Update README.md

README.md

@@ -62,6 +62,10 @@ configs:
   data_files:
   - path: data/natural/it-en/*/*.parquet
     split: train
+- config_name: natural
+  data_files:
+  - path: data/natural/*/*/*.parquet
+    split: train
 - config_name: code
   data_files:
   - path: data/code/*/*/*parquet
@@ -403,3 +407,97 @@ configs:
   - path: data/natural/*/YouTube/*.parquet
     split: train
 ---
+
+# Dataset Card
+
+The Lucie Training Dataset is a curated collection of text data
+in English, French, German, Spanish and Italian,
+from the web,
+video subtitles,
+collections of books, newspapers, monographies, and magazines processed by Optical Character Recognition (OCR),
+as well as collections of files in diverse programming languages.
+
+It was used to pretrain [Lucie-7B](https://huggingface.co/OpenLLM-France/Lucie-7B),
+a foundation LLM with strong capabilities in French and English.
+
+## Dataset Description
+
+This dataset was made to provide an extensive and diverse dataset for training Large Language Models (LLM),
+with the following motivations in mind:
+* Data mix:
+    * French is as well represented as English
+    (Lucie Training Dataset is one of the biggest of collection of French text data with a minimum of quality),
+    * German, Spanish and Italian are also represented to some extend,
+    * Code is also included to boost the reasoning capabilities of LLM.
+* Data filtering and deduplication:
+    * The dataset is cleaned low-quality data
+    * The dataset is cleaned from duplicates to some extend, following best practices.
+* Ethics:
+    * A special care was taken to respect copyright laws and the privacy of individuals.
+      All books, newspapers, monographies, and magazines are in the public domain
+  (which depends on the author's death date, and the country of publication).
+    * There is no data from the web for which robots.txt files forbid crawling.
+
+### Dataset Structure
+
+The corpus contains the following information for each text sample:
+* `text`: the text sample itself.
+* `source`: an identifier for the source(s) of the text sample (`Wikipedia`, `RedPajama`, `Gutenberg`, …).
+  The list of all sources is described in this document.
+* `id`: an identifier that is unique among the source.
+* `language`: the language of the text sample, which can be:
+    * the ISO 639-1 code of a natural language: `en`, `fr`, `de`, `es`, or `it`;
+    * the common name prefixed by "`code:`" of a programming language:  `code:python`, `code:c++`, …; or
+    * a list of ISO 639-1 codes separated by commas, if the text sample is multilingual: `fr,en`, `de,fr`, `es,en`, `it,en`
+     (or in the opposite order if the languages appear in the opposite order in the text).
+* `url` (optional): the URL of the original text sample on the web, if available.
+* `title` (optional): the title of the original text sample, if available.
+* `author` (optional): the author of the original text sample, if available.
+   Usually the author name in plain text, except for `Gutenberg` where it is the JSON serialized object of the author metadata.
+* `date` (optional): the publication date of the original text sample, if available. The text format of the source depends on the source.
+* `quality_signals` (optional): a list of quality signals about the text sample (that could be used for further filtering or sample weighting).
+  It can include indicators computed by `fasttext` and `CCNet`, statistics about occurrences of characters, words, special characters, etc.
+  This field is always a JSON serialized object.
+* `extra` (optional): JSON serialized extra information about the text sample.
+  This can include metadata about the source subset, the rights, etc.
+
+Examples of metadata (except from `text`) are shown for each source in [metadata_examples.json](metadata_examples.json).
+
+### Example use in python
+
+Load the dataset using the `datasets` library:    
+```python
+from datasets import load_dataset
+
+kwargs = {"split": "train", "streaming": True}
+
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", **kwargs)
+```
+
+Several configurations are available to select a language, a source, or both, illustrated in the following examples.
+
+Only load data in French:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "fr", **kwargs)
+```
+Load data that is aligned in French and English:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "fr,en", **kwargs)
+```
+Only load data corresponding to programming languages:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "code", **kwargs)
+```
+Only load data in python:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "code:python", **kwargs)
+```
+Only load data from Wikipedia:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "Wikipedia", **kwargs)
+```
+Only load data from Wikipedia in French:
+```python
+dataset = load_dataset("OpenLLM-France/Lucie-Training-Dataset", "Wikipedia-fr", **kwargs)
+```
+