired split.
### Environment info
- `datasets` version: 2.17.1
- Platform: Linux-5.15.133+-x86_64-with-glibc2.31
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0
| 6,697 |
https://github.com/huggingface/datasets/issues/6695 | Support JSON file with an array of strings | [
"https://huggingface.co/datasets/CausalLM/Refined-Anime-Text/discussions/1 has been fixed, but how can we check if there are other datasets with the same error, in datasets-server's database? I don't know how to get the list of erroneous cache entries, since we only copied `Error code: JobManagerCrashedError`, but not the traceback in `details`... Do you remember the error message, or the underlying exception, we had?"
] | Support loading a dataset from a JSON file with an array of strings.
See: https://huggingface.co/datasets/CausalLM/Refined-Anime-Text/discussions/1 | 6,695 |
https://github.com/huggingface/datasets/issues/6691 | load_dataset() does not support tsv | [
"#self-assign",
"Hi @dipsivenkatesh,\r\n\r\nPlease note that this functionality is already implemented. Our CSV builder uses `pandas.read_csv` under the hood, and you can pass the parameter `delimiter=\"\\t\"` to read TSV files.\r\n\r\nSee the list of CSV config parameters in our docs: https://huggingface.co/docs/datasets/package_reference/loading_methods#datasets.packaged_modules.csv.CsvConfig"
] | ### Feature request
the load_dataset() for local functions support file types like csv, json etc but not of type tsv (tab separated values).
### Motivation
cant easily load files of type tsv, have to convert them to another type like csv then load
### Your contribution
Can try by raising a PR with a little help, currently went through the code but didn't fully understand | 6,691 |
https://github.com/huggingface/datasets/issues/6690 | Add function to convert a script-dataset to Parquet | [] | Add function to convert a script-dataset to Parquet and push it to the Hub, analogously to the Space: "Convert a Hugging Face dataset to Parquet" | 6,690 |
https://github.com/huggingface/datasets/issues/6689 | .load_dataset() method defaults to zstandard | [
"The dataset is made of JSON files compressed using zstandard, as you can see here: https://huggingface.co/datasets/cerebras/SlimPajama-627B/tree/main/test/chunk1\r\n\r\nThat's why it asks for zstandard to be installed.\r\n\r\nThough I'm intrigued that you manage to load the dataset without zstandard installed. Maybe `pyarrow` that we use to load JSON data under the hood got support for zstandard at one point.",
"> The dataset is made of JSON files compressed using zstandard, as you can see here: https://huggingface.co/datasets/cerebras/SlimPajama-627B/tree/main/test/chunk1\r\n> \r\n> That's why it asks for zstandard to be installed.\r\n> \r\n> Though I'm intrigued that you manage to load the dataset without zstandard installed. Maybe `pyarrow` that we use to load JSON data under the hood got support for zstandard at one point.\r\n\r\nQuestion, then.\r\n\r\nWhen I loaded this dataset back in October, it downloaded all the files, and then loaded into memory just fine.\r\n\r\nNOW, it has to sit there and unpack all these zstd files (3.6TB worth). Further, when they're in my harddrive, they're regular json files. It's only when looking at the LFS, or when the loading script runs, that I get asked to install zstd.\r\n\r\nMy question is, **is this normal?** As far as I can tell, there's no reason the dataset or the loading methods should have changed between then and now. Was my old behavior flawed, and the new behavior correct?\r\n\r\nI mean, I got it working eventually, but it was pulling teeth, and it still doesn't load right, as I had to unpack each chunk separately, so there's no clean mapping between the chunks and the broader dataset.",
"The `ZstdExtractor` has been added 3 years ago and we haven't touched it since then. Same for the JSON loader.\r\n\r\n`zstandard` is required as soon as you try to load a file with the `.zstd` extension or if a file starts with the Zstandard magic number `b\"\\x28\\xb5\\x2f\\xfd\"` (used to recognize Zstandard files).\r\n\r\nNote that the extraction only has to happen once - if you reload the dataset it will be reloaded from your cache directly.\r\n\r\nNot sure what happened between October and now unfortunately",
"Understood, thank you for clarifying that for me.\r\n\r\nI'll look into how best to collate my stack of batches w/o creating duplicate arrow tables for each one."
] | ### Describe the bug
Regardless of what method I use, datasets defaults to zstandard for unpacking my datasets.
This is poor behavior, because not only is zstandard not a dependency in the huggingface package (and therefore, your dataset loading will be interrupted while it asks you to install the package), but it happens on datasets that are uploaded in json format too, meaning the dataset loader will attempt to convert the data to a zstandard compatible format, and THEN try to unpackage it.
My 4tb drive runs out of room when using zstandard on slimpajama. It loads fine on 1.5tb when using json, however I lack the understanding of the "magic numbers" system used to select the unpackaging algorithm, so I can't push a change myself.
Commenting out this line, in "/datasets/utils/extract.py" fixes the issue, and causes SlimPajama to properly extract using rational amounts of storage, however it completely disables zstandard, which is probably undesirable behavior. Someone with an understanding of the "magic numbers" system should probably take a pass over this issue.
```
class Extractor:
# Put zip file to the last, b/c it is possible wrongly detected as zip (I guess it means: as tar or gzip)
extractors: Dict[str, Type[BaseExtractor]] = {
"tar": TarExtractor,
"gzip": GzipExtractor,
"zip": ZipExtractor,
"xz": XzExtractor,
#"zstd": ZstdExtractor, # This line needs to go, in order for datasets to work w/o non-dependent packages
"rar": RarExtractor,
"bz2": Bzip2Extractor,
"7z": SevenZipExtractor, # <Added version="2.4.0"/>
"lz4": Lz4Extractor, # <Added version="2.4.0"/>
}
```
### Steps to reproduce the bug
'''
from datasaets import load_dataset
load_dataset(path="/cerebras/SlimPajama-627B")
'''
This alone should trigger the error on any system that does not have zstandard pip installed.
### Expected behavior
This repository (which is encoded in json format, not zstandard) should check whether zstandard is installed before defaulting to it. Additionally, using zstandard should not use more than 3x the required space that other extraction mechanisms use.
### Environment info
- `datasets` version: 2.17.1
- Platform: Linux-6.5.0-18-generic-x86_64-with-glibc2.35
- Python version: 3.12.0
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0 | 6,689 |
https://github.com/huggingface/datasets/issues/6688 | Tensor type (e.g. from `return_tensors`) ignored in map | [
"Hi, this is expected behavior since all the tensors are converted to Arrow data (the storage type behind a Dataset).\r\n\r\nTo get pytorch tensors back, you can set the dataset format to \"torch\":\r\n\r\n```python\r\nds = ds.with_format(\"torch\")\r\n```",
"Thanks. Just one additional question. During the pipeline `<framework> -> arrow -> <framework>`, does `.with_format` zero-copies the tensors or is it a deep copy? And is this behavior framework-dependent?\r\n\r\nThanks again.",
"We do zero-copy Arrow <-> NumPy <-> PyTorch when the output dtype matches the original dtype, but for other frameworks it depends. For example JAX doesn't allow zero-copy NumPy -> JAX at all IIRC.\r\n\r\nCurrently tokenized data are formatted using a copy though, since tokens are stored as int32 and returned as int64 torch tensors."
] | ### Describe the bug
I don't know if it is a bug or an expected behavior, but the tensor type seems to be ignored after applying map. For example, mapping over to tokenize text with a transformers' tokenizer always returns lists and it ignore the `return_tensors` argument.
If this is an expected behaviour (e.g., for caching/Arrow compatibility/etc.) it should be clearly documented. For example, current documentation (see [here](https://huggingface.co/docs/datasets/v2.17.1/en/nlp_process#map)) clearly state to "set `return_tensors="np"` when you tokenize your text" to have Numpy arrays.
### Steps to reproduce the bug
```py
# %%%
import datasets
import numpy as np
import tensorflow as tf
import torch
from transformers import AutoTokenizer
# %%
ds = datasets.load_dataset("cnn_dailymail", "1.0.0", split="train[:1%]")
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
#%%
for return_tensors in [None, "np", "pt", "tf", "jax"]:
print(f"********** no map, return_tensors={return_tensors} **********")
_ds = tokenizer(ds["article"], return_tensors=return_tensors, truncation=True, padding=True)
print('Type <input_ids>:', type(_ds["input_ids"]))
# %%
for return_tensors in [None, "np", "pt", "tf", "jax"]:
print(f"********** map, return_tensors={return_tensors} **********")
_ds = ds.map(
lambda examples: tokenizer(examples["article"], return_tensors=return_tensors, truncation=True, padding=True),
batched=True,
remove_columns=["article"],
)
print('Type <input_ids>:', type(_ds[0]["input_ids"]))
```
### Expected behavior
The output from the script above. I would expect the second half to be the same.
```
********** no map, return_tensors=None **********
Type <input_ids>: <class 'list'>
********** no map, return_tensors=np **********
Type <input_ids>: <class 'numpy.ndarray'>
********** no map, return_tensors=pt **********
Type <input_ids>: <class 'torch.Tensor'>
********** no map, return_tensors=tf **********
Type <input_ids>: <class 'tensorflow.python.framework.ops.EagerTensor'>
********** no map, return_tensors=jax **********
Type <input_ids>: <class 'jaxlib.xla_extension.ArrayImpl'>
********** map, return_tensors=None **********
Type <input_ids>: <class 'list'>
********** map, return_tensors=np **********
Type <input_ids>: <class 'list'>
********** map, return_tensors=pt **********
Type <input_ids>: <class 'list'>
********** map, return_tensors=tf **********
Type <input_ids>: <class 'list'>
********** map, return_tensors=jax **********
Type <input_ids>: <class 'list'>
```
### Environment info
- `datasets` version: 2.17.1
- Platform: Redacted (linux)
- Python version: 3.10.12
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.1.3
- `fsspec` version: 2023.10.0 | 6,688 |
https://github.com/huggingface/datasets/issues/6686 | Question: Is there any way for uploading a large image dataset? | [
"```\r\nimport pandas as pd\r\nfrom datasets import Dataset, Image\r\n\r\n# Read the CSV file\r\ndata = pd.read_csv(\"XXXX.csv\")\r\n\r\n# Create a Hugging Face Dataset\r\ndataset = Dataset.from_pandas(data)\r\ndataset = dataset.cast_column(\"file_name\", Image())\r\n\r\n# Upload to Hugging Face Hub (make sure authentication is set up)\r\ndataset.push_to_hub(\"XXXXX\"\")\r\n```\r\n\r\nstuck in \"Casting the dataset\r\n\r\n\"\r\n"
] | I am uploading an image dataset like this:
```
dataset = load_dataset(
"json",
data_files={"train": "data/custom_dataset/train.json", "validation": "data/custom_dataset/val.json"},
)
dataset = dataset.cast_column("images", Sequence(Image()))
dataset.push_to_hub("StanfordAIMI/custom_dataset", max_shard_size="1GB")
```
where it takes a long time in the `Map` process. Do you think I can use multi-processing to map all the image data to the memory first? For the `Map()` function, I can set `num_proc`. But for `push_to_hub` and `cast_column`, I can not find it.
Thanks in advance!
Best, | 6,686 |
https://github.com/huggingface/datasets/issues/6679 | Node.js 16 GitHub Actions are deprecated | [] | `Node.js` 16 GitHub Actions are deprecated. See: https://github.blog/changelog/2023-09-22-github-actions-transitioning-from-node-16-to-node-20/
We should update them to Node 20.
See warnings in our CI, e.g.: https://github.com/huggingface/datasets/actions/runs/7957295009?pr=6678
> Node.js 16 actions are deprecated. Please update the following actions to use Node.js 20: actions/checkout@v3, actions/setup-python@v4. For more information see: https://github.blog/changelog/2023-09-22-github-actions-transitioning-from-node-16-to-node-20/.
| 6,679 |
https://github.com/huggingface/datasets/issues/6676 | Can't Read List of JSON Files Properly | [
"Found the issue, if there are other files in the directory, it gets caught into this `*` so essentially it should be `*.json`. Could we possibly to check for list of files to make sure the pattern matches json files and raise error if not?",
"I don't think we should filter for `*.json` as this might silently remove desired files for many users. And this could be a major breaking change for many organizations.\r\n\r\nYou could do the globbing yourself which would keep the code clean.\r\n\r\n```python\r\nfrom glob import glob\r\n\r\nDataset.from_json(glob('folder/*.json'))\r\n```",
"I think it should still be fine to log a warning message in case the folder contains different files? I also don't get why would this be breaking as in the end using `from_FILE_TYPE` should be able to read a specific file type only. Maybe some other use case I am not aware of but since globbing or this case not mentioned anywhere in the doc, I spent quite a bit of time trying to figure out where the issue was. Just making sure it's clear for users."
] | ### Describe the bug
Trying to read a bunch of JSON files into Dataset class but default approach doesn't work. I don't get why it works when I read it one by one but not when I pass as a list :man_shrugging:
The code fails with
```
ArrowInvalid: JSON parse error: Invalid value. in row 0
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x80 in position 0: invalid start byte
DatasetGenerationError: An error occurred while generating the dataset
```
### Steps to reproduce the bug
This doesn't work
```
from datasets import Dataset
# dir contains 100 json files.
Dataset.from_json("/PUT SOME PATH HERE/*")
```
This works:
```
from datasets import concatenate_datasets
ls_ds = []
for file in list_of_json_files:
ls_ds.append(Dataset.from_json(file))
ds = concatenate_datasets(ls_ds)
```
### Expected behavior
I expect this to read json files properly as error is not clear
### Environment info
- `datasets` version: 2.17.0
- Platform: Linux-6.5.0-15-generic-x86_64-with-glibc2.35
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.2
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0
| 6,676 |
https://github.com/huggingface/datasets/issues/6675 | Allow image model (color conversion) to be specified as part of datasets Image() decode | [
"It would be a great addition indeed :)\r\n\r\nThis can be implemented the same way we have `sampling_rate` for Audio(): we just add a new parameter to the Image() type and take this parameter into account in `Image.decode_example`\r\n\r\nEDIT: adding an example of how it can be used:\r\n\r\n```python\r\nds = ds.cast_column(\"image\", Image(mode=...))\r\n```"
] | ### Feature request
Typical torchvision / torch Datasets in image applications apply color conversion in the Dataset portion of the code as part of image decode, separately from the image transform stack. This is true for PIL.Image where convert is usually called in dataset, for native torchvision https://pytorch.org/vision/main/generated/torchvision.io.decode_jpeg.html, and similarly in tensorflow.data pipelines decode_jpeg or https://www.tensorflow.org/api_docs/python/tf/io/decode_and_crop_jpeg have a channels arg that allows controlling the image mode in the decode step.
datasets currently requires this pattern (from [examples](https://huggingface.co/docs/datasets/main/en/image_process)):
```
from torchvision.transforms import Compose, ColorJitter, ToTensor
jitter = Compose(
[
ColorJitter(brightness=0.25, contrast=0.25, saturation=0.25, hue=0.7),
ToTensor(),
]
)
def transforms(examples):
examples["pixel_values"] = [jitter(image.convert("RGB")) for image in examples["image"]]
return examples
```
### Motivation
It would be nice to be able to handle `image.convert("RGB")` (or other modes) in the decode step, before applying torchvision transforms, this would reduce differences in code when handling pipelines that can handle torchvision, webdatset, or hf datasets with fewer code differences and without needing to handle image mode argument passing in two different stages of the pipelines...
### Your contribution
Can do a PR with guidance on how mode should be passed / set on the dataset. | 6,675 |
https://github.com/huggingface/datasets/issues/6674 | Depprcated Overview.ipynb Link to new Quickstart Notebook invalid | [
"Good catch! Feel free to open a PR to fix the link."
] | ### Describe the bug
For the dreprecated notebook found [here](https://github.com/huggingface/datasets/blob/main/notebooks/Overview.ipynb). The link to the new notebook is broken.
### Steps to reproduce the bug
Click the [Quickstart notebook](https://github.com/huggingface/notebooks/blob/main/datasets_doc/quickstart.ipynb) link in the notebook.
### Expected behavior
I believe is it suposed to link [here](https://github.com/huggingface/notebooks/blob/main/datasets_doc/en/quickstart.ipynb). That is mentioned in the readme.
### Environment info
Colab | 6,674 |
https://github.com/huggingface/datasets/issues/6673 | IterableDataset `set_epoch` is ignored when DataLoader `persistent_workers=True` | [] | ### Describe the bug
When persistent workers are enabled, the epoch that's set via the IterableDataset instance held by the training process is ignored by the workers as they are disconnected across processes.
PyTorch samplers for non-iterable datasets have a mechanism to sync this, datasets.IterableDataset does not.
