louisbrulenaudet commited on
Commit
8e4280e
·
verified ·
1 Parent(s): c52be45

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +238 -1
README.md CHANGED
@@ -44,4 +44,241 @@ size_categories:
44
 
45
  ## Dataset Description
46
  - **Repository:** https://huggingface.co/datasets/louisbrulenaudet/lemone-docs-embeded
47
- - **Point of Contact:** [Louis Brulé Naudet](mailto:[email protected])
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
 
45
  ## Dataset Description
46
  - **Repository:** https://huggingface.co/datasets/louisbrulenaudet/lemone-docs-embeded
47
+ - **Point of Contact:** [Louis Brulé Naudet](mailto:[email protected])
48
+
49
+ <img src="assets/thumbnail.webp">
50
+
51
+ # Lemone-embeded, pre-built embeddings dataset for French taxation.
52
+
53
+ <div class="not-prose bg-gradient-to-r from-gray-50-to-white text-gray-900 border" style="border-radius: 8px; padding: 0.5rem 1rem;">
54
+ <p>This database presents the embeddings generated by the Lemone-embed-pro model and aims at a large-scale distribution of the model even for the GPU-poor.</p>
55
+ </div>
56
+
57
+ This sentence transformers model, specifically designed for French taxation, has been fine-tuned on a dataset comprising 43 million tokens, integrating a blend of semi-synthetic and fully synthetic data generated by GPT-4 Turbo and Llama 3.1 70B, which have been further refined through evol-instruction tuning and manual curation.
58
+
59
+ The model is tailored to meet the specific demands of information retrieval across large-scale tax-related corpora, supporting the implementation of production-ready Retrieval-Augmented Generation (RAG) applications. Its primary purpose is to enhance the efficiency and accuracy of legal processes in the taxation domain, with an emphasis on delivering consistent performance in real-world settings, while also contributing to advancements in legal natural language processing research.
60
+
61
+ This is a sentence-transformers model finetuned from Alibaba-NLP/gte-multilingual-base. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
62
+
63
+ ## Usage with ChromaDB
64
+
65
+ We recommend integration via a vector-store to produce an optimal RAG pipeline. Here's a code extract for producing such a database with ChromaDB:
66
+
67
+ ```python
68
+ import chromadb
69
+ import polars as pl
70
+
71
+ from chromadb.config import Settings
72
+ from chromadb.utils import embedding_functions
73
+ from torch.cuda import is_available
74
+
75
+ client = chromadb.PersistentClient(
76
+ path="./chroma.db",
77
+ settings=Settings(anonymized_telemetry=False)
78
+ )
79
+
80
+ sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(
81
+ model_name="louisbrulenaudet/lemone-embed-pro",
82
+ device="cuda" if is_available() else "cpu",
83
+ trust_remote_code=True
84
+ )
85
+
86
+ collection = client.get_or_create_collection(
87
+ name="tax",
88
+ embedding_function=sentence_transformer_ef
89
+ )
90
+
91
+ dataframe = pl.scan_parquet('hf://datasets/louisbrulenaudet/lemone-docs-embeded/data/train-00000-of-00001.parquet').filter(
92
+ pl.col(
93
+ "text"
94
+ ).is_not_null()
95
+ ).collect()
96
+
97
+ collection.add(
98
+ embeddings=dataframe["lemone_pro_embeddings"].to_list(),
99
+ documents=dataframe["text"].to_list(),
100
+ metadatas=dataframe.drop(
101
+ [
102
+ "lemone_pro_embeddings",
103
+ "text"
104
+ ]
105
+ ).to_dicts(),
106
+ ids=[
107
+ str(i) for i in range(0, dataframe.shape[0])
108
+ ]
109
+ )
110
+ ```
111
+
112
+ Here is a code for reproduction of this dataset:
113
+
114
+ ```python
115
+ import hashlib
116
+
117
+ from datetime import datetime
118
+ from typing import (
119
+ IO,
120
+ TYPE_CHECKING,
121
+ Any,
122
+ Dict,
123
+ List,
124
+ Type,
125
+ Tuple,
126
+ Union,
127
+ Mapping,
128
+ TypeVar,
129
+ Callable,
130
+ Optional,
131
+ Sequence,
132
+ )
133
+
134
+ import chromadb
135
+ import polars as pl
136
+
137
+ from chromadb.config import Settings
138
+ from chromadb.utils import embedding_functions
139
+ from torch.cuda import is_available
140
+
141
