Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
Jeronymous commited on
Commit
c5ed47b
·
verified ·
1 Parent(s): ec75a27

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +98 -0
README.md CHANGED
@@ -62,6 +62,10 @@ configs:
62
  data_files:
63
  - path: data/natural/it-en/*/*.parquet
64
  split: train
 
 
 
 
65
  - config_name: code
66
  data_files:
67
  - path: data/code/*/*/*parquet
@@ -403,3 +407,97 @@ configs:
403
  - path: data/natural/*/YouTube/*.parquet
404
  split: train
405
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  data_files:
63
  - path: data/natural/it-en/*/*.parquet
64
  split: train
65
+ - config_name: natural
66
+ data_files:
67
+ - path: data/natural/*/*/*.parquet
68
+ split: train
69
  - config_name: code
70
  data_files:
71
  - path: data/code/*/*/*parquet
 
407
  - path: data/natural/*/YouTube/*.parquet
408
  split: train
409
  ---
410
+
411
+ # Dataset Card
412
+
413
+ The Lucie Training Dataset is a curated collection of text data
414
+ in English, French, German, Spanish and Italian,
415
+ from the web,
416
+ video subtitles,
417
+ collections of books, newspapers, monographies, and magazines processed by Optical Character Recognition (OCR),
418
+ as well as collections of files in diverse programming languages.
419
+
420
+ It was used to pretrain [Lucie-7B](https://huggingface.co/OpenLLM-France/Lucie-7B),
421
+ a foundation LLM with strong capabilities in French and English.
422
+
423
+ ## Dataset Description
424
+
425
+ This dataset was made to provide an extensive and diverse dataset for training Large Language Models (LLM),
426
+ with the following motivations in mind:
427
+ * Data mix:
428
+ * French is as well represented as English
429
+ (Lucie Training Dataset is one of the biggest of collection of French text data with a minimum of quality),
430
+ * German, Spanish and Italian are also represented to some extend,
431
+ * Code is also included to boost the reasoning capabilities of LLM.
432
+ * Data filtering and deduplication:
433
+ * The dataset is cleaned low-quality data
434
+ * The dataset is cleaned from duplicates to some extend, following best practices.
435
+ * Ethics:
436
+ * A special care was taken to respect copyright laws and the privacy of individuals.
437
+ All books, newspapers, monographies, and magazines are in the public domain
438
+ (which depends on the author's death date, and the country of publication).
439
+ * There is no data from the web for which robots.txt files forbid crawling.
440
+
441
+ ### Dataset Structure
442
+
443
+ The corpus contains the following information for each text sample:
444
+ * `text`: the text sample itself.
445
+ * `source`: an identifier for the source(s) of the text sample (`Wikipedia`, `RedPajama`, `Gutenberg`, …).
446
+ The list of all sources is described in this document.
447
+ * `id`: an identifier that is unique among the source.
448
+ * `language`: the language of the text sample, which can be:
449
+ * the ISO 639-1 code of a natural language: `en`, `fr`, `de`, `es`, or `it`;
450
+ * the common name prefixed by "`code:`" of a programming language: `code:python`, `code:c++`, …; or
451
+ * a list of ISO 639-1 codes separated by commas, if the text sample is multilingual: `fr,en`, `de,fr`, `es,en`, `it,en`
452
+ (or in the opposite order if the languages appear in the opposite order in the text).
453
+ * `url` (optional): the URL of the original text sample on the web, if available.
454
+ * `title` (optional): the title of the original text sample, if available.
455
+ * `author` (optional): the author of the original text sample, if available.
456
+ Usually the author name in plain text, except for `Gutenberg` where it is the JSON serialized object of the author metadata.
457
+ * `date` (optional): the publication date of the original text sample, if available. The text format of the source depends on the source.
458
+ * `quality_signals` (optional): a list of quality signals about the text sample (that could be used for further filtering or sample weighting).
459
+ It can include indicators computed by `fasttext` and `CCNet`, statistics about occurrences of characters, words, special characters, etc.
460
+ This field is always a JSON serialized object.
461
+ * `extra` (optional): JSON serialized extra information about the text sample.
462
+ This can include metadata about the source subset, the rights, etc.
463
+
464
+ Examples of metadata (except from `text`) are shown for each source in [metadata_examples.json](metadata_examples.json).
465
+
466
+ ### Example use in python
467
+
468
+ Load the dataset using the `datasets` library:
469
+ ```python
470
+ from datasets import load_dataset
471
+
472
+ kwargs = {"split": "train", "streaming": True}
473
+
474
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", **kwargs)
475
+ ```
476
+
477
+ Several configurations are available to select a language, a source, or both, illustrated in the following examples.
478
+
479
+ Only load data in French:
480
+ ```python
481
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "fr", **kwargs)
482
+ ```
483
+ Load data that is aligned in French and English:
484
+ ```python
485
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "fr,en", **kwargs)
486
+ ```
487
+ Only load data corresponding to programming languages:
488
+ ```python
489
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "code", **kwargs)
490
+ ```
491
+ Only load data in python:
492
+ ```python
493
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "code:python", **kwargs)
494
+ ```
495
+ Only load data from Wikipedia:
496
+ ```python
497
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "Wikipedia", **kwargs)
498
+ ```
499
+ Only load data from Wikipedia in French:
500
+ ```python
501
+ dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "Wikipedia-fr", **kwargs)
502
+ ```
503
+