In my own use of IterableDatasets I usually track the epoch count which crosses process boundaries in a multiprocessing.Value
### Steps to reproduce the bug
Use a streaming dataset (Iterable) w/ the recommended pattern below and `persistent_workers=True` in the torch DataLoader.
```
for epoch in range(epochs):
shuffled_dataset.set_epoch(epoch)
for example in shuffled_dataset:
...
```
### Expected behavior
When the canonical bit of code above is used with `num_workers > 0` and `persistent_workers=True`, the epoch set via `set_epoch()` is propagated to the IterableDataset instances in the worker processes
### Environment info
N/A | 6,673 |
https://github.com/huggingface/datasets/issues/6671 | CSV builder raises deprecation warning on verbose parameter | [] | CSV builder raises a deprecation warning on `verbose` parameter:
```
FutureWarning: The 'verbose' keyword in pd.read_csv is deprecated and will be removed in a future version.
```
See:
- https://github.com/pandas-dev/pandas/pull/56556
- https://github.com/pandas-dev/pandas/pull/57450 | 6,671 |
https://github.com/huggingface/datasets/issues/6670 | ValueError | [
"Hi @prashanth19bolukonda,\r\n\r\nYou have to restart the notebook runtime session after the installation of `datasets`.\r\n\r\nDuplicate of:\r\n- #5923",
"Thank you soo much\r\n\r\nOn Fri, Feb 16, 2024 at 8:14β―PM Albert Villanova del Moral <\r\n***@***.***> wrote:\r\n\r\n> Closed #6670 <https://github.com/huggingface/datasets/issues/6670> as\r\n> completed.\r\n>\r\n> β\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/huggingface/datasets/issues/6670#event-11829788289>,\r\n> or unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/A2Y44YDQOBUFUWMR4C5O3QTYT5WDJAVCNFSM6AAAAABDL24S5SVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJRHAZDSNZYHAZDQOI>\r\n> .\r\n> You are receiving this because you were mentioned.Message ID:\r\n> ***@***.***>\r\n>\r\n"
] | ### Describe the bug
ValueError Traceback (most recent call last)
[<ipython-input-11-9b99bc80ec23>](https://localhost:8080/#) in <cell line: 11>()
9 import numpy as np
10 import matplotlib.pyplot as plt
---> 11 from datasets import DatasetDict, Dataset
12 from transformers import AutoTokenizer, AutoModelForSequenceClassification
13 from transformers import Trainer, TrainingArguments
5 frames
[/usr/local/lib/python3.10/dist-packages/datasets/__init__.py](https://localhost:8080/#) in <module>
16 __version__ = "2.17.0"
17
---> 18 from .arrow_dataset import Dataset
19 from .arrow_reader import ReadInstruction
20 from .builder import ArrowBasedBuilder, BeamBasedBuilder, BuilderConfig, DatasetBuilder, GeneratorBasedBuilder
[/usr/local/lib/python3.10/dist-packages/datasets/arrow_dataset.py](https://localhost:8080/#) in <module>
65
66 from . import config
---> 67 from .arrow_reader import ArrowReader
68 from .arrow_writer import ArrowWriter, OptimizedTypedSequence
69 from .data_files import sanitize_patterns
[/usr/local/lib/python3.10/dist-packages/datasets/arrow_reader.py](https://localhost:8080/#) in <module>
27
28 import pyarrow as pa
---> 29 import pyarrow.parquet as pq
30 from tqdm.contrib.concurrent import thread_map
31
[/usr/local/lib/python3.10/dist-packages/pyarrow/parquet/__init__.py](https://localhost:8080/#) in <module>
18 # flake8: noqa
19
---> 20 from .core import *
[/usr/local/lib/python3.10/dist-packages/pyarrow/parquet/core.py](https://localhost:8080/#) in <module>
34 import pyarrow as pa
35 import pyarrow.lib as lib
---> 36 import pyarrow._parquet as _parquet
37
38 from pyarrow._parquet import (ParquetReader, Statistics, # noqa
/usr/local/lib/python3.10/dist-packages/pyarrow/_parquet.pyx in init pyarrow._parquet()
ValueError: pyarrow.lib.IpcWriteOptions size changed, may indicate binary incompatibility. Expected 88 from C header, got 72 from PyObject
### Steps to reproduce the bug
ValueError: pyarrow.lib.IpcWriteOptions size changed, may indicate binary incompatibility. Expected 88 from C header, got 72 from PyObject
### Expected behavior
Resolve the binary incompatibility
### Environment info
Google Colab Note book | 6,670 |
https://github.com/huggingface/datasets/issues/6669 | attribute error when writing trainer.train() | [
"Hi! Kaggle notebooks use an outdated version of `datasets`, so you should update the `datasets` installation (with `!pip install -U datasets`) to avoid the error.",
"Thank you for your response\r\n\r\nOn Thu, Feb 29, 2024 at 10:55β―PM Mario Ε aΕ‘ko ***@***.***>\r\nwrote:\r\n\r\n> Closed #6669 <https://github.com/huggingface/datasets/issues/6669> as\r\n> completed.\r\n>\r\n> β\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/huggingface/datasets/issues/6669#event-11969246964>,\r\n> or unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/A2Y44YG2RRVMYONNKPLBVE3YV5SAPAVCNFSM6AAAAABDLZ3BTSVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJRHE3DSMRUGY4TMNA>\r\n> .\r\n> You are receiving this because you authored the thread.Message ID:\r\n> ***@***.***>\r\n>\r\n"
] | ### Describe the bug
AttributeError Traceback (most recent call last)
Cell In[39], line 2
1 # Start the training process
----> 2 trainer.train()
File /opt/conda/lib/python3.10/site-packages/transformers/trainer.py:1539, in Trainer.train(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
1537 hf_hub_utils.enable_progress_bars()
1538 else:
-> 1539 return inner_training_loop(
1540 args=args,
1541 resume_from_checkpoint=resume_from_checkpoint,
1542 trial=trial,
1543 ignore_keys_for_eval=ignore_keys_for_eval,
1544 )
File /opt/conda/lib/python3.10/site-packages/transformers/trainer.py:1836, in Trainer._inner_training_loop(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)
1833 rng_to_sync = True
1835 step = -1
-> 1836 for step, inputs in enumerate(epoch_iterator):
1837 total_batched_samples += 1
1839 if self.args.include_num_input_tokens_seen:
File /opt/conda/lib/python3.10/site-packages/accelerate/data_loader.py:451, in DataLoaderShard.__iter__(self)
449 # We iterate one batch ahead to check when we are at the end
450 try:
--> 451 current_batch = next(dataloader_iter)
452 except StopIteration:
453 yield
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataloader.py:630, in _BaseDataLoaderIter.__next__(self)
627 if self._sampler_iter is None:
628 # TODO([https://github.com/pytorch/pytorch/issues/76750)](https://github.com/pytorch/pytorch/issues/76750)%3C/span%3E)
629 self._reset() # type: ignore[call-arg]
--> 630 data = self._next_data()
631 self._num_yielded += 1
632 if self._dataset_kind == _DatasetKind.Iterable and \
633 self._IterableDataset_len_called is not None and \
634 self._num_yielded > self._IterableDataset_len_called:
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataloader.py:674, in _SingleProcessDataLoaderIter._next_data(self)
672 def _next_data(self):
673 index = self._next_index() # may raise StopIteration
--> 674 data = self._dataset_fetcher.fetch(index) # may raise StopIteration
675 if self._pin_memory:
676 data = _utils.pin_memory.pin_memory(data, self._pin_memory_device)
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py:51, in _MapDatasetFetcher.fetch(self, possibly_batched_index)
49 data = self.dataset.__getitems__(possibly_batched_index)
50 else:
---> 51 data = [self.dataset[idx] for idx in possibly_batched_index]
52 else:
53 data = self.dataset[possibly_batched_index]
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py:51, in <listcomp>(.0)
49 data = self.dataset.__getitems__(possibly_batched_index)
50 else:
---> 51 data = [self.dataset[idx] for idx in possibly_batched_index]
52 else:
53 data = self.dataset[possibly_batched_index]
File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:1764, in Dataset.__getitem__(self, key)
1762 def __getitem__(self, key): # noqa: F811
1763 """Can be used to index columns (by string names) or rows (by integer index or iterable of indices or bools)."""
-> 1764 return self._getitem(
1765 key,
1766 )
File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:1749, in Dataset._getitem(self, key, decoded, **kwargs)
1747 formatter = get_formatter(format_type, features=self.features, decoded=decoded, **format_kwargs)
1748 pa_subtable = query_table(self._data, key, indices=self._indices if self._indices is not None else None)
-> 1749 formatted_output = format_table(
1750 pa_subtable, key, formatter=formatter, format_columns=format_columns, output_all_columns=output_all_columns
1751 )
1752 return formatted_output
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:540, in format_table(table, key, formatter, format_columns, output_all_columns)
538 else:
539 pa_table_to_format = pa_table.drop(col for col in pa_table.column_names if col not in format_columns)
--> 540 formatted_output = formatter(pa_table_to_format, query_type=query_type)
541 if output_all_columns:
542 if isinstance(formatted_output, MutableMapping):
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:281, in Formatter.__call__(self, pa_table, query_type)
279 def __call__(self, pa_table: pa.Table, query_type: str) -> Union[RowFormat, ColumnFormat, BatchFormat]:
280 if query_type == "row":
--> 281 return self.format_row(pa_table)
282 elif query_type == "column":
283 return self.format_column(pa_table)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/torch_formatter.py:57, in TorchFormatter.format_row(self, pa_table)
56 def format_row(self, pa_table: pa.Table) -> dict:
---> 57 row = self.numpy_arrow_extractor().extract_row(pa_table)
58 return self.recursive_tensorize(row)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:154, in NumpyArrowExtractor.extract_row(self, pa_table)
153 def extract_row(self, pa_table: pa.Table) -> dict:
--> 154 return _unnest(self.extract_batch(pa_table))
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:160, in NumpyArrowExtractor.extract_batch(self, pa_table)
159 def extract_batch(self, pa_table: pa.Table) -> dict:
--> 160 return {col: self._arrow_array_to_numpy(pa_table[col]) for col in pa_table.column_names}
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:160, in <dictcomp>(.0)
159 def extract_batch(self, pa_table: pa.Table) -> dict:
--> 160 return {col: self._arrow_array_to_numpy(pa_table[col]) for col in pa_table.column_names}
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:196, in NumpyArrowExtractor._arrow_array_to_numpy(self, pa_array)
194 array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only).tolist()
195 if len(array) > 0:
--> 196 if any(
197 (isinstance(x, np.ndarray) and (x.dtype == np.object or x.shape != array[0].shape))
198 or (isinstance(x, float) and np.isnan(x))
199 for x in array
200 ):
201 return np.array(array, copy=False, **{**self.np_array_kwargs, "dtype": np.object})
202 return np.array(array, copy=False, **self.np_array_kwargs)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:197, in <genexpr>(.0)
194 array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only).tolist()
195 if len(array) > 0:
196 if any(
--> 197 (isinstance(x, np.ndarray) and (x.dtype == np.object or x.shape != array[0].shape))
198 or (isinstance(x, float) and np.isnan(x))
199 for x in array
200 ):
201 return np.array(array, copy=False, **{**self.np_array_kwargs, "dtype": np.object})
202 return np.array(array, copy=False, **self.np_array_kwargs)
File /opt/conda/lib/python3.10/site-packages/numpy/__init__.py:324, in __getattr__(attr)
319 warnings.warn(
320 f"In the future `np.{attr}` will be defined as the "
321 "corresponding NumPy scalar.", FutureWarning, stacklevel=2)
323 if attr in __former_attrs__:
--> 324 raise AttributeError(__former_attrs__[attr])
326 if attr == 'testing':
327 import numpy.testing as testing
AttributeError: module 'numpy' has no attribute 'object'.
`np.object` was a deprecated alias for the builtin `object`. To avoid this error in existing code, use `object` by itself. Doing this will not modify any behavior and is safe.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecationsAttributeError Traceback (most recent call last)
Cell In[39], line 2
1 # Start the training process
----> 2 trainer.train()
File /opt/conda/lib/python3.10/site-packages/transformers/trainer.py:1539, in Trainer.train(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
1537 hf_hub_utils.enable_progress_bars()
1538 else:
-> 1539 return inner_training_loop(
1540 args=args,
1541 resume_from_checkpoint=resume_from_checkpoint,
1542 trial=trial,
1543 ignore_keys_for_eval=ignore_keys_for_eval,
1544 )
File /opt/conda/lib/python3.10/site-packages/transformers/trainer.py:1836, in Trainer._inner_training_loop(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)
1833 rng_to_sync = True
1835 step = -1
-> 1836 for step, inputs in enumerate(epoch_iterator):
1837 total_batched_samples += 1
1839 if self.args.include_num_input_tokens_seen:
File /opt/conda/lib/python3.10/site-packages/accelerate/data_loader.py:451, in DataLoaderShard.__iter__(self)
449 # We iterate one batch ahead to check when we are at the end
450 try:
--> 451 current_batch = next(dataloader_iter)
452 except StopIteration:
453 yield
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataloader.py:630, in _BaseDataLoaderIter.__next__(self)
627 if self._sampler_iter is None:
628 # TODO([https://github.com/pytorch/pytorch/issues/76750)](https://github.com/pytorch/pytorch/issues/76750)%3C/span%3E)
629 self._reset() # type: ignore[call-arg]
--> 630 data = self._next_data()
631 self._num_yielded += 1
632 if self._dataset_kind == _DatasetKind.Iterable and \
633 self._IterableDataset_len_called is not None and \
634 self._num_yielded > self._IterableDataset_len_called:
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataloader.py:674, in _SingleProcessDataLoaderIter._next_data(self)
672 def _next_data(self):
673 index = self._next_index() # may raise StopIteration
--> 674 data = self._dataset_fetcher.fetch(index) # may raise StopIteration
675 if self._pin_memory:
676 data = _utils.pin_memory.pin_memory(data, self._pin_memory_device)
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py:51, in _MapDatasetFetcher.fetch(self, possibly_batched_index)
49 data = self.dataset.__getitems__(possibly_batched_index)
50 else:
---> 51 data = [self.dataset[idx] for idx in possibly_batched_index]
52 else:
53 data = self.dataset[possibly_batched_index]
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py:51, in <listcomp>(.0)
49 data = self.dataset.__getitems__(possibly_batched_index)
50 else:
---> 51 data = [self.dataset[idx] for idx in possibly_batched_index]
52 else:
53 data = self.dataset[possibly_batched_index]
File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:1764, in Dataset.__getitem__(self, key)
1762 def __getitem__(self, key): # noqa: F811
1763 """Can be used to index columns (by string names) or rows (by integer index or iterable of indices or bools)."""
-> 1764 return self._getitem(
1765 key,
1766 )
File /opt/conda/lib/python3.10/site-packages/datasets/arrow_dataset.py:1749, in Dataset._getitem(self, key, decoded, **kwargs)
1747 formatter = get_formatter(format_type, features=self.features, decoded=decoded, **format_kwargs)
1748 pa_subtable = query_table(self._data, key, indices=self._indices if self._indices is not None else None)
-> 1749 formatted_output = format_table(
1750 pa_subtable, key, formatter=formatter, format_columns=format_columns, output_all_columns=output_all_columns
1751 )
1752 return formatted_output
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:540, in format_table(table, key, formatter, format_columns, output_all_columns)
538 else:
539 pa_table_to_format = pa_table.drop(col for col in pa_table.column_names if col not in format_columns)
--> 540 formatted_output = formatter(pa_table_to_format, query_type=query_type)
541 if output_all_columns:
542 if isinstance(formatted_output, MutableMapping):
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:281, in Formatter.__call__(self, pa_table, query_type)
279 def __call__(self, pa_table: pa.Table, query_type: str) -> Union[RowFormat, ColumnFormat, BatchFormat]:
280 if query_type == "row":
--> 281 return self.format_row(pa_table)
282 elif query_type == "column":
283 return self.format_column(pa_table)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/torch_formatter.py:57, in TorchFormatter.format_row(self, pa_table)
56 def format_row(self, pa_table: pa.Table) -> dict:
---> 57 row = self.numpy_arrow_extractor().extract_row(pa_table)
58 return self.recursive_tensorize(row)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:154, in NumpyArrowExtractor.extract_row(self, pa_table)
153 def extract_row(self, pa_table: pa.Table) -> dict:
--> 154 return _unnest(self.extract_batch(pa_table))
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:160, in NumpyArrowExtractor.extract_batch(self, pa_table)
159 def extract_batch(self, pa_table: pa.Table) -> dict:
--> 160 return {col: self._arrow_array_to_numpy(pa_table[col]) for col in pa_table.column_names}
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:160, in <dictcomp>(.0)
159 def extract_batch(self, pa_table: pa.Table) -> dict:
--> 160 return {col: self._arrow_array_to_numpy(pa_table[col]) for col in pa_table.column_names}
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:196, in NumpyArrowExtractor._arrow_array_to_numpy(self, pa_array)
194 array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only).tolist()
195 if len(array) > 0:
--> 196 if any(
197 (isinstance(x, np.ndarray) and (x.dtype == np.object or x.shape != array[0].shape))
198 or (isinstance(x, float) and np.isnan(x))
199 for x in array
200 ):
201 return np.array(array, copy=False, **{**self.np_array_kwargs, "dtype": np.object})
202 return np.array(array, copy=False, **self.np_array_kwargs)
File /opt/conda/lib/python3.10/site-packages/datasets/formatting/formatting.py:197, in <genexpr>(.0)
194 array: List = pa_array.to_numpy(zero_copy_only=zero_copy_only).tolist()
195 if len(array) > 0:
196 if any(
--> 197 (isinstance(x, np.ndarray) and (x.dtype == np.object or x.shape != array[0].shape))
198 or (isinstance(x, float) and np.isnan(x))
199 for x in array
200 ):
201 return np.array(array, copy=False, **{**self.np_array_kwargs, "dtype": np.object})
202 return np.array(array, copy=False, **self.np_array_kwargs)
File /opt/conda/lib/python3.10/site-packages/numpy/__init__.py:324, in __getattr__(attr)
319 warnings.warn(
320 f"In the future `np.{attr}` will be defined as the "
321 "corresponding NumPy scalar.", FutureWarning, stacklevel=2)
323 if attr in __former_attrs__:
--> 324 raise AttributeError(__former_attrs__[attr])
326 if attr == 'testing':
327 import numpy.testing as testing
AttributeError: module 'numpy' has no attribute 'object'.