+ client = chromadb.Client(
142
+ settings=Settings(anonymized_telemetry=False)
143
+ )
144
+
145
+ sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(
146
+ model_name="louisbrulenaudet/lemone-embed-pro",
147
+ device="cuda" if is_available() else "cpu",
148
+ trust_remote_code=True
149
+ )
150
+
151
+ collection = client.get_or_create_collection(
152
+ name="tax",
153
+ embedding_function=sentence_transformer_ef
154
+ )
155
+
156
+ bofip_dataframe = pl.scan_parquet(
157
+ "hf://datasets/louisbrulenaudet/bofip/data/train-00000-of-00001.parquet"
158
+ ).with_columns(
159
+ [
160
+ (
161
+ pl.lit("Bulletin officiel des finances publiques - impôts").alias(
162
+ "title_main"
163
+ )
164
+ ),
165
+ (
166
+ pl.col("debut_de_validite")
167
+ .str.strptime(pl.Date, format="%Y-%m-%d")
168
+ .dt.strftime("%Y-%m-%d 00:00:00")
169
+ ).alias("date_publication"),
170
+ (
171
+ pl.col("contenu")
172
+ .map_elements(lambda x: hashlib.sha256(str(x).encode()).hexdigest(), return_dtype=pl.Utf8)
173
+ .alias("hash")
174
+ )
175
+ ]
176
+ ).rename(
177
+ {
178
+ "contenu": "text",
179
+ "permalien": "url_sourcepage",
180
+ "identifiant_juridique": "id_sub",
181
+ }
182
+ ).select(
183
+ [
184
+ "text",
185
+ "title_main",
186
+ "id_sub",
187
+ "url_sourcepage",
188
+ "date_publication",
189
+ "hash"
190
+ ]
191
+ )
192
+
193
+ books: List[str] = [
194
+ "hf://datasets/louisbrulenaudet/code-douanes/data/train-00000-of-00001.parquet",
195
+ "hf://datasets/louisbrulenaudet/code-impots/data/train-00000-of-00001.parquet",
196
+ "hf://datasets/louisbrulenaudet/code-impots-annexe-i/data/train-00000-of-00001.parquet",
197
+ "hf://datasets/louisbrulenaudet/code-impots-annexe-ii/data/train-00000-of-00001.parquet",
198
+ "hf://datasets/louisbrulenaudet/code-impots-annexe-iii/data/train-00000-of-00001.parquet",
199
+ "hf://datasets/louisbrulenaudet/code-impots-annexe-iv/data/train-00000-of-00001.parquet",
200
+ "hf://datasets/louisbrulenaudet/code-impositions-biens-services/data/train-00000-of-00001.parquet",
201
+ "hf://datasets/louisbrulenaudet/livre-procedures-fiscales/data/train-00000-of-00001.parquet"
202
+ ]
203
+
204
+ legi_dataframe = pl.concat(
205
+ [
206
+ pl.scan_parquet(
207
+ book
208
+ ) for book in books
209
+ ]
210
+ ).with_columns(
211
+ [
212
+ (
213
+ pl.lit("https://www.legifrance.gouv.fr/codes/article_lc/")
214
+ .add(pl.col("id"))
215
+ .alias("url_sourcepage")
216
+ ),
217
+ (
218
+ pl.col("dateDebut")
219
+ .cast(pl.Int64)
220
+ .map_elements(
221
+ lambda x: datetime.fromtimestamp(x / 1000).strftime("%Y-%m-%d %H:%M:%S"),
222
+ return_dtype=pl.Utf8
223
+ )
224
+ .alias("date_publication")
225
+ ),
226
+ (
227
+ pl.col("texte")
228
+ .map_elements(lambda x: hashlib.sha256(str(x).encode()).hexdigest(), return_dtype=pl.Utf8)
229
+ .alias("hash")
230
+ )
231
+ ]
232
+ ).rename(
233
+ {
234
+ "texte": "text",
235
+ "num": "id_sub",
236
+ }
237
+ ).select(
238
+ [
239
+ "text",
240
+ "title_main",
241
+ "id_sub",
242
+ "url_sourcepage",
243
+ "date_publication",
244
+ "hash"
245
+ ]
246
+ )
247
+
248
+ print("Starting embeddings production...")
249
+
250
+ dataframe = pl.concat(
251
+ [
252
+ bofip_dataframe,
253
+ legi_dataframe
254
+ ]
255
+ ).filter(
256
+ pl.col(
257
+ "text"
258
+ ).is_not_null()
259
+ ).with_columns(
260
+ pl.col("text").map_elements(
261
+ lambda x: sentence_transformer_ef(
262
+ [x]
263
+ )[0].tolist(),
264
+ return_dtype=pl.List(pl.Float64)
265
+ ).alias("lemone_pro_embeddings")
266
+ ).collect()
267
+ ```
268
+
269
+ ## Citation
270
+
271
+ If you use this code in your research, please use the following BibTeX entry.
272
+
273
+ ```BibTeX
274
+ @misc{louisbrulenaudet2024,
275
+ author = {Louis Brulé Naudet},
276
+ title = {Lemone-Embed: A Series of Fine-Tuned Embedding Models for French Taxation},
277
+ year = {2024}
278
+ howpublished = {\url{https://huggingface.co/datasets/louisbrulenaudet/lemone-embed-pro}},
279
+ }
280
+ ```
281
+
282
+ ## Feedback
283
+
284
+ If you have any feedback, please reach out at [[email protected]](mailto:[email protected]).