`np.object` was a deprecated alias for the builtin `object`. To avoid this error in existing code, use `object` by itself. Doing this will not modify any behavior and is safe.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
Please help me to resolve the above error
### Steps to reproduce the bug
Please resolve the issue of deprecated function np.object to object in the numpy
### Expected behavior
np.object should be written as object only
### Environment info
kaggle notebook | 6,669 |
https://github.com/huggingface/datasets/issues/6668 | Chapter 6 - Issue Loading `cnn_dailymail` dataset | [] | ### Describe the bug
So I am getting this bug when I try to run cell 4 of the Chapter 6 notebook code:
`dataset = load_dataset("ccdv/cnn_dailymail", version="3.0.0")`
Error Message:
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[4], line 4
1 #hide_output
2 from datasets import load_dataset
----> 4 dataset = load_dataset("ccdv/cnn_dailymail", version="3.0.0")
7 # dataset = load_dataset("ccdv/cnn_dailymail", version="3.0.0", trust_remote_code=True)
8 print(f"Features: {dataset['train'].column_names}")
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\load.py:2587, in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, ignore_verifications, keep_in_memory, save_infos, revision, token, use_auth_token, task, streaming, num_proc, storage_options, trust_remote_code, **config_kwargs)
2583 # Build dataset for splits
2584 keep_in_memory = (
2585 keep_in_memory if keep_in_memory is not None else is_small_dataset(builder_instance.info.dataset_size)
2586 )
-> 2587 ds = builder_instance.as_dataset(split=split, verification_mode=verification_mode, in_memory=keep_in_memory)
2588 # Rename and cast features to match task schema
2589 if task is not None:
2590 # To avoid issuing the same warning twice
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\builder.py:1244, in DatasetBuilder.as_dataset(self, split, run_post_process, verification_mode, ignore_verifications, in_memory)
1241 verification_mode = VerificationMode(verification_mode or VerificationMode.BASIC_CHECKS)
1243 # Create a dataset for each of the given splits
-> 1244 datasets = map_nested(
1245 partial(
1246 self._build_single_dataset,
1247 run_post_process=run_post_process,
1248 verification_mode=verification_mode,
1249 in_memory=in_memory,
1250 ),
1251 split,
1252 map_tuple=True,
1253 disable_tqdm=True,
1254 )
1255 if isinstance(datasets, dict):
1256 datasets = DatasetDict(datasets)
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\utils\py_utils.py:477, in map_nested(function, data_struct, dict_only, map_list, map_tuple, map_numpy, num_proc, parallel_min_length, types, disable_tqdm, desc)
466 mapped = [
467 map_nested(
468 function=function,
(...)
474 for obj in iterable
475 ]
476 elif num_proc != -1 and num_proc <= 1 or len(iterable) < parallel_min_length:
--> 477 mapped = [
478 _single_map_nested((function, obj, types, None, True, None))
479 for obj in hf_tqdm(iterable, disable=disable_tqdm, desc=desc)
480 ]
481 else:
482 with warnings.catch_warnings():
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\utils\py_utils.py:478, in <listcomp>(.0)
466 mapped = [
467 map_nested(
468 function=function,
(...)
474 for obj in iterable
475 ]
476 elif num_proc != -1 and num_proc <= 1 or len(iterable) < parallel_min_length:
477 mapped = [
--> 478 _single_map_nested((function, obj, types, None, True, None))
479 for obj in hf_tqdm(iterable, disable=disable_tqdm, desc=desc)
480 ]
481 else:
482 with warnings.catch_warnings():
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\utils\py_utils.py:370, in _single_map_nested(args)
368 # Singleton first to spare some computation
369 if not isinstance(data_struct, dict) and not isinstance(data_struct, types):
--> 370 return function(data_struct)
372 # Reduce logging to keep things readable in multiprocessing with tqdm
373 if rank is not None and logging.get_verbosity() < logging.WARNING:
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\builder.py:1274, in DatasetBuilder._build_single_dataset(self, split, run_post_process, verification_mode, in_memory)
1271 split = Split(split)
1273 # Build base dataset
-> 1274 ds = self._as_dataset(
1275 split=split,
1276 in_memory=in_memory,
1277 )
1278 if run_post_process:
1279 for resource_file_name in self._post_processing_resources(split).values():
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\builder.py:1348, in DatasetBuilder._as_dataset(self, split, in_memory)
1346 if self._check_legacy_cache():
1347 dataset_name = self.name
-> 1348 dataset_kwargs = ArrowReader(cache_dir, self.info).read(
1349 name=dataset_name,
1350 instructions=split,
1351 split_infos=self.info.splits.values(),
1352 in_memory=in_memory,
1353 )
1354 fingerprint = self._get_dataset_fingerprint(split)
1355 return Dataset(fingerprint=fingerprint, **dataset_kwargs)
File ~\anaconda3\envs\nlp-transformers\lib\site-packages\datasets\arrow_reader.py:254, in BaseReader.read(self, name, instructions, split_infos, in_memory)
252 if not files:
253 msg = f'Instruction "{instructions}" corresponds to no data!'
--> 254 raise ValueError(msg)
255 return self.read_files(files=files, original_instructions=instructions, in_memory=in_memory)
**ValueError: Instruction "validation" corresponds to no data!**
````
Looks like the data is not being loaded. Any advice would be appreciated. Thanks!
### Steps to reproduce the bug
Run all cells of Chapter 6 notebook.
### Expected behavior
Data should load correctly without any errors.
### Environment info
- `datasets` version: 2.17.0
- Platform: Windows-10-10.0.19045-SP0
- Python version: 3.9.18
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0 | 6,668 |
https://github.com/huggingface/datasets/issues/6667 | Default config for squad is incorrect | [
"you can try: pip install datasets==2.16.1"
] | ### Describe the bug
If you download Squad, it will download the plain_text version, but the config still specifies "default", so if you set the offline mode the cache will try to look it up according to the config_id which is "default" and this will say;
ValueError: Couldn't find cache for squad for config 'default'
Available configs in the cache: ['plain_text']
### Steps to reproduce the bug
1. export HF_DATASETS_OFFLINE=0
2. load_dataset("squad")
3. export HF_DATASETS_OFFLINE=1
4. load_dataset("squad")
### Expected behavior
We should change the config_name I guess?
### Environment info
linux, latest version of datasets | 6,667 |
https://github.com/huggingface/datasets/issues/6663 | `write_examples_on_file` and `write_batch` are broken in `ArrowWriter` | [
"Thanks for reporting! I've left some comments on the PR on how to fix this recent change rather than reverting it.",
"> Thanks for reporting! I've left some comments on the PR on how to fix this recent change rather than reverting it.\r\n\r\nI feel that'd be good, but it'd be great to release a hotfix ASAP (a revert is a fast thing to do) so people can continue using this library and then focus on still applying the improvement.",
"Fixed by #6664 "
] | ### Describe the bug
`write_examples_on_file` and `write_batch` are broken in `ArrowWriter` since #6636. The order between the columns and the schema is not preserved anymore. So these functions don't work anymore unless the order happens to align well.
### Steps to reproduce the bug
Try to do `write_batch` with anything that has many columns, and it's likely to break.
### Expected behavior
I expect these functions to work, instead of it trying to cast a column to its incorrect type.
### Environment info
- `datasets` version: 2.17.0
- Platform: Linux-5.15.0-1040-aws-x86_64-with-glibc2.35
- Python version: 3.10.13
- `huggingface_hub` version: 0.19.4
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0 | 6,663 |
https://github.com/huggingface/datasets/issues/6661 | Import error on Google Colab | [
"Hi! This can happen if an incompatible `pyarrow` version (`pyarrow<12.0.0`) has been imported before the `datasets` installation and the Colab session hasn't been restarted afterward. To avoid the error, go to \"Runtime -> Restart session\" after `!pip install -U datasets` and before `import datasets`, or insert the `import os; os.kill(os.getpid(), 9)` cell between `!pip install -U datasets` and `import datasets` to do the same programmatically.",
"One possible cause might be the one pointed out by @mariosasko above, and you get the following warning on Colab:\r\n```\r\nWARNING: The following packages were previously imported in this runtime:\r\n [pyarrow]\r\nYou must restart the runtime in order to use newly installed versions.\r\n```\r\n\r\nOn the other hand, if the old version of `pyarrow` is not previously imported (before the installation of `datasets`), the reported issue here is not reproducible: `datasets` can be installed, imported and used on Colab.",
"Duplicate of:\r\n- #5923",
"Google Colab now pre-installs PyArrow 14.0.2, making this issue unlikely to happen. So, I'm unpinning it."
] | ### Describe the bug
Cannot be imported on Google Colab, the import throws the following error:
ValueError: pyarrow.lib.IpcWriteOptions size changed, may indicate binary incompatibility. Expected 88 from C header, got 72 from PyObject
### Steps to reproduce the bug
1. `! pip install -U datasets`
2. `import datasets`
### Expected behavior
Should be possible to use the library
### Environment info
- `datasets` version: 2.17.0
- Platform: Linux-6.1.58+-x86_64-with-glibc2.35
- Python version: 3.10.12
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 1.5.3
- `fsspec` version: 2023.6.0 | 6,661 |
https://github.com/huggingface/datasets/issues/6657 | Release not pushed to conda channel | [
"Thanks for reporting, @atulsaurav.\r\n\r\nWe are investigating the issue. ",
"I can't fix this issue because I do not appear as a team member of the huggingface datasets project: https://anaconda.org/huggingface/datasets\r\n\r\n@lhoestq could you please add `datasets` team members to the corresponding Anaconda project?\r\n\r\nOnce this done, I could recreate and update the Anaconda token, as mentioned above it seems the current one has expired.",
"I think @LysandreJik has access ?",
"FYI it failed for 2.18.0 too: https://github.com/huggingface/datasets/actions/runs/8117132330/job/22188677936",
"We updated the token and I re-ran the conda releases :)"
] | ### Describe the bug
The github actions step to publish the release 2.17.0 to conda channel has failed due to expired token. Can some one please update the anaconda token rerun the failed action? @albertvillanova ?
### Steps to reproduce the bug
Please see this actions [link](https://github.com/huggingface/datasets/actions/runs/7842473662)
### Expected behavior
The action runs successfully and the latest release is pushed to HuggingFace conda channel
### Environment info
Not applicable. | 6,657 |
https://github.com/huggingface/datasets/issues/6656 | Error when loading a big local json file | [
"I get similar when dealing with a large jsonl file (6k lines), \r\n\r\n> TypeError: Couldn't cast array of type timestamp[us] to null\r\n\r\nYet when I split it into 1k lines, files, load_dataset works fine!\r\n\r\nhttps://github.com/huggingface/course/issues/692\r\n\r\n"
] | ### Describe the bug
When trying to load big json files from a local directory, `load_dataset` throws the following error
```
Traceback (most recent call last):
File "/miniconda3/envs/conda-env/lib/python3.10/site-packages/datasets/builder.py", line 1989, in _prepare_split_single
writer.write_table(table)
File "miniconda3/envs/conda-env/lib/python3.10/site-packages/datasets/arrow_writer.py", line 573, in write_table
pa_table = pa_table.combine_chunks()
File "pyarrow/table.pxi", line 3638, in pyarrow.lib.Table.combine_chunks
File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: offset overflow while concatenating arrays
```
### Steps to reproduce the bug
1. Download a big file, e.g. `https://dl.fbaipublicfiles.com/dpr/data/retriever/biencoder-nq-train.json.gz`
2. Load it like `data = load_dataset("json", data_files=["nq-train.json"], split="train")`
```python
from datasets import load_dataset
data = load_dataset("json", data_files=["nq-train.json"], split="train")
```
A similarly formatted but smaller file, e.g. e.g. `https://dl.fbaipublicfiles.com/dpr/data/retriever/biencoder-nq-dev.json.gz` is loaded without issues
```python
from datasets import load_dataset
data = load_dataset("json", data_files=["nq-dev.json"], split="train")
```
### Expected behavior
It should load normally
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-5.18.10-76051810-generic-x86_64-with-glibc2.31
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0 | 6,656 |
https://github.com/huggingface/datasets/issues/6655 | Cannot load the dataset go_emotions | [
"Thanks for reporting, @arame.\r\n\r\nI guess you have an old version of `transformers` (that submodule is present in `transformers` since version 3.0.1, since nearly 4 years ago). If you update it, the error should disappear:\r\n```shell\r\npip install -U transformers\r\n```\r\n\r\nOn the other hand, I am wondering: does it make sense to use `transformers` in this case, even if we don't need it to load the `go_emotions` dataset (already converted to Parquet files)?\r\n- Maybe @mariosasko can give some insight, as he included these code lines:\r\n - #6454\r\n\r\nhttps://github.com/huggingface/datasets/blob/9751fb14594d354e952f0ebdfaf31cb203b011e7/src/datasets/utils/_dill.py#L60-L63\r\n",
"The linked code lazily registers a custom reducer for `transformers.PreTrainedTokenizerBase` only if `transformers` have already been imported (imports are expensive, so we check `sys.modules`).\r\n\r\nHowever, the logic does not account for `transformers<3`, so we should add a version check to fix that.",
"> The linked code lazily registers a custom reducer for `transformers.PreTrainedTokenizerBase` only if `transformers` have already been imported (imports are expensive, so we check `sys.modules`).\r\n> \r\n> However, the logic does not account for `transformers<3`, so we should add a version check to fix that.\r\n\r\nThank you for that Mario. Would this fix solve the problem and do you have any idea when it will be done? \r\nI tried the pip install suggested by Albert and it made no difference.",
"I tried running the code today and the problem appears to be fixed."
] | ### Describe the bug
When I run the following code I get an exception;
`go_emotions = load_dataset("go_emotions")`
> AttributeError Traceback (most recent call last)
Cell In[6], [line 1](vscode-notebook-cell:?execution_count=6&line=1)
----> [1](vscode-notebook-cell:?execution_count=6&line=1) go_emotions = load_dataset("go_emotions")
[2](vscode-notebook-cell:?execution_count=6&line=2) data = go_emotions.data
File [c:\Users\hijik\anaconda3\Lib\site-packages\datasets\load.py:2523](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2523), in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, ignore_verifications, keep_in_memory, save_infos, revision, token, use_auth_token, task, streaming, num_proc, storage_options, trust_remote_code, **config_kwargs)
[2518](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2518) verification_mode = VerificationMode(
[2519](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2519) (verification_mode or VerificationMode.BASIC_CHECKS) if not save_infos else VerificationMode.ALL_CHECKS
[2520](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2520) )
[2522](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2522) # Create a dataset builder
-> [2523](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2523) builder_instance = load_dataset_builder(
[2524](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2524) path=path,
[2525](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2525) name=name,
[2526](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2526) data_dir=data_dir,
[2527](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2527) data_files=data_files,
[2528](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2528) cache_dir=cache_dir,
[2529](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2529) features=features,
[2530](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2530) download_config=download_config,
[2531](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2531) download_mode=download_mode,
[2532](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2532) revision=revision,
[2533](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2533) token=token,
[2534](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2534) storage_options=storage_options,
[2535](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2535) trust_remote_code=trust_remote_code,
[2536](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/load.py:2536) _require_default_config_name=name is None,
...
---> [63](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/utils/_dill.py:63) if issubclass(obj_type, transformers.PreTrainedTokenizerBase):
[64](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/utils/_dill.py:64) pklregister(obj_type)(_save_transformersPreTrainedTokenizerBase)
[66](file:///C:/Users/hijik/anaconda3/Lib/site-packages/datasets/utils/_dill.py:66) # Unwrap `torch.compile`-ed functions
AttributeError: module 'transformers' has no attribute 'PreTrainedTokenizerBase'
Output is truncated. View as a [scrollable element](command:cellOutput.enableScrolling?10bc0728-6947-456e-9a3e-f056872b04c6) or open in a [text editor](command:workbench.action.openLargeOutput?10bc0728-6947-456e-9a3e-f056872b04c6). Adjust cell output [settings](command:workbench.action.openSettings?%5B%22%40tag%3AnotebookOutputLayout%22%5D)...
### Steps to reproduce the bug
```
from datasets import load_dataset
go_emotions = load_dataset("go_emotions")
```
### Expected behavior
Should simply load the variable with the data from the file
### Environment info
Copy-and-paste the text below in your GitHub issue.
- `datasets` version: 2.16.1
- Platform: Windows-10-10.0.22631-SP0
- Python version: 3.11.4
- `huggingface_hub` version: 0.20.3
- PyArrow version: 11.0.0
- Pandas version: 1.5.3
- `fsspec` version: 2023.10.0 | 6,655 |
https://github.com/huggingface/datasets/issues/6654 | Batched dataset map throws exception that cannot cast fixed length array to Sequence | [
"Hi ! This issue has been fixed by https://github.com/huggingface/datasets/pull/6283\r\n\r\nCan you try again with the new release 2.17.0 ?\r\n\r\n```\r\npip install -U datasets\r\n```\r\n\r\n",
"Amazing! It's indeed fixed now. Thanks!"
] | ### Describe the bug
I encountered a TypeError when batch processing a dataset with Sequence features in datasets package version 2.16.1. The error arises from a mismatch in handling fixed-size list arrays during the map function execution. Debugging pinpoints the issue to an if-statement in datasets/table.py, line 2093, failing to correctly process sequence lengths.
### Steps to reproduce the bug
Create virtual environment and activate
```
virtualenv venv
source venv/bin/activate
```
Then install the datasets package (I'm using the latest version)
```
pip install datasets==2.16.1
```
Then run
```python
# bug.py
from datasets import Dataset
from datasets.features import Features, Sequence, Value
data = {
"num": [[1, 2], [3, 4]],
}
features = Features({'num': Sequence(feature=Value(dtype='int32'), length=2)})
dataset = Dataset.from_dict(data, features=features)
dataset.map(lambda x: x, batched=True, batch_size=1)
```
### Expected behavior
I get the following stack trace
```
Map: 50%|βββββ | 1/2 [00:00<00:00, 423.92 examples/s]
Traceback (most recent call last):
File "/PATH/TO/BUG_PORT/bug.py", line 9, in <module>
dataset.map(lambda x: x, batched=True, batch_size=1)
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 592, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 557, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 3093, in map
for rank, done, content in Dataset._map_single(**dataset_kwargs):
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 3489, in _map_single
writer.write_batch(batch)
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 551, in write_batch
array = cast_array_to_feature(col_values, col_type) if col_type is not None else col_values
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/table.py", line 1797, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/table.py", line 1797, in <listcomp>
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
File "/PATH/TO/BUG_PORT/venv/lib/python3.9/site-packages/datasets/table.py", line 2111, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{array.type}\nto\n{feature}")
TypeError: Couldn't cast array of type
fixed_size_list<item: int32>[2]
to
Sequence(feature=Value(dtype='int32', id=None), length=2, id=None)
```
After some debugging, I found that the if-statement that is actually failing is line 2093 in `datasets/table.py`
```python
# datasets/table.py
...
2093 if feature.length * len(array) == len(array_values):
2094 return pa.FixedSizeListArray.from_arrays(_c(array_values, feature.feature), feature.length)
...
```
### Environment info
Platform: MacOS
Datasets version: datasets==2.16.1
Python version: 3.9.6 | 6,654 |
https://github.com/huggingface/datasets/issues/6651 | Slice splits support for datasets.load_from_disk | [] | ### Feature request
Support for slice splits in `datasets.load_from_disk`, similar to how it's already supported for `datasets.load_dataset`.
### Motivation
Slice splits are convienient in a numer of cases - adding support to `datasets.load_from_disk` would make working with local datasets easier and homogenize the APIs of load_from_disk and load_dataset.
### Your contribution
Sure, if the devs think the feature request is sensible. | 6,651 |
https://github.com/huggingface/datasets/issues/6650 | AttributeError: 'InMemoryTable' object has no attribute '_batches' | [
"Hi! Does running the following code also return the same error on your machine? \r\n\r\n```python\r\nimport copy\r\nimport pyarrow as pa\r\nfrom datasets.table import InMemoryTable\r\n\r\ncopy.deepcopy(InMemoryTable(pa.table({\"a\": [1, 2, 3], \"b\": [\"foo\", \"bar\", \"foobar\"]})))\r\n```",
"No, it doesn't, it runs fine. But what's really strange is that the error just went away after I reran the data prep script for conversion from csv to a datasets object. I realize that's not very helpful since the problem isn't reproducible. ",
"Feel free to close the issue then :)."
] | ### Describe the bug
```
Traceback (most recent call last):
File "finetune.py", line 103, in <module>
main(args)
File "finetune.py", line 45, in main
data_tokenized = data.map(partial(funcs.tokenize_function, tokenizer,
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/dataset_dict.py", line 868, in map
{
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/dataset_dict.py", line 869, in <dictcomp>
k: dataset.map(
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 592, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 557, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 3093, in map
for rank, done, content in Dataset._map_single(**dataset_kwargs):
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 3432, in _map_single
arrow_formatted_shard = shard.with_format("arrow")
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 2667, in with_format
dataset = copy.deepcopy(self)
File "/opt/conda/envs/ptca/lib/python3.8/copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
File "/opt/conda/envs/ptca/lib/python3.8/copy.py", line 270, in _reconstruct
state = deepcopy(state, memo)
File "/opt/conda/envs/ptca/lib/python3.8/copy.py", line 146, in deepcopy
y = copier(x, memo)
File "/opt/conda/envs/ptca/lib/python3.8/copy.py", line 230, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
File "/opt/conda/envs/ptca/lib/python3.8/copy.py", line 153, in deepcopy
y = copier(memo)
File "/opt/conda/envs/ptca/lib/python3.8/site-packages/datasets/table.py", line 176, in __deepcopy__
memo[id(self._batches)] = list(self._batches)
AttributeError: 'InMemoryTable' object has no attribute '_batches'
```
### Steps to reproduce the bug
I'm running an MLOps flow using AzureML.
The error appears when I run the following function in my training script:
```python
data_tokenized = data.map(partial(funcs.tokenize_function, tokenizer,
seq_length),
batched=True,
batch_size=batch_size,
remove_columns=['col1', 'col2'])
```
```python
def tokenize_function(tok, seq_length, example)
# Pad so that each batch has the same sequence length
inp = tok(example['col1'], padding=True, truncation=True)
outp = tok(example['col2'], padding="max_length", max_length=seq_length)
res = {
'input_ids': inp['input_ids'],
'attention_mask': inp['attention_mask'],
'decoder_input_ids': outp['input_ids'],
'labels': outp['input_ids'],
'decoder_attention_mask': outp['attention_mask']
}
return res
```
### Expected behavior
Processing proceeds without errors. I ran this same workflow 2 weeks ago without a problem. I recreated the environment since then but it doesn't appear that datasets versions have changed since Dec. '23.
### Environment info
datasets 2.16.1
transformers 4.35.2
pyarrow 15.0.0
pyarrow-hotfix 0.6
torch 2.0.1
I'm not using the latest transformers version because there was an error due to a conflict with Azure mlflow when I tried the last time. | 6,650 |
https://github.com/huggingface/datasets/issues/6645 | Support fsspec 2024.2 | [
"I'd be very grateful. This upper bound banished me straight into dependency hell today. :("
] | Support fsspec 2024.2.
First, we should address:
- #6644 | 6,645 |
https://github.com/huggingface/datasets/issues/6644 | Support fsspec 2023.12 | [
"The pinned fsspec version range dependency conflict has been affecting several of our users in https://github.com/iterative/dvc. I've opened an initial PR that I think should resolve the glob behavior changes with using datasets + the latest fsspec release.\r\n\r\nPlease let us know if there's any other fsspec related behavior in datasets that needs to be updated to get 2024.2 supported, we'd like to get this conflict resolved as quickly as possible and we're willing to contribute any additional work that's required here.\r\n\r\ncc @dberenbaum"
] | Support fsspec 2023.12 by handling previous and new glob behavior. | 6,644 |
https://github.com/huggingface/datasets/issues/6643 | Faiss GPU index cannot be serialised when passed to trainer | [
"Hi ! make sure your query embeddings are numpy arrays, not torch tensors ;)",
"Hi Quentin, not sure how that solves the problem number 1. I am trying to pass on a dataset with a faiss gpu for training to the standard trainer but getting this serialisation error. What is a workaround this? I do not want to remove the faiss index, as I would want to use it to create batches of retrieved samples from the dataset. \r\nThanks in advance for your help!",
"Issue number one seems to be an issue with FAISS indexes not being compatible with copy.deepcopy.\r\n\r\nMaybe you try to not remove the columns, e.g. by passing `remove_unused_columns=False`"
] | ### Describe the bug
I am working on a retrieval project and encountering I have encountered two issues in the hugging face faiss integration:
1. I am trying to pass in a dataset with a faiss index to the Huggingface trainer. The code works for a cpu faiss index, but doesn't for a gpu one, getting error:
```
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/transformers/trainer.py", line 1543, in train
return inner_training_loop(
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/transformers/trainer.py", line 1555, in _inner_training_loop
train_dataloader = self.get_train_dataloader()
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/transformers/trainer.py", line 831, in get_train_dataloader
train_dataset = self._remove_unused_columns(train_dataset, description="training")
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/transformers/trainer.py", line 725, in _remove_unused_columns
return dataset.remove_columns(ignored_columns)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 592, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 557, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/fingerprint.py", line 481, in wrapper
out = func(dataset, *args, **kwargs)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 2146, in remove_columns
dataset = copy.deepcopy(self)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 146, in deepcopy
y = copier(x, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 146, in deepcopy
y = copier(x, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 146, in deepcopy
y = copier(x, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/copy.py", line 161, in deepcopy
rv = reductor(4)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/faiss/__init__.py", line 556, in index_getstate
return {"this": serialize_index(self).tobytes()}
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/faiss/__init__.py", line 1607, in serialize_index
write_index(index, writer)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/faiss/swigfaiss.py", line 9843, in write_index
return _swigfaiss.write_index(*args)
RuntimeError: Error in void faiss::write_index(const faiss::Index*, faiss::IOWriter*) at /project/faiss/faiss/impl/index_write.cpp:590: don't know how to serialize this type of index
```
The index was created with the add_faiss_index method
```
train_dataset.add_faiss_index(
column='embeddings',
index_name='embeddings',
string_factory=faiss_index_string,
train_size=config.faiss_train_size,
device=0, # Use -1 for CPU, or specify GPU device ID
faiss_verbose=True
)
```
2. Athough faiss is written to be compatible on the gpu for searching [https://github.com/facebookresearch/faiss/wiki/Faiss-on-the-GPU](https://github.com/facebookresearch/faiss/wiki/Faiss-on-the-GPU) I am getting error when trying to use the hugggingface code to do the search on gpu. This seems to be caused by this line https://github.com/huggingface/datasets/blob/f9975f636542df7f95c27065ea93147440d690b7/src/datasets/search.py#L376 producing error
```
total_scores, total_examples = self.dataset.get_nearest_examples_batch('embeddings', embeddings, k=self.k)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/search.py", line 773, in get_nearest_examples_batch
total_scores, total_indices = self.search_batch(index_name, queries, k, **kwargs)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/search.py", line 727, in search_batch
return self._indexes[index_name].search_batch(queries, k, **kwargs)
File "/users/rubman/.conda/envs/protein_npt_env/lib/python3.10/site-packages/datasets/search.py", line 376, in search_batch
if not queries.flags.c_contiguous:
AttributeError: 'Tensor' object has no attribute 'flags'
```
### Steps to reproduce the bug
```
train_dataset.add_faiss_index(
column='embeddings',
index_name='embeddings',
string_factory=faiss_index_string,
train_size=config.faiss_train_size,
device=0, # Use -1 for CPU, or specify GPU device ID
faiss_verbose=True
)
Trainer(
model=model,
args=args,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
data_collator=data_collator,
tokenizer=tokenizer
)
train_dataset.get_nearest_examples_batch('embeddings', embeddings, k=self.k)
```
### Expected behavior
I would expect the faiss database code to be gpu compatible
### Environment info
huggingface Version: 2.16.1 | 6,643 |
https://github.com/huggingface/datasets/issues/6642 | Differently dataset object saved than it is loaded. | [
"I see now, that I have to use `load_from_disk`, in order to load dataset properly, not `load_dataset`. Why is this behavior split? Why do we need both, `load_dataset` and `load_from_disk`?\r\n\r\nUnless answered, I believe this might be helpful for other hf datasets newbies.\r\n\r\nAnyway, made a `load_dataset` compatible dataset in a following way. I created a directory, and just copied jsonl there as `train.jsonl/test.jsonl`.\r\n```python\r\noutput_folder = os.path.join(args.output_folder, f\"{task_meta_type}_{task_type}\")\r\nos.makedirs(output_folder, exist_ok=True)\r\nfile = f\"{task_meta_type}_{task_type}_train.jsonl\"\r\nshutil.copy(os.path.join(input_folder, file),\r\n os.path.join(output_folder, \"train.jsonl\"))\r\n# now test\r\nfile = f\"{task_meta_type}_{task_type}_test.jsonl\"\r\nshutil.copy(os.path.join(input_folder, file),\r\n os.path.join(output_folder, \"test.jsonl\"))\r\n```\r\n",
"Hi @MFajcik, \r\n\r\nYou can find information about save_to_disk/load_from_disk in our docs:\r\n- https://huggingface.co/docs/datasets/v2.16.1/en/process#save\r\n- https://huggingface.co/docs/datasets/v2.16.1/en/package_reference/main_classes#datasets.Dataset.save_to_disk\r\n- https://huggingface.co/docs/datasets/v2.16.1/en/package_reference/main_classes#datasets.Dataset.load_from_disk"
] | ### Describe the bug
Differently sized object is saved than it is loaded.
### Steps to reproduce the bug
Hi, I save dataset in a following way:
```
dataset = load_dataset("json",
data_files={
"train": os.path.join(input_folder, f"{task_meta_type}_{task_type}_train.jsonl"),
"test": os.path.join(input_folder, f"{task_meta_type}_{task_type}_test.jsonl")})
print(os.path.join(output_folder, f"{task_meta_type}_{task_type}"))
print(f"Length of train dataset: {len(dataset['train'])}")
print(f"Length of test dataset: {len(dataset['test'])}")
dataset.save_to_disk(os.path.join(output_folder, f"{task_meta_type}_{task_type}"))
```
this yields output
```
.data/hf_dataset/propaganda_zanr
Length of train dataset: 7642
Length of test dataset: 1000
```
Everything looks fine.
Then I load the dataset
```python
from datasets import load_dataset
dataset_path = ".data/hf_dataset/propaganda_zanr"
dataset = load_dataset(dataset_path)
print(f"Length of train dataset: {len(dataset['train'])}")
print(f"Length of test dataset: {len(dataset['test'])}")
```
this prints
```
Generating train split: 1 examples [00:00, 72.10 examples/s]
Generating test split: 1 examples [00:00, 100.69 examples/s]
Length of train dataset: 1
Length of test dataset: 1
```
I dont' understand :(
### Expected behavior
same object is loaded
### Environment info
datasets==2.16.1 | 6,642 |
https://github.com/huggingface/datasets/issues/6641 | unicodedecodeerror: 'utf-8' codec can't decode byte 0xac in position 25: invalid start byte | [
"Hi @Hughhuh. \r\n\r\nI have formatted the issue because it was not easily readable. Additionally, the environment info is incomplete: it seems you did not run the proposed CLI command `datasets-cli env` and essential information is missing: version of `datasets`, version of `pyarrow`,...\r\n\r\nWith the information you provided, it seems an issue with the specific \"samsum\" dataset. I'm transferring the issue to the corresponding dataset page: https://huggingface.co/datasets/samsum/discussions/5"
] | ### Describe the bug
unicodedecodeerror: 'utf-8' codec can't decode byte 0xac in position 25: invalid start byte
### Steps to reproduce the bug
```
import sys
sys.getdefaultencoding()
'utf-8'
from datasets import load_dataset
print(f"Train dataset size: {len(dataset['train'])}")
print(f"Test dataset size: {len(dataset['test'])}")
Resolving data files: 100%
159/159 [00:00<00:00, 9909.28it/s]
Using custom data configuration samsum-0b1209637541c9e6
Downloading and preparing dataset json/samsum to C:/Users/Administrator/.cache/huggingface/datasets/json/samsum-0b1209637541c9e6/0.0.0/0f7e3662623656454fcd2b650f34e886a7db4b9104504885bd462096cc7a9f51...
Downloading data files: 100%
3/3 [00:00<00:00, 119.99it/s]
Extracting data files: 100%
3/3 [00:00<00:00, 9.54it/s]
Generating train split:
88392/0 [00:15<00:00, 86848.17 examples/s]
Generating test split:
0/0 [00:00<?, ? examples/s]
---------------------------------------------------------------------------
ArrowInvalid Traceback (most recent call last)
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\packaged_modules\json\json.py:132, in Json._generate_tables(self, files)
131 try:
--> 132 pa_table = paj.read_json(
133 io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
134 )
135 break
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\pyarrow\_json.pyx:290, in pyarrow._json.read_json()
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\pyarrow\error.pxi:144, in pyarrow.lib.pyarrow_internal_check_status()
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\pyarrow\error.pxi:100, in pyarrow.lib.check_status()
ArrowInvalid: JSON parse error: Invalid value. in row 0
During handling of the above exception, another exception occurred:
UnicodeDecodeError Traceback (most recent call last)
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\builder.py:1819, in ArrowBasedBuilder._prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)
1818 _time = time.time()
-> 1819 for _, table in generator:
1820 if max_shard_size is not None and writer._num_bytes > max_shard_size:
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\packaged_modules\json\json.py:153, in Json._generate_tables(self, files)
152 with open(file, encoding="utf-8") as f:
--> 153 dataset = json.load(f)
154 except json.JSONDecodeError:
File ~\AppData\Local\Programs\Python\Python310\lib\json\__init__.py:293, in load(fp, cls, object_hook, parse_float, parse_int, parse_constant, object_pairs_hook, **kw)
276 """Deserialize ``fp`` (a ``.read()``-supporting file-like object containing
277 a JSON document) to a Python object.
278
(...)
291 kwarg; otherwise ``JSONDecoder`` is used.
292 """
--> 293 return loads(fp.read(),
294 cls=cls, object_hook=object_hook,
295 parse_float=parse_float, parse_int=parse_int,
296 parse_constant=parse_constant, object_pairs_hook=object_pairs_hook, **kw)
File ~\AppData\Local\Programs\Python\Python310\lib\codecs.py:322, in BufferedIncrementalDecoder.decode(self, input, final)
321 data = self.buffer + input
--> 322 (result, consumed) = self._buffer_decode(data, self.errors, final)
323 # keep undecoded input until the next call
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xac in position 25: invalid start byte
The above exception was the direct cause of the following exception:
DatasetGenerationError Traceback (most recent call last)
Cell In[81], line 5
1 from datasets import load_dataset
3 # Load dataset from the hub
4 #dataset = load_dataset("json",data_files="C:/Users/Administrator/Desktop/samsum/samsum/data/corpus/train.json",field="data")
----> 5 dataset = load_dataset('json',"samsum")
6 #dataset = load_dataset("samsum")
7 print(f"Train dataset size: {len(dataset['train'])}")
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\load.py:1758, in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, num_proc, **config_kwargs)
1755 try_from_hf_gcs = path not in _PACKAGED_DATASETS_MODULES
1757 # Download and prepare data
-> 1758 builder_instance.download_and_prepare(
1759 download_config=download_config,
1760 download_mode=download_mode,
1761 ignore_verifications=ignore_verifications,
1762 try_from_hf_gcs=try_from_hf_gcs,
1763 num_proc=num_proc,
1764 )
1766 # Build dataset for splits
1767 keep_in_memory = (
1768 keep_in_memory if keep_in_memory is not None else is_small_dataset(builder_instance.info.dataset_size)
1769 )
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\builder.py:860, in DatasetBuilder.download_and_prepare(self, output_dir, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, file_format, max_shard_size, num_proc, storage_options, **download_and_prepare_kwargs)
858 if num_proc is not None:
859 prepare_split_kwargs["num_proc"] = num_proc
--> 860 self._download_and_prepare(
861 dl_manager=dl_manager,
862 verify_infos=verify_infos,
863 **prepare_split_kwargs,
864 **download_and_prepare_kwargs,
865 )
866 # Sync info
867 self.info.dataset_size = sum(split.num_bytes for split in self.info.splits.values())
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\builder.py:953, in DatasetBuilder._download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
949 split_dict.add(split_generator.split_info)
951 try:
952 # Prepare split will record examples associated to the split
--> 953 self._prepare_split(split_generator, **prepare_split_kwargs)
954 except OSError as e:
955 raise OSError(
956 "Cannot find data file. "
957 + (self.manual_download_instructions or "")
958 + "\nOriginal error:\n"
959 + str(e)
960 ) from None
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\builder.py:1708, in ArrowBasedBuilder._prepare_split(self, split_generator, file_format, num_proc, max_shard_size)
1706 gen_kwargs = split_generator.gen_kwargs
1707 job_id = 0
-> 1708 for job_id, done, content in self._prepare_split_single(
1709 gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
1710 ):
1711 if done:
1712 result = content
File ~\AppData\Local\Programs\Python\Python310\lib\site-packages\datasets\builder.py:1851, in ArrowBasedBuilder._prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)
1849 if isinstance(e, SchemaInferenceError) and e.__context__ is not None:
1850 e = e.__context__
-> 1851 raise DatasetGenerationError("An error occurred while generating the dataset") from e
1853 yield job_id, True, (total_num_examples, total_num_bytes, writer._features, num_shards, shard_lengths)
DatasetGenerationError: An error occurred while generating the dataset
```
### Expected behavior
can't load dataset
### Environment info
dataset:samsum
system :win10
gpu:m40 24G | 6,641 |
https://github.com/huggingface/datasets/issues/6640 | Sign Language Support | [] | ### Feature request
Currently, there are only several Sign Language labels, I would like to propose adding all the Signed Languages as new labels which are described in this ISO standard: https://www.evertype.com/standards/iso639/sign-language.html
### Motivation
Datasets currently only have labels for several signed languages. There are more signed languages in the world. Furthermore, some signed languages that have a lot of online data cannot be found because of this reason (for instance, German Sign Language, and there is no German Sign Language label on huggingface datasets even though there are a lot of readily available sign language datasets exist for German Sign Language, which are used very frequently in Sign Language Processing papers, and models.)
### Your contribution
I can submit a PR for this as well, adding the ISO codes and languages to the labels in datasets. | 6,640 |
https://github.com/huggingface/datasets/issues/6638 | Cannot download wmt16 dataset | [
"Looks like it works with latest datasets repository\r\n```\r\n- `datasets` version: 2.16.2.dev0\r\n- Platform: Linux-5.15.0-92-generic-x86_64-with-glibc2.29\r\n- Python version: 3.8.10\r\n- `huggingface_hub` version: 0.20.3\r\n- PyArrow version: 15.0.0\r\n- Pandas version: 2.0.1\r\n- `fsspec` version: 2023.10.0\r\n```\r\n\r\nCould you explain which is the minimum version that fixes this?\r\nEdit: Looks like that's 2.16.0, will close out issue"
] | ### Describe the bug
As of this morning (PST) 2/1/2024, seeing the wmt16 dataset is missing from opus , could you suggest an alternative?
```
Downloading data files: 0%| | 0/4 [00:00<?, ?it/s]Traceback (most recent call last):
File "test.py", line 2, in <module>
raw_datasets = load_dataset("wmt16","ro-en",split="train")
File "/usr/local/lib/python3.8/dist-packages/datasets/load.py", line 2153, in load_dataset
builder_instance.download_and_prepare(
File "/usr/local/lib/python3.8/dist-packages/datasets/builder.py", line 954, in download_and_prepare
self._download_and_prepare(
File "/usr/local/lib/python3.8/dist-packages/datasets/builder.py", line 1717, in _download_and_prepare
super()._download_and_prepare(
File "/usr/local/lib/python3.8/dist-packages/datasets/builder.py", line 1027, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/root/.cache/huggingface/modules/datasets_modules/datasets/wmt16/746749a11d25c02058042da7502d973ff410e73457f3d305fc1177dc0e8c4227/wmt_utils.py", line 754, in _split_generators
downloaded_files = dl_manager.download_and_extract(urls_to_download)
File "/usr/local/lib/python3.8/dist-packages/datasets/download/download_manager.py", line 565, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/usr/local/lib/python3.8/dist-packages/datasets/download/download_manager.py", line 428, in download
downloaded_path_or_paths = map_nested(
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/py_utils.py", line 464, in map_nested
mapped = [
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/py_utils.py", line 465, in <listcomp>
_single_map_nested((function, obj, types, None, True, None))
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/py_utils.py", line 384, in _single_map_nested
mapped = [_single_map_nested((function, v, types, None, True, None)) for v in pbar]
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/py_utils.py", line 384, in <listcomp>
mapped = [_single_map_nested((function, v, types, None, True, None)) for v in pbar]
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/py_utils.py", line 367, in _single_map_nested
return function(data_struct)
File "/usr/local/lib/python3.8/dist-packages/datasets/download/download_manager.py", line 454, in _download
return cached_path(url_or_filename, download_config=download_config)
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/file_utils.py", line 182, in cached_path
output_path = get_from_cache(
File "/usr/local/lib/python3.8/dist-packages/datasets/utils/file_utils.py", line 596, in get_from_cache
raise FileNotFoundError(f"Couldn't find file at {url}")
FileNotFoundError: Couldn't find file at https://opus.nlpl.eu/download.php?f=SETIMES/v2/tmx/en-ro.tmx.gz
```
### Steps to reproduce the bug
```
from datasets import load_dataset
raw_datasets = load_dataset("wmt16","ro-en",split="train")
```
### Expected behavior
Expect the dataset to be downloaded/ at least a clean exit with error explaining dataset is missing and a suggestion for next steps
### Environment info
- `datasets` version: 2.14.7
- Platform: Linux-5.15.0-92-generic-x86_64-with-glibc2.29
- Python version: 3.8.10
- Huggingface_hub version: 0.17.3
- PyArrow version: 15.0.0
- Pandas version: 2.0.1
| 6,638 |
https://github.com/huggingface/datasets/issues/6637 | 'with_format' is extremely slow when used together with 'interleave_datasets' or 'shuffle' on IterableDatasets | [
"The \"torch\" formatting is usually fast because we do zero-copy conversion from the Arrow data on your disk to Torch tensors. However IterableDataset shuffling seems to do data copies that slow down the pipeline, and it shuffles python objects instead of Arrow data.\r\n\r\nTo fix this we need to implement `BufferShuffledExamplesIterable.iter_arrow()` (same as regular `BufferShuffledExamplesIterable.__iter__()` but yields Arrow tables)\r\n\r\nhttps://github.com/huggingface/datasets/blob/b7d854b7fd3e9a330e21b76ee8421d4a7ebb4a7a/src/datasets/iterable_dataset.py#L968-L974\r\n"
] | ### Describe the bug
If you:
1. Interleave two iterable datasets together with the interleave_datasets function, or shuffle an iterable dataset
2. Set the output format to torch tensors with .with_format('torch')
Then iterating through the dataset becomes over 100x slower than it is if you don't apply the torch formatting.
### Steps to reproduce the bug
```python
import datasets
import torch
from tqdm import tqdm
rand_a = torch.randn(3,224,224)
rand_b = torch.randn(3,224,224)
a = torch.stack([rand_a] * 1000)
b = torch.stack([rand_b] * 1000)
features = datasets.Features({"tensor": datasets.Array3D(shape=(3,224,224), dtype="float32")})
ds_a = datasets.Dataset.from_dict({"tensor": a}, features=features).to_iterable_dataset()
ds_b = datasets.Dataset.from_dict({"tensor": b}, features=features).to_iterable_dataset()
# Iterating through either dataset with torch formatting is really fast (2000it/s on my machine)
for example in tqdm(ds_a.with_format('torch')):
pass
# Iterating through either dataset shuffled is also pretty fast (100it/s on my machine)
for example in tqdm(ds_a.shuffle()):
pass
# Iterating through this interleaved dataset is pretty fast (200it/s on my machine)
ds_fast = datasets.interleave_datasets([ds_a, ds_b])
for example in tqdm(ds_fast):
pass
# Iterating through either dataset with torch formatting *after shuffling* is really slow... (<2it/s on my machine)
for example in tqdm(ds_a.shuffle().with_format('torch')):
pass
# Iterating through this torch formatted interleaved dataset is also really slow (<2it/s on my machine)...
ds_slow = datasets.interleave_datasets([ds_a, ds_b]).with_format('torch')
for example in tqdm(ds_slow):
pass
# Even doing this is way faster!! (70it/s on my machine)
for example in tqdm(ds_fast):
test = torch.tensor(example['tensor'])
```
### Expected behavior
Applying torch formatting to the interleaved dataset shouldn't increase the time taken to iterate through the dataset by very much, since even explicitly converting every example is over 70x faster than calling .with_format('torch').
### Environment info
- `datasets` version: 2.16.1
- Platform: Linux-6.5.0-15-generic-x86_64-with-glibc2.38
- Python version: 3.11.6
- `huggingface_hub` version: 0.20.3
- PyArrow version: 15.0.0
- Pandas version: 2.2.0
- `fsspec` version: 2023.10.0
| 6,637 |
https://github.com/huggingface/datasets/issues/6624 | How to download the laion-coco dataset | [
"Hi, this dataset has been disabled by the authors, so unfortunately it's no longer possible to download it."
] | The laion coco dataset is not available now. How to download it
https://huggingface.co/datasets/laion/laion-coco | 6,624 |
https://github.com/huggingface/datasets/issues/6623 | streaming datasets doesn't work properly with multi-node | [
"@mariosasko, @lhoestq, @albertvillanova\r\nhey guys! can anyone help? or can you guys suggest who can help with this?",
"Hi ! \r\n\r\n1. When the dataset is running of of examples, the last batches received by the GPU can be incomplete or empty/missing. We haven't implemented yet a way to ignore the last batch. It might require the datasets to provide the number of examples per shard though, so that we can know when to stop.\r\n2. Samplers are not compatible with IterableDatasets in pytorch\r\n3. if `dataset.n_shards % world_size != 0` then all the nodes will read/stream the full dataset in order (possibly reading/streaming the same data multiple times), BUT will only yield one example out of `world_size` so that each example goes to one exactly one GPU.\r\n4. no, sharding should be down up-front and can take some time depending on the dataset size and format",
"> if dataset.n_shards % world_size != 0 then all the nodes will read/stream the full dataset in order (possibly reading/streaming the same data multiple times), BUT will only yield one example out of world_size so that each example goes to one exactly one GPU.\r\n\r\nconsidering there's just 1 shard and 2 worker nodes, do you mean each worker node will load the whole dataset but still receive half of that shard while streaming?",
"Yes both nodes will stream from the 1 shard, but each node will skip half of the examples. This way in total each example is seen once and exactly once during you distributed training.\r\n\r\nThough it terms of I/O, the dataset is effectively read/streamed twice.",
"what if the number of samples in that shard % num_nodes != 0? it will break/get stuck? or is the data repeated in that case for gradient sync?",
"In the case one at least one of the noes will get an empty/incomplete batch. The data is not repeated in that case. If the training loop doesn't take this into account it can lead to unexpected behaviors indeed.\r\n\r\nIn the future we'd like to add a feature that would allow the nodes to ignore the last batch, this way all the nodes would only have full batches.",
"> In the case one at least one of the noes will get an empty/incomplete batch. The data is not repeated in that case. If the training loop doesn't take this into account it can lead to unexpected behaviors indeed.\r\n> \r\n> In the future we'd like to add a feature that would allow the nodes to ignore the last batch, this way all the nodes would only have full batches.\r\n\r\nIs there any method to modify one dataset's n_shard? modify the number of files is ok? one file == one shard?",
"> modify the number of files is ok? one file == one shard?\r\n\r\nYep, one file == one shard :)"
] | ### Feature request
Letβs say I have a dataset with 5 samples with values [1, 2, 3, 4, 5], with 2 GPUs (for DDP) and batch size of 2. This dataset is an `IterableDataset` since I am streaming it.
Now I split the dataset using `split_dataset_by_node` to ensure it doesnβt get repeated. And since itβs already splitted, I donβt have to use `DistributedSampler` (also they don't work with iterable datasets anyway)?
But in this case I noticed that the:
First iteraton:
first GPU will get β [1, 2]
first GPU will get β [3, 4]
Second iteraton:
first GPU will get β [5]
first GPU will get β Nothing
which actually creates an issue since in case of `DistributedSampler`, the samples are repeated internally to ensure non of the GPUs at any iteration is missing any data for gradient sync.
So my questions are:
1. Here since splitting is happening before hand, how to make sure each GPU getβs a batch at each iteration to avoid gradient sync issues?
2. Do we need to use `DistributedSampler`? If yes, how?
3. in the docstrings of `split_dataset_by_node`, this is mentioned: *"If the dataset has a number of shards that is a factor of `world_size` (i.e. if `dataset.n_shards % world_size == 0`), then the shards are evenly assigned across the nodes, which is the most optimized. Otherwise, each node keeps 1 example out of `world_size`, skipping the other examples."* Can you explain the last part here?
4. If `dataset.n_shards % world_size != 0`, is it possible to shard the streaming dataset on the fly to avoid the case where data is missing?
### Motivation
Somehow streaming datasets should work with DDP since for big LLMs a lot of data is required and DDP/multi-node is mostly used to train such models and streaming can actually help solve the data part of it.
### Your contribution
Yes, I can help in submitting the PR once we get mutual understanding on how it should behave. | 6,623 |
https://github.com/huggingface/datasets/issues/6622 | multi-GPU map does not work | [
"This should now be fixed by https://github.com/huggingface/datasets/pull/6550 and updated with https://github.com/huggingface/datasets/pull/6646\r\n\r\nFeel free to re-open if you're still having issues :)"
] | ### Describe the bug
Here is the code for single-GPU processing: https://pastebin.com/bfmEeK2y
Here is the code for multi-GPU processing: https://pastebin.com/gQ7i5AQy
Here is the video showing that the multi-GPU mapping does not work as expected (there are so many things wrong here, it's better to watch the 3-minute video than explain here):
https://youtu.be/RNbdPkSppc4
### Steps to reproduce the bug
-
### Expected behavior
-
### Environment info
x2 RTX A4000 | 6,622 |
https://github.com/huggingface/datasets/issues/6621 | deleted | [] | ... | 6,621 |
https://github.com/huggingface/datasets/issues/6620 | wiki_dpr.py error (ID mismatch between lines {id} and vector {vec_id} | [
"Thanks for reporting, @kiehls90.\r\n\r\nAs this seems an issue with the specific \"wiki_dpr\" dataset, I am transferring the issue to the corresponding dataset page: https://huggingface.co/datasets/wiki_dpr/discussions/13"
] | ### Describe the bug
I'm trying to run a rag example, and the dataset is wiki_dpr.
wiki_dpr download and extracting have been completed successfully.
However, at the generating train split stage, an error from wiki_dpr.py keeps popping up.
Especially in "_generate_examples" :
1. The following error occurs in the line **id, text, title = line.strip().split("\t")**
ValueError: not enough values ββto unpack (expected 3, got 2)
-> This part handles exceptions so that even if an error occurs, it passes.
2. **ID mismatch between lines {id} and vector {vec_id}**
This error seems to occur at the line " assert int(id) == int(vec_id),".
After I handled the exception in the split error, generating train split progressed to 80%, but an id mismatch error occurred at about the 16200000th vector id.
Debugging is even more difficult because it takes a long time to download and split wiki_dpr. I need help. thank you in advance!!
### Steps to reproduce the bug
Occurs in the generating train split step when running the rag example in the transformers repository.
Specifically, it is an error in wiki_dpr.py.
### Expected behavior
.
### Environment info
python 3.8 | 6,620 |
https://github.com/huggingface/datasets/issues/6618 | While importing load_dataset from datasets | [
"Hi! Can you please share the error's stack trace so we can see where it comes from?",
"We cannot reproduce the issue and we do not have enough information: environment info (need to run `datasets-cli env`), stack trace,...\r\n\r\nI am closing the issue. Feel free to reopen it (with additional information) if the problem persists.",
"Yeah π\r\n\r\nOn Tue, 6 Feb 2024 at 2:56 PM, Albert Villanova del Moral <\r\n***@***.***> wrote:\r\n\r\n> We cannot reproduce the issue and we do not have enough information:\r\n> environment info (need to run datasets-cli env), stack trace,...\r\n>\r\n> I am closing the issue. Feel free to reopen it (with additional\r\n> information) if the problem persists.\r\n>\r\n> β\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/huggingface/datasets/issues/6618#issuecomment-1929102334>,\r\n> or unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/ASS4PJ3XOIIWISPY3VX3QRTYSHZK5AVCNFSM6AAAAABCL3BT4SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMRZGEYDEMZTGQ>\r\n> .\r\n> You are receiving this because you authored the thread.Message ID:\r\n> ***@***.***>\r\n>\r\n",
"Please downgrade the version of urllib3 if you have the same issue:\r\n\r\n!pip install urllib3==1.25.11",
"> Please downgrade the version of urllib3 if you have the same issue:\r\n> \r\n> !pip install urllib3==1.25.11\r\n\r\nThis worked for me. Thanks.\r\n\r\nI use python 3.11 and datasets==2.20.0. Downgrading urllib3 to 1.25.11 worked in my case."
] | ### Describe the bug
cannot import name 'DEFAULT_CIPHERS' from 'urllib3.util.ssl_' this is the error i received
### Steps to reproduce the bug
from datasets import load_dataset
### Expected behavior
No errors
### Environment info
python 3.11.5 | 6,618 |
https://github.com/huggingface/datasets/issues/6615 | ... | [
"Sorry I posted in the wrong repo, please delete.. thanks!"
] | ... | 6,615 |
https://github.com/huggingface/datasets/issues/6614 | `datasets/downloads` cleanup tool | [] | ### Feature request
Splitting off https://github.com/huggingface/huggingface_hub/issues/1997 - currently `huggingface-cli delete-cache` doesn't take care of cleaning `datasets` temp files
e.g. I discovered having millions of files under `datasets/downloads` cache, I had to do:
```
sudo find /data/huggingface/datasets/downloads -type f -mtime +3 -exec rm {} \+
sudo find /data/huggingface/datasets/downloads -type d -empty -delete
```
could the cleanup be integrated into `huggingface-cli` or a different tool provided to keep the folders tidy and not consume inodes and space
e.g. there were tens of thousands of `.lock` files - I don't know why they never get removed - lock files should be temporary for the duration of the operation requiring the lock and not remain after the operation finished, IMHO.
Also I think one should be able to nuke `datasets/downloads` w/o hurting the cache, but I think there are some datasets that rely on files extracted under this dir - or at least they did in the past - which is very difficult to manage since one has no idea what is safe to delete and what not.
Thank you
@Wauplin (requested to be tagged) | 6,614 |
https://github.com/huggingface/datasets/issues/6612 | cnn_dailymail repeats itself | [
"Hi ! We recently updated `cnn_dailymail` and now `datasets>=2.14` is needed to load it.\r\n\r\nYou can update `datasets` with\r\n\r\n```\r\npip install -U datasets\r\n```"
] | ### Describe the bug
When I try to load `cnn_dailymail` dataset, it takes longer than usual and when I checked the dataset it's 3x bigger than it's supposed to be.
Check https://huggingface.co/datasets/cnn_dailymail: it says 287k rows for train. But when I check length of train split it says 861339.
Also I checked data:
```
>>> ds['train']['highlights'][0]
"Harry Potter star Daniel Radcliffe gets Β£20M fortune as he turns 18 Monday . Young actor says he has no plans to fritter his cash away . Radcliffe's earnings from first five Potter films have been held in trust fund ."````
>>> ds['train']['highlights'][0]
"Harry Potter star Daniel Radcliffe gets Β£20M fortune as he turns 18 Monday . Young actor says he has no plans to fritter his cash away . Radcliffe's earnings from first five Potter films have been held in trust fund ."````
>>> ds['train']['highlights'][287113]
"Harry Potter star Daniel Radcliffe gets Β£20M fortune as he turns 18 Monday .\nYoung actor says he has no plans to fritter his cash away .\nRadcliffe's earnings from first five Potter films have been held in trust fund ."````
>>> ds['train']['highlights'][574226]
"Harry Potter star Daniel Radcliffe gets Β£20M fortune as he turns 18 Monday .\nYoung actor says he has no plans to fritter his cash away .\nRadcliffe's earnings from first five Potter films have been held in trust fund ."
```
The datasets seems to be updated 6 days ago to convert it to Parquet. Probably, there is some issue with backward compatability.
### Steps to reproduce the bug
1.
```
from datasets import load_dataset
ds = load_dataset('cnn_dailymail', '3.0.0')
len(ds['train'])
```
### Expected behavior
It should not repeat itself.
### Environment info
datasets==2.13.2
Python==3.7.13 | 6,612 |
https://github.com/huggingface/datasets/issues/6611 | `load_from_disk` with large dataset from S3 runs into `botocore.exceptions.ClientError` | [] | ### Describe the bug
When loading a large dataset (>1000GB) from S3 I run into the following error:
```
Traceback (most recent call last):
File "/home/alp/.local/lib/python3.10/site-packages/s3fs/core.py", line 113, in _error_wrapper
return await func(*args, **kwargs)
File "/home/alp/.local/lib/python3.10/site-packages/aiobotocore/client.py", line 383, in _make_api_call
raise error_class(parsed_response, operation_name)
botocore.exceptions.ClientError: An error occurred (RequestTimeTooSkewed) when calling the GetObject operation: The difference between the request time and the current time is too large.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/home/alp/phoneme-classification.monorepo/aws_sagemaker/data_processing/inspect_final_dataset.py", line 13, in <module>
dataset = load_from_disk("s3://speech-recognition-processed-data/whisper/de/train_data/", storage_options=storage_options)
File "/home/alp/.local/lib/python3.10/site-packages/datasets/load.py", line 1902, in load_from_disk
return Dataset.load_from_disk(dataset_path, keep_in_memory=keep_in_memory, storage_options=storage_options)
File "/home/alp/.local/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 1686, in load_from_disk
fs.download(src_dataset_path, [dest_dataset_path.as](http://dest_dataset_path.as/)_posix(), recursive=True)
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/spec.py", line 1480, in download
return self.get(rpath, lpath, recursive=recursive, **kwargs)
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/asyn.py", line 121, in wrapper
return sync(self.loop, func, *args, **kwargs)
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/asyn.py", line 106, in sync
raise return_result
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/asyn.py", line 61, in _runner
result[0] = await coro
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/asyn.py", line 604, in _get
return await _run_coros_in_chunks(
File "/home/alp/.local/lib/python3.10/site-packages/fsspec/asyn.py", line 257, in _run_coros_in_chunks
await asyncio.gather(*chunk, return_exceptions=return_exceptions),
File "/usr/lib/python3.10/asyncio/tasks.py", line 408, in wait_for
return await fut
File "/home/alp/.local/lib/python3.10/site-packages/s3fs/core.py", line 1193, in _get_file
body, content_length = await _open_file(range=0)
File "/home/alp/.local/lib/python3.10/site-packages/s3fs/core.py", line 1184, in _open_file
resp = await self._call_s3(
File "/home/alp/.local/lib/python3.10/site-packages/s3fs/core.py", line 348, in _call_s3
return await _error_wrapper(
File "/home/alp/.local/lib/python3.10/site-packages/s3fs/core.py", line 140, in _error_wrapper
raise err
PermissionError: The difference between the request time and the current time is too large.
```
The usual problem for this error is that the time on my local machine is out of sync with the current time. However, this is not the case here. I checked the time and even reset it with no success. See resources here:
- https://stackoverflow.com/questions/4770635/s3-error-the-difference-between-the-request-time-and-the-current-time-is-too-la
- https://stackoverflow.com/questions/25964491/aws-s3-upload-fails-requesttimetooskewed
The error does not appear when loading a smaller dataset (e.g. our test set) from the same s3 path.
### Steps to reproduce the bug
1. Create large dataset
2. Try loading it from s3 using:
```
dataset = load_from_disk("s3://...", storage_options=storage_options)
```
### Expected behavior
Load dataset without running into this error.
### Environment info
- `datasets` version: 2.13.1
- Platform: Linux-5.15.0-91-generic-x86_64-with-glibc2.35
- Python version: 3.10.12
- Huggingface_hub version: 0.19.3
- PyArrow version: 12.0.1
- Pandas version: 2.0.3 | 6,611 |
https://github.com/huggingface/datasets/issues/6610 | cast_column to Sequence(subfeatures_dict) has err | [
"Hi! You are passing the wrong feature type to `cast_column`. This is the fixed call:\r\n```python\r\nais_dataset = ais_dataset.cast_column(\"my_labeled_bbox\", {\"bbox\": Sequence(Value(dtype=\"int64\")), \"label\": ClassLabel(names=[\"cat\", \"dog\"])})\r\n```",
"> Hi! You are passing the wrong feature type to `cast_column`. This is the fixed call:\r\n> \r\n> ```python\r\n> ais_dataset = ais_dataset.cast_column(\"my_labeled_bbox\", {\"bbox\": Sequence(Value(dtype=\"int64\")), \"label\": ClassLabel(names=[\"cat\", \"dog\"])})\r\n> ```\r\n\r\nthanks"
] | ### Describe the bug
I am working with the following demo code:
```
from datasets import load_dataset
from datasets.features import Sequence, Value, ClassLabel, Features
ais_dataset = load_dataset("/data/ryan.gao/ais_dataset_cache/raw/1978/")
ais_dataset = ais_dataset["train"]
def add_class(example):
example["my_labeled_bbox"] = {"bbox": [100,100,200,200], "label": "cat"}
return example
ais_dataset = ais_dataset.map(add_class, batched=False, num_proc=32)
ais_dataset = ais_dataset.cast_column("my_labeled_bbox", Sequence(
{
"bbox": Sequence(Value(dtype="int64")),
"label": ClassLabel(names=["cat", "dog"])
}))
print(ais_dataset[0])
```
However, executing this code results in an error:
```
File "/home/protoss.gao/.local/lib/python3.9/site-packages/datasets/table.py", line 2111, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{array.type}\nto\n{feature}")
TypeError: Couldn't cast array of type
int64
to
Sequence(feature=Value(dtype='int64', id=None), length=-1, id=None)
```
Upon examining the source code in datasets/table.py at line 2035:
```
if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
feature = {
name: Sequence(subfeature, length=feature.length) for name, subfeature in feature.feature.items()
}
```
I noticed that if subfeature is of type Sequence, the code results in Sequence(Sequence(...), ...) and Sequence(ClassLabel(...), ...), which appears to be the source of the error.
### Steps to reproduce the bug
run my demo code
### Expected behavior
no exception
### Environment info
python 3.9
datasets: 2.16.1 | 6,610 |
https://github.com/huggingface/datasets/issues/6609 | Wrong path for cache directory in offline mode | [
"+1",
"same error in 2.16.1",
"@kongjiellx any luck with the issue?",
"I opened https://github.com/huggingface/datasets/pull/6632 to fix this issue. Once it's merged we'll do a new release of `datasets`",
"Thanks @lhoestq !"
] | ### Describe the bug
Dear huggingfacers,
I'm trying to use a subset of the-stack dataset. When I run the command the first time
```
dataset = load_dataset(
path='bigcode/the-stack',
data_dir='data/fortran',
split='train' )
```
It downloads the files and caches them normally.
Nevertheless, since my compute nodes are not online (`HF_DATASETS_OFFLINE=1`) . Whenever I try to run the command again, the library is passing the wrong cache path:
`Cache directory for the-stack doesn't exist at /Users/user/.cache/huggingface/datasets/bigcode___the-stack/default-data_dir=data%2Ffortran-data_dir=data%2Ffortran`
when the right path is:
`'/Users/user/.cache/huggingface/datasets/bigcode___the-stack/default-data_dir=data\%2Ffortran`
Not sure why those redundancies are included in the path. If I try adding the correct path through the the cache_dir argument it throws an error:
ConnectionError: Couldn't reach the Hugging Face Hub for dataset 'bigcode/the-stack': Offline mode is enabled.
Your help with this issue is greatly appreciated. Thanks a lot for the great work.
### Steps to reproduce the bug
1:
`dataset = load_dataset(
path='bigcode/the-stack',
data_dir='data/fortran',
split='train' )`
2:
`HF_DATASETS_OFFLINE=1`
3:
`dataset = load_dataset(
path='bigcode/the-stack',
data_dir='data/fortran',
split='train' )`
### Expected behavior
being able to use the cached data
### Environment info
several different systems | 6,609 |
https://github.com/huggingface/datasets/issues/6605 | ELI5 no longer available, but referenced in example code | [
"Addressed in https://github.com/huggingface/transformers/pull/28715."
] | Here, an example code is given:
https://huggingface.co/docs/transformers/tasks/language_modeling
This code + article references the ELI5 dataset.
ELI5 is no longer available, as the ELI5 dataset page states: https://huggingface.co/datasets/eli5
"Defunct: Dataset "eli5" is defunct and no longer accessible due to unavailability of the source data.
Reddit recently [changed the terms of access](https://www.reddit.com/r/reddit/comments/12qwagm/an_update_regarding_reddits_api/) to its API, making the source data for this dataset unavailable.
"
Please change the example code to use a different dataset. | 6,605 |
https://github.com/huggingface/datasets/issues/6604 | Transform fingerprint collisions due to setting fixed random seed | [
"I've opened a PR with a fix.",
"I don't think the PR fixes the root cause, since it still relies on the `random` library which will often have its seed fixed. I think the builtin `uuid.uuid4()` is a better choice: https://docs.python.org/3/library/uuid.html"
] | ### Describe the bug
The transform fingerprinting logic relies on the `random` library for random bits when the function is not hashable (e.g. bound methods as used in `trl`: https://github.com/huggingface/trl/blob/main/trl/trainer/dpo_trainer.py#L356). This causes collisions when the training code sets a fixed random seed, which is common practice: https://github.com/huggingface/alignment-handbook/blob/main/recipes/zephyr-7b-beta/sft/config_full.yaml#L45.
This results in fingerprint collisions which leads to silently loading incorrect cache files corresponding to completely different datasets.
### Steps to reproduce the bug
n/a
### Expected behavior
Use `uuid` v4 instead of `random.getrandbits()`
### Environment info
`datasets` main branch | 6,604 |
https://github.com/huggingface/datasets/issues/6603 | datasets map `cache_file_name` does not work | [
"Unfortunately, I'm unable to reproduce this error. Can you share the reproducer?",
"```\r\nds = datasets.Dataset.from_dict(dict(a=[i for i in range(100)]))\r\nds.map(lambda item: dict(b=item['a'] * 2), cache_file_name=\"/tmp/whatever-fn\") # this worked\r\nds.map(lambda item: dict(b=item['a'] * 2), cache_file_name=\"/tmp/whatever-folder/filename\") # this failed\r\nds.map(lambda item: dict(b=item['a'] * 2), cache_file_name=\"/tmp/whatever-folder/\") # this failed\r\n\r\n\r\nFileNotFoundError: [Errno 2] No such file or directory: '/tmp/whatever-folder/tmp1_izxvoo'\r\n```\r\n\r\nIt will fail if the filename parents do not exists. If we have `os.makedirs(\"/tmp/whatever-folder\")`, then it worked.\r\n\r\nMaybe add the `mkdir -p` into the map function?"
] | ### Describe the bug
In the documentation `datasets.Dataset.map` arg `cache_file_name` is said to be a string, but it doesn't work.
### Steps to reproduce the bug
1. pick a dataset
2. write a map function
3. do `ds.map(..., cache_file_name='some_filename')`
4. it crashes
### Expected behavior
It will tell you the filename you specified does not exist or it will generate a new file and tell you the filename does not exist.
### Environment info
- `datasets` version: 2.16.0
- Platform: Linux-5.10.201-168.748.amzn2int.x86_64-x86_64-with-glibc2.26
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.2
- Pandas version: 2.1.4
- `fsspec` version: 2023.12.2 | 6,603 |
https://github.com/huggingface/datasets/issues/6602 | Index error when data is large | [] | ### Describe the bug
At `save_to_disk` step, the `max_shard_size` by default is `500MB`. However, one row of the dataset might be larger than `500MB` then the saving will throw an index error. Without looking at the source code, the bug is due to wrong calculation of number of shards which i think is
`total_size / min(max_shard_size, row_size)` which should be `total_size / max(max_shard_size, row_size)`
The fix is setting a larger `max_shard_size`
### Steps to reproduce the bug
1. create a dataset with large dense tensors per row
2. set a small `max_shard_size` say 1MB
3. `save_to_disk`
### Expected behavior
```
raise IndexError(f"Index {index} out of range for dataset of size {size}.")
IndexError: Index 10 out of range for dataset of size 10.
```
### Environment info
- `datasets` version: 2.16.0
- Platform: Linux-5.10.201-168.748.amzn2int.x86_64-x86_64-with-glibc2.26
- Python version: 3.10.13
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.2
- Pandas version: 2.1.4
- `fsspec` version: 2023.12.2 | 6,602 |
https://github.com/huggingface/datasets/issues/6600 | Loading CSV exported dataset has unexpected format | [
"Hi! Parquet is the only format that supports complex/nested features such as `Translation`. So, this should work:\r\n```python\r\ntest_dataset = load_dataset(\"opus100\", name=\"en-fr\", split=\"test\")\r\n\r\n# Save with .to_parquet()\r\ntest_parquet_path = \"try_testset_save.parquet\"\r\ntest_dataset.to_parquet(test_parquet_path)\r\n\r\n# Load dataset from the Parquet\r\nloaded_dataset = load_dataset(\"parquet\", data_files=test_parquet_path)\r\nprint(test_dataset_fromfile[0][\"translation\"])\r\nprint(test_dataset_fromfile[0][\"translation\"][\"en\"])\r\n```",
"Indeed this works great, thank you !"
] | ### Describe the bug
I wanted to be able to save a HF dataset for translations and load it again in another script, but I'm a bit confused with the documentation and the result I've got so I'm opening this issue to ask if this behavior is as expected.
### Steps to reproduce the bug
The documentation I've mainly consulted is https://huggingface.co/docs/datasets/v2.16.1/en/package_reference/loading_methods#datasets.load_dataset and https://huggingface.co/docs/datasets/package_reference/main_classes#datasets.Dataset (where I've found `.to_csv()`)
```python
# Load a dataset of translations
test_dataset = load_dataset("opus100", name="en-fr", split="test")
# Save with .to_csv()
test_csv_path = "try_testset_save.csv"
test_dataset.to_csv(test_csv_path)
# Load dataset from the CSV
loaded_dataset = load_dataset("csv", data_files=test_csv_path)
print(test_dataset_fromfile[0]["translation"])
print(test_dataset_fromfile[0]["translation"]["en"])
```
```
Creating CSV from Arrow format: 100%
2/2 [00:00<00:00, 47.99ba/s]
Downloading data files: 100%
1/1 [00:00<00:00, 65.33it/s]
Extracting data files: 100%
1/1 [00:00<00:00, 42.10it/s]
Generating train split:
2000/0 [00:00<00:00, 47486.09 examples/s]
{'en': "She wasn't going to vaccinate her kid against polio, no way.", 'fr': 'Elle ne vaccinerait pas son enfant contre la polio. Pas question.'}
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[29], line 11
9 loaded_dataset = load_dataset("csv", data_files=test_csv_path)
10 print(test_dataset_fromfile[0]["translation"])
---> 11 print(test_dataset_fromfile[0]["translation"]["en"])
TypeError: string indices must be integers, not 'str'
```
### Expected behavior
Each translation was saved as a stringified dict like `"{'en': ""She wasn't going to vaccinate her kid against polio, no way."", 'fr': 'Elle ne vaccinerait pas son enfant contre la polio. Pas question.'}"` where I would have expected 2 columns (1st with English segments, and 2nd with French segments), and I was expecting `load_dataset` to infer the type of feature automatically as I haven't seen anything about it in the documentation.
Do you have an example of how to effectively save and load datasets of translations ?
### Environment info
- `datasets` version: 2.15.0
- Platform: Linux-3.10.0-1160.36.2.el7.x86_64-x86_64-with-glibc2.17
- Python version: 3.11.5
- `huggingface_hub` version: 0.16.4
- PyArrow version: 14.0.2
- Pandas version: 2.1.4
- `fsspec` version: 2023.10.0 | 6,600 |
https://github.com/huggingface/datasets/issues/6599 | Easy way to segment into 30s snippets given an m4a file and a vtt file | [
"Hi! Non-generic data processing is out of this library's scope, so it's downstream libraries/users' responsibility to implement such logic.",
"That's fair. Thanks"
] | ### Feature request
Uploading datasets is straightforward thanks to the ability to push Audio to hub. However, it would be nice if the data (text and audio) could be segmented when being pushed (if not possible already).
### Motivation
It's easy to create a vtt file from an audio file. If there could be auto-segmenting, this would make the creation of datasets much faster.
### Your contribution
I have made a custom script to do this but it's not all that clean - uses librosa and pydub. | 6,599 |
https://github.com/huggingface/datasets/issues/6598 | Unexpected keyword argument 'hf' when downloading CSV dataset from S3 | [
"I am facing similar issue while reading a csv file from s3. Wondering if somebody has found a workaround. ",
"same thing happened to other formats like parquet",
"I am facing similar issue while reading a parquet file from s3.\r\ni try with every version between 2.14 to 2.16.1 but it dosen't work ",
"Re-define the DownloadConfig might work:\r\n\r\n```\r\nclass ReviseDownloadConfig(DownloadConfig):\r\n def __post_init__(self, use_auth_token):\r\n if use_auth_token != \"deprecated\":\r\n warnings.warn(\r\n \"'use_auth_token' was deprecated in favor of 'token' in version 2.14.0 and will be removed in 3.0.0.\\n\"\r\n f\"You can remove this warning by passing 'token={use_auth_token}' instead.\",\r\n FutureWarning,\r\n )\r\n self.token = use_auth_token\r\n\r\n def copy(self):\r\n return self.__class__(**{k: copy.deepcopy(v) for k, v in self.__dict__.items()})\r\n\r\ndownloadconfig = ReviseDownloadConfig()\r\n```\r\n",
"> Re-define the DownloadConfig might work:\r\n> \r\n> ```\r\n> class ReviseDownloadConfig(DownloadConfig):\r\n> def __post_init__(self, use_auth_token):\r\n> if use_auth_token != \"deprecated\":\r\n> warnings.warn(\r\n> \"'use_auth_token' was deprecated in favor of 'token' in version 2.14.0 and will be removed in 3.0.0.\\n\"\r\n> f\"You can remove this warning by passing 'token={use_auth_token}' instead.\",\r\n> FutureWarning,\r\n> )\r\n> self.token = use_auth_token\r\n> ```\r\nThis seemed to work for me.\r\n",
"use pandas and then convert to `Dataset`",
"I am currently facing the same issue while using a custom loading script with files located in a remote S3 instance. I was using the `download_custom` functionality but now it is deprecated mentioning that I should use the native S3 loading, which is not working. \r\n\r\nAs stated before, the library forces the existence of a `hf` key in the `storage_options` variable, which is **not** accepted by `s3fs` : \r\n\r\n```python\r\n.../site-packages/s3fs/core.py\", line 516, in set_session\r\n self.session = aiobotocore.session.AioSession(**self.kwargs)\r\nTypeError: __init__() got an unexpected keyword argument 'hf'.\r\n````\r\n\r\nMeanwhile, if my `storage_options` var stays like:\r\n```python\r\n{'key': '...',\r\n 'secret': '...',\r\n 'client_kwargs': {'endpoint_url': '...'}}\r\n```\r\nit works alright. "
] | ### Describe the bug
I receive this error message when using `load_dataset` with "csv" path and `dataset_files=s3://...`:
```
TypeError: Session.__init__() got an unexpected keyword argument 'hf'
```
I found a similar issue here: https://stackoverflow.com/questions/77596258/aws-issue-load-dataset-from-s3-fails-with-unexpected-keyword-argument-error-in
Full stacktrace:
```
.../site-packages/datasets/load.py:2549: in load_dataset
builder_instance.download_and_prepare(
.../site-packages/datasets/builder.py:1005: in download_and_prepare
self._download_and_prepare(
.../site-packages/datasets/builder.py:1078: in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
.../site-packages/datasets/packaged_modules/csv/csv.py:147: in _split_generators
data_files = dl_manager.download_and_extract(self.config.data_files)
.../site-packages/datasets/download/download_manager.py:562: in download_and_extract
return self.extract(self.download(url_or_urls))
.../site-packages/datasets/download/download_manager.py:426: in download
downloaded_path_or_paths = map_nested(
.../site-packages/datasets/utils/py_utils.py:466: in map_nested
mapped = [
.../site-packages/datasets/utils/py_utils.py:467: in <listcomp>
_single_map_nested((function, obj, types, None, True, None))
.../site-packages/datasets/utils/py_utils.py:387: in _single_map_nested
mapped = [_single_map_nested((function, v, types, None, True, None)) for v in pbar]
.../site-packages/datasets/utils/py_utils.py:387: in <listcomp>
mapped = [_single_map_nested((function, v, types, None, True, None)) for v in pbar]
.../site-packages/datasets/utils/py_utils.py:370: in _single_map_nested
return function(data_struct)
.../site-packages/datasets/download/download_manager.py:451: in _download
out = cached_path(url_or_filename, download_config=download_config)
.../site-packages/datasets/utils/file_utils.py:188: in cached_path
output_path = get_from_cache(
...1/site-packages/datasets/utils/file_utils.py:511: in get_from_cache
response = fsspec_head(url, storage_options=storage_options)
.../site-packages/datasets/utils/file_utils.py:316: in fsspec_head
fs, _, paths = fsspec.get_fs_token_paths(url, storage_options=storage_options)
.../site-packages/fsspec/core.py:622: in get_fs_token_paths
fs = filesystem(protocol, **inkwargs)
.../site-packages/fsspec/registry.py:290: in filesystem
return cls(**storage_options)
.../site-packages/fsspec/spec.py:79: in __call__
obj = super().__call__(*args, **kwargs)
.../site-packages/s3fs/core.py:187: in __init__
self.s3 = self.connect()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
self = <s3fs.core.S3FileSystem object at 0x1500a1310>, refresh = True
def connect(self, refresh=True):
"""
Establish S3 connection object.
Parameters
----------
refresh : bool
Whether to create new session/client, even if a previous one with
the same parameters already exists. If False (default), an
existing one will be used if possible
"""
if refresh is False:
# back compat: we store whole FS instance now
return self.s3
anon, key, secret, kwargs, ckwargs, token, ssl = (
self.anon, self.key, self.secret, self.kwargs,
self.client_kwargs, self.token, self.use_ssl)
if not self.passed_in_session:
> self.session = botocore.session.Session(**self.kwargs)
E TypeError: Session.__init__() got an unexpected keyword argument 'hf'
```
### Steps to reproduce the bug
1. Assuming a valid CSV file located at `s3://bucket/data.csv`
2. Run the below code:
```
storage_options = {
"key": "...",
"secret": "...",
"client_kwargs": {
"endpoint_url": "...",
}
}
load_dataset("csv", data_files="s3://bucket/data.csv", storage_options=storage_options)
```
Encountered in version `2.16.1` but also reproduced in `2.16.0` and `2.15.0`.
Note: I encountered this in a unit test using a `moto` mock for S3, however since the error occurs before the session is instantiated, it should not be the issue.
### Expected behavior
No exception is raised, the boto3 session is created successfully, and the CSV file is downloaded successfully and returned as a dataset.
===
After some research I found that `DownloadConfig` has a `__post_init__` method that always forces this value to be set in its `storage_options`, even though in case of an S3 location the storage options get passed on to the S3 Session which does not expect this parameter. I assume this parameter is needed when reading from the huggingface hub and should not be set in this context.
Unfortunately there is nothing the user can do to work around it. Even if you manually do something like:
```
download_config = DownloadConfig()
del download_config.storage_options["hf"]
load_dataset("csv", data_files="s3://bucket/data.csv", download_config=download_config)
```
the library will still reinsert this parameter when `download_config = self.download_config.copy()` in line 418 of `download_manager.py` (`DownloadManager.download`).
Therefore `load_dataset` currently cannot be used to read a dataset in CSV format from an S3 location.
### Environment info
- `datasets` version: 2.16.1
- Platform: macOS-14.2.1-arm64-arm-64bit
- Python version: 3.11.7
- `huggingface_hub` version: 0.20.2
- PyArrow version: 14.0.2
- Pandas version: 2.1.4
- `fsspec` version: 2023.10.0
| 6,598 |
https://github.com/huggingface/datasets/issues/6597 | Dataset.push_to_hub of a canonical dataset creates an additional dataset under the user namespace | [
"It is caused by these code lines: https://github.com/huggingface/datasets/blob/9d6d16117a30ba345b0236407975f701c5b288d4/src/datasets/dataset_dict.py#L1688-L1694",
"Also note the information in the docstring: https://github.com/huggingface/datasets/blob/9d6d16117a30ba345b0236407975f701c5b288d4/src/datasets/dataset_dict.py#L1582-L1585\r\n\r\n> Also accepts `<dataset_name>`, which will default to the namespace of the logged-in user.\r\n\r\nThis behavior was \"reverted\" by the PR: \r\n- #6519\r\n\r\nWe have therefore contradictory requirements. We should decide:\r\n- whether to support passing dataset_namespace without user/org that defaults to the logged-in user (and not support canonical datasets)\r\n- or vice-versa, to support canonical datasets and not support passing only dataset_name\r\n\r\nAs canonical datasets are \"deprecated\" (and will eventually disappear), I would choose the first option. However, if so, the Space to convert datasets to Parquet will not work for canonical datasets: https://huggingface.co/spaces/albertvillanova/convert-dataset-to-parquet",
"IIUC, this could also be \"fixed\" by `create_repo(\"dataset_name\")` not defaulting to `create_repo(\"user/dataset_name\")` (when the user's token is available), which would be consistent with the rest of the `HfApi` ops used in the `push_to_hub` implementation. This is a (small) breaking change for `huggingface_hub`, but justified to make the API more consistent.",
"I tag @Wauplin to have his opinion as well.",
"Hmm, creating repo with implicit namespace (e.g. `create_repo(\"dataset_name\")`) is a convenient feature used in a lot of integrations. It is not consistent with other HfApi methods specifically because it is the method to create repos. Once the repo is created, the return value provides the explicit repo_id (`namespace/repo_name`) that has to be passed to every `HfApi` method. Otherwise, libraries/scripts would often need to do a `whoami` call to get the namespace before creating a repo.\r\n\r\n Another solution for https://github.com/huggingface/datasets/issues/6597#issuecomment-1893746690 could be that implicit namespace is allowed (same as today) except if the `repo_id` is in a hard-coded list of canonical datasets. This list can be maintained automatically and should be slowly decreasing. **Caveat:** as a normal user I wouldn't be able to implicitly push to `imagenet-1k` if I wanted to push to `Wauplin/imagenet-1k`. Shouldn't be too problematic, no? Worse case, would need to add a `whoami` call and allow implicit-canonical-name for non-HF users for instance (a bit too over-engineered IMO but doable). ",
"As canonical datasets are going to disappear in the following couple of months, I would not make any effort on their support.\r\n\r\nI propose reverting #6519, so that the behavior of `push_to_hub` is aligned with the one described in its dosctring: \"Also accepts `<dataset_name>`, which will default to the namespace of the logged-in user.\"\r\n\r\nI'm opening a PR."
] | While using `Dataset.push_to_hub` of a canonical dataset, an additional dataset was created under my user namespace.
## Steps to reproduce the bug
The command:
```python
commit_info = ds.push_to_hub(
"caner",
config_name="default",
commit_message="Convert dataset to Parquet",
commit_description="Convert dataset to Parquet.",
create_pr=True,
token=token,
)
```
creates the additional dataset `albertvillanova/caner`. | 6,597 |
https://github.com/huggingface/datasets/issues/6595 | Loading big dataset raises pyarrow.lib.ArrowNotImplementedError 2 | [
"Hi ! I think the issue comes from the \"float16\" features that are not supported yet in Parquet\r\n\r\nFeel free to open an issue in `pyarrow` about this. In the meantime, I'd encourage you to use \"float32\" for your \"pooled_prompt_embeds\" and \"prompt_embeds\" features.\r\n\r\nYou can cast them to \"float32\" using\r\n\r\n```python\r\nfrom datasets import Value\r\n\r\nds = ds.cast_column(\"pooled_prompt_embeds\", Value(\"float32\"))\r\nds = ds.cast_column(\"prompt_embeds\", Value(\"float32\"))\r\n```",
"@lhoestq hm. Thank you very much.\r\n\r\nDo you think it won't have any impact on the training? That it won't break it or the quality won't degrade because of this?\r\n\r\nI need to use it for [SDXL training](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_sdxl.py)",
"Increasing the precision should not degrade training (it only increases the precision), but make sure that it doesn't break your pytorch code (e.g. if it expects a float16 instead of a float32 somewhere)",
"@lhoestq just fyi pyarrow 15.0.0 (just released) supports float16 as the underlying parquetcpp does as well now :)",
"Oh that's amazing ! (and great timing ^^)\r\n\r\n@kopyl can you try to update `pyarrow` and try again ?\r\n\r\nBtw @assignUser there seems to be some casting implementations missing with float16 in 15.0.0, e.g.\r\n\r\n```\r\nArrowNotImplementedError: Unsupported cast from int64 to halffloat using function cast_half_float\r\n```\r\n\r\n```\r\nArrowNotImplementedError: Unsupported cast from double to halffloat using function cast_half_float\r\n```",
"Ah you are right casting is not implemented yet, it's even mentioned in the docs. This pr references the relevant issues if you'd like to track them\nhttps://github.com/apache/arrow/pull/38494",
"Cool thank you :)",
"@lhoestq i just recently found out that it's supported in 15.0.0, but wanted to try it first before telling you...\r\n\r\nTrying this right now and it seemingly works (although i need to wait till the end to make sure there is nothing wrong). Will update you when it's finished.\r\n\r\n<img width=\"918\" alt=\"image\" src=\"https://github.com/huggingface/datasets/assets/17604849/4821e215-e782-4736-8c76-d06187078175\">\r\n\r\nA couple of questions though:\r\n\r\n1. What does that missing casting implementation mean for my specific case and what does it mean in general?\r\n2. Do you know how to `push_to_hub` with multiple processes?",
"@lhoestq also it's strange that there was no error for a dataset with the same features, same data type, but smaller (much smaller).\r\n\r\nAltho i'm not sure about this, but chances are the dataset was loaded directly, not `load_from_disk`.... Maybe because of this.",
"> What does that missing casting implementation mean for my specific case and what does it mean in general?\r\n\r\nNothing for you, just that casting to float16 using `.cast_column(\"my_column_name\", Value(\"float16\"))` raises an error\r\n\r\n> Do you know how to push_to_hub with multiple processes?\r\n\r\nIt's not possible (yet ?). Mostly because we haven't implemented yet how to do parallel uploads to the Hub from `datasets`.\r\nThough if you want faster uploads you can already enable `hf_transfer` \r\n\r\n```\r\npip install hf_transfer\r\n```\r\n\r\nand setting `HF_HUB_ENABLE_HF_TRANSFER=1` as an environment variable\r\n\r\nsee https://huggingface.co/docs/huggingface_hub/guides/upload#tips-and-tricks-for-large-uploads",
"@lhoestq thank you very much.\r\n\r\nThat would be amazing, I need to create a feature request for this :)\r\n\r\nBy the way, in short, how does hf_transfer improves the upload speed under the hood?",
"@lhoestq i was just able to successfully upload without the dataset with the new pyarrow update and without increasing the precision :)",
"Awesome !\r\n\r\nRegarding hf_transfer: it's been optimized in rust ;)",
"@lhoestq wow, cool :)"
] | ### Describe the bug
I'm aware of the issue #5695 .
I'm using a modified SDXL trainer: https://github.com/kopyl/diffusers/blob/5e70f604155aeecee254a5c63c5e4236ad4a0d3d/examples/text_to_image/train_text_to_image_sdxl.py#L1027C16-L1027C16
So i
1. Map dataset
2. Save to disk
3. Try to upload:
```
import datasets
from datasets import load_from_disk
dataset = load_from_disk("ds")
datasets.config.DEFAULT_MAX_BATCH_SIZE = 1
dataset.push_to_hub("kopyl/ds", private=True, max_shard_size="500MB")
```
And i get this error:
`pyarrow.lib.ArrowNotImplementedError: Unhandled type for Arrow to Parquet schema conversion: halffloat`
Full traceback:
```
>>> dataset.push_to_hub("kopyl/3M_icons_monochrome_only_no_captioning_mapped-for-SDXL-2", private=True, max_shard_size="500MB")
Map: 100%|βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ| 1451/1451 [00:00<00:00, 6827.40 examples/s]
Uploading the dataset shards: 0%| | 0/2099 [00:00<?, ?it/s]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/local/lib/python3.10/dist-packages/datasets/dataset_dict.py", line 1705, in push_to_hub
split_additions, uploaded_size, dataset_nbytes = self[split]._push_parquet_shards_to_hub(
File "/usr/local/lib/python3.10/dist-packages/datasets/arrow_dataset.py", line 5208, in _push_parquet_shards_to_hub
shard.to_parquet(buffer)
File "/usr/local/lib/python3.10/dist-packages/datasets/arrow_dataset.py", line 4931, in to_parquet
return ParquetDatasetWriter(self, path_or_buf, batch_size=batch_size, **parquet_writer_kwargs).write()
File "/usr/local/lib/python3.10/dist-packages/datasets/io/parquet.py", line 129, in write
written = self._write(file_obj=self.path_or_buf, batch_size=batch_size, **self.parquet_writer_kwargs)
File "/usr/local/lib/python3.10/dist-packages/datasets/io/parquet.py", line 141, in _write
writer = pq.ParquetWriter(file_obj, schema=schema, **parquet_writer_kwargs)
File "/usr/local/lib/python3.10/dist-packages/pyarrow/parquet/core.py", line 1016, in __init__
self.writer = _parquet.ParquetWriter(
File "pyarrow/_parquet.pyx", line 1869, in pyarrow._parquet.ParquetWriter.__cinit__
File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
pyarrow.lib.ArrowNotImplementedError: Unhandled type for Arrow to Parquet schema conversion: halffloat
```
Smaller datasets with the same way of saving and pushing work wonders. Big ones are not.
I'm currently trying to upload dataset like this:
`HfApi().upload_folder...`
But i'm not sure that in this case "load_dataset" would work well.
This setting num_shards does not help too:
```
dataset.push_to_hub("kopyl/3M_icons_monochrome_only_no_captioning_mapped-for-SDXL-2", private=True, num_shards={'train': 500})
```
Tried 3000, 500, 478, 100
Also do you know if it's possible to push a dataset with multiple processes? It would take an eternity pushing 1TB...
### Steps to reproduce the bug
Described above
### Expected behavior
Should be able to upload...
### Environment info
Total dataset size: 978G
Amount of `.arrow` files: 2101
Each `.arrow` file size: 477M (i know 477 megabytes * 2101 does not equal 978G, but i just checked the size of a couple `.arrow` files, i don't know if some might have different size)
Some files:
- "ds/train/state.json": https://pastebin.com/tJ3ZLGAg
- "ds/train/dataset_info.json": https://pastebin.com/JdXMQ5ih | 6,595 |
https://github.com/huggingface/datasets/issues/6594 | IterableDataset sharding logic needs improvement | [] | ### Describe the bug
The sharding of IterableDatasets with respect to distributed and dataloader worker processes appears problematic with significant performance traps and inconsistencies wrt to distributed train processes vs worker processes.
Splitting across num_workers (per train process loader processes) and world_size (distributed training processes) appears inconsistent.
* worker split: https://github.com/huggingface/datasets/blob/9d6d16117a30ba345b0236407975f701c5b288d4/src/datasets/iterable_dataset.py#L1266-L1283
* distributed split: https://github.com/huggingface/datasets/blob/9d6d16117a30ba345b0236407975f701c5b288d4/src/datasets/iterable_dataset.py#L1335-L1356
In the case of the distributed split, there is a modulus check that flips between two very different behaviours, why is this different than splitting across the data loader workers? For IterableDatasets the DataLoaders worker processes are independent, so whether it's workers within one train process or across a distributed world the shards should be distributed the same, across `world_size * num_worker` independent workers in either case...
Further, the fallback case when the `n_shards % world_size == 0` check fails is a rather extreme change. I argue it is not desirable to do that implicitly, it should be an explicit case for specific scenarios (ie reliable validation). A train scenario would likely be much better handled with improved wrapping / stopping behaviour to eg also fix #6437. Changing from stepping shards to stepping samples means that every single process reads ALL of the shards. This was never an intended default for sharded training, shards gain their performance advantage in large scale distributed training by explicitly avoiding the need to have every process overlapping in the data they read, by default, only the data allocated to each process via their assigned shards should be read in each pass of the dataset.
Using a large scale CLIP example, some of the larger datasets have 10-20k shards across 100+TB of data. Training with 1000 GPUs we are switching between reading 100 terabytes per epoch to 100 petabytes if say change 20k % 1000 and drop one gpu-node to 20k % 992.
The 'step over samples' case might be worth the overhead in specific validation scenarios where gaurantees of at least/most once samples seen are more important and do not make up a significant portion of train time or are done in smaller world sizes outside of train.
### Steps to reproduce the bug
N/A
### Expected behavior
We have an iterable dataset with N shards, to split across workers
* shuffle shards (same seed across all train processes)
* step shard iterator across distributed processes
* step shard iterator across dataloader worker processes
* shuffle samples in every worker via shuffle buffer (different seed in each worker, but ideally controllable (based on base seed + worker id + epoch).
* end up with (possibly uneven) number of shards per worker but each shard only ever accessed by 1 worker per pass (epoch)
### Environment info
N/A | 6,594 |
https://github.com/huggingface/datasets/issues/6592 | Logs are delayed when doing .map when `docker logs` | [
"Hi! `tqdm` doesn't work well in non-interactive environments, so there isn't much we can do about this. It's best to [disable it](https://huggingface.co/docs/datasets/v2.16.1/en/package_reference/utilities#datasets.disable_progress_bars) in such environments and instead use logging to track progress."
] | ### Describe the bug
When I run my SD training in a Docker image and then listen to logs like `docker logs train -f`, the progress bar is delayed.
It's updating every few percent.
When you have a large dataset that has to be mapped (like 1+ million samples), it's crucial to see the updates in real-time, not every couple hours to make sure nothing got frozen or broken
### Steps to reproduce the bug
1. Run any huge dataset processing as a Docker image
2. `docker logs image_name` to it
### Expected behavior
...
### Environment info
... | 6,592 |
https://github.com/huggingface/datasets/issues/6591 | The datasets models housed in Dropbox can't support a lot of users downloading them | [
"Hi! Indeed, Dropbox is not a reliable host. I've just merged https://huggingface.co/datasets/PolyAI/minds14/discussions/24 to fix this by hosting the data files inside the repo."
] | ### Describe the bug
I'm using the datasets
```
from datasets import load_dataset, Audio
dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")
```
And it seems that sometimes when I imagine a lot of users are accessing the same resources, the Dropbox host fails:
`raise ConnectionError(f"Couldn't reach {url} (error {response.status_code})") ConnectionError: Couldn't reach https://www.dropbox.com/s/e2us0hcs3ilr20e/MInDS-14.zip?dl=1 (error 429)`
My question is if we can somehow host these files elsewhere or can you change the limit of simultaneous users accessing those resources or any other solution?
Also, has anyone had this issue before?
Thanks
### Steps to reproduce the bug
1: Create a python script like so:
```
from datasets import load_dataset, Audio
dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")
```
2: Execute this by a certain number of users at the same time
### Expected behavior
I woudl expect that this shouldnt happen unless its a huge amount of users, which it is not the case
### Environment info
This was done in an Ubuntu 22 environment. | 6,591 |
https://github.com/huggingface/datasets/issues/6590 | Feature request: Multi-GPU dataset mapping for SDXL training | [] | ### Feature request
We need to speed up SDXL dataset pre-process. Please make it possible to use multiple GPUs for the [official SDXL trainer](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_sdxl.py) :)
### Motivation
Pre-computing 3 million of images takes around 2 days.
Would be nice to be able to be able to do multi-GPU (or even better β multi-GPU + multi-node) vae and embedding precompute...
### Your contribution
I'm not sure i can wrap my head around the multi-GPU mapping...
Plus it's too expensive for me to take x2 A100 and spend a day just figuring out the staff since I don't have a job right now. | 6,590 |
https://github.com/huggingface/datasets/issues/6589 | After `2.16.0` version, there are `PermissionError` when users use shared cache_dir | [
"We'll do a new release of `datasets` in the coming days with a fix !",
"@lhoestq Thank you very much!"
] | ### Describe the bug
- We use shared `cache_dir` using `HF_HOME="{shared_directory}"`
- After dataset version 2.16.0, datasets uses `filelock` package for file locking #6445
- But, `filelock` package make `.lock` file with `644` permission
- Dataset is not available to other users except the user who created the lock file via `load_dataset`.
### Steps to reproduce the bug
1. `pip install datasets==2.16.0`
2. `export HF_HOME="{shared_directory}"`
3. download dataset with `load_dataset`
4. logout and login another user
5. `pip install datasets==2.16.0`
6. `export HF_HOME="{shared_directory}"`
7. download dataset with `load_dataset`
8. `PermissionError` occurs
### Expected behavior
- Users can share `cache_dir` using environment variable `HF_HOME`
### Environment info
- python == 3.9.10
- datasets == 2.16.0
- ubuntu 22.04
- shared_directory has ACL
- users are same group (developers)
| 6,589 |