Commit
•
1312980
1
Parent(s):
ca6bd3d
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,13 +1,18 @@
|
|
1 |
---
|
|
|
2 |
language:
|
3 |
- fr
|
4 |
-
license: apache-2.0
|
5 |
multilinguality:
|
6 |
- monolingual
|
7 |
-
|
8 |
-
-
|
|
|
|
|
|
|
|
|
9 |
source_datasets:
|
10 |
- original
|
|
|
11 |
task_categories:
|
12 |
- text-generation
|
13 |
- table-question-answering
|
@@ -15,40 +20,10 @@ task_categories:
|
|
15 |
- text-retrieval
|
16 |
- question-answering
|
17 |
- text-classification
|
18 |
-
|
19 |
-
|
20 |
-
- finetuning
|
21 |
-
- legal
|
22 |
-
- french law
|
23 |
-
- droit français
|
24 |
-
- Code de la mutualité
|
25 |
-
dataset_info:
|
26 |
-
features:
|
27 |
-
- name: instruction
|
28 |
-
dtype: string
|
29 |
-
- name: input
|
30 |
-
dtype: string
|
31 |
-
- name: output
|
32 |
-
dtype: string
|
33 |
-
- name: start
|
34 |
-
dtype: string
|
35 |
-
- name: expiration
|
36 |
-
dtype: string
|
37 |
-
- name: num
|
38 |
-
dtype: string
|
39 |
-
splits:
|
40 |
-
- name: train
|
41 |
-
num_bytes: 693958
|
42 |
-
num_examples: 526
|
43 |
-
download_size: 275793
|
44 |
-
dataset_size: 693958
|
45 |
-
configs:
|
46 |
-
- config_name: default
|
47 |
-
data_files:
|
48 |
-
- split: train
|
49 |
-
path: data/train-*
|
50 |
---
|
51 |
-
# Code de la mutualité, non-instruct (2024-
|
52 |
|
53 |
This project focuses on fine-tuning pre-trained language models to create efficient and accurate models for legal practice.
|
54 |
|
@@ -64,6 +39,120 @@ Instruction-based fine-tuning significantly enhances the performance of LLMs in
|
|
64 |
- Interpretability: Instruction-based fine-tuning also makes LLM behavior more interpretable. Since the instructions are human-readable, it becomes easier to understand and control model outputs.
|
65 |
- Adaptive Behavior: LLMs, post instruction-based fine-tuning, exhibit adaptive behavior that is responsive to both explicit task descriptions and implicit cues within the provided text.
|
66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
## Dataset generation
|
68 |
|
69 |
This JSON file is a list of dictionaries, each dictionary contains the following fields:
|
|
|
1 |
---
|
2 |
+
license: apache-2.0
|
3 |
language:
|
4 |
- fr
|
|
|
5 |
multilinguality:
|
6 |
- monolingual
|
7 |
+
tags:
|
8 |
+
- finetuning
|
9 |
+
- legal
|
10 |
+
- french law
|
11 |
+
- droit français
|
12 |
+
- Code de la mutualité
|
13 |
source_datasets:
|
14 |
- original
|
15 |
+
pretty_name: Code de la mutualité
|
16 |
task_categories:
|
17 |
- text-generation
|
18 |
- table-question-answering
|
|
|
20 |
- text-retrieval
|
21 |
- question-answering
|
22 |
- text-classification
|
23 |
+
size_categories:
|
24 |
+
- 1K<n<10K
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
---
|
26 |
+
# Code de la mutualité, non-instruct (2024-04-01)
|
27 |
|
28 |
This project focuses on fine-tuning pre-trained language models to create efficient and accurate models for legal practice.
|
29 |
|
|
|
39 |
- Interpretability: Instruction-based fine-tuning also makes LLM behavior more interpretable. Since the instructions are human-readable, it becomes easier to understand and control model outputs.
|
40 |
- Adaptive Behavior: LLMs, post instruction-based fine-tuning, exhibit adaptive behavior that is responsive to both explicit task descriptions and implicit cues within the provided text.
|
41 |
|
42 |
+
## Concurrent reading of the LegalKit
|
43 |
+
|
44 |
+
To use all the legal data published on LegalKit, you can use this code snippet:
|
45 |
+
```python
|
46 |
+
# -*- coding: utf-8 -*-
|
47 |
+
import concurrent.futures
|
48 |
+
import os
|
49 |
+
|
50 |
+
import datasets
|
51 |
+
from tqdm.notebook import tqdm
|
52 |
+
|
53 |
+
def dataset_loader(
|
54 |
+
name:str,
|
55 |
+
streaming:bool=True
|
56 |
+
) -> datasets.Dataset:
|
57 |
+
"""
|
58 |
+
Helper function to load a single dataset in parallel.
|
59 |
+
|
60 |
+
Parameters
|
61 |
+
----------
|
62 |
+
name : str
|
63 |
+
Name of the dataset to be loaded.
|
64 |
+
|
65 |
+
streaming : bool, optional
|
66 |
+
Determines if datasets are streamed. Default is True.
|
67 |
+
|
68 |
+
Returns
|
69 |
+
-------
|
70 |
+
dataset : datasets.Dataset
|
71 |
+
Loaded dataset object.
|
72 |
+
|
73 |
+
Raises
|
74 |
+
------
|
75 |
+
Exception
|
76 |
+
If an error occurs during dataset loading.
|
77 |
+
"""
|
78 |
+
try:
|
79 |
+
return datasets.load_dataset(
|
80 |
+
name,
|
81 |
+
split="train",
|
82 |
+
streaming=streaming
|
83 |
+
)
|
84 |
+
|
85 |
+
except Exception as exc:
|
86 |
+
logging.error(f"Error loading dataset {name}: {exc}")
|
87 |
+
|
88 |
+
return None
|
89 |
+
|
90 |
+
|
91 |
+
def load_datasets(
|
92 |
+
req:list,
|
93 |
+
streaming:bool=True
|
94 |
+
) -> list:
|
95 |
+
"""
|
96 |
+
Downloads datasets specified in a list and creates a list of loaded datasets.
|
97 |
+
|
98 |
+
Parameters
|
99 |
+
----------
|
100 |
+
req : list
|
101 |
+
A list containing the names of datasets to be downloaded.
|
102 |
+
|
103 |
+
streaming : bool, optional
|
104 |
+
Determines if datasets are streamed. Default is True.
|
105 |
+
|
106 |
+
Returns
|
107 |
+
-------
|
108 |
+
datasets_list : list
|
109 |
+
A list containing loaded datasets as per the requested names provided in 'req'.
|
110 |
+
|
111 |
+
Raises
|
112 |
+
------
|
113 |
+
Exception
|
114 |
+
If an error occurs during dataset loading or processing.
|
115 |
+
|
116 |
+
Examples
|
117 |
+
--------
|
118 |
+
>>> datasets = load_datasets(["dataset1", "dataset2"], streaming=False)
|
119 |
+
"""
|
120 |
+
datasets_list = []
|
121 |
+
|
122 |
+
with concurrent.futures.ThreadPoolExecutor() as executor:
|
123 |
+
future_to_dataset = {executor.submit(dataset_loader, name): name for name in req}
|
124 |
+
|
125 |
+
for future in tqdm(concurrent.futures.as_completed(future_to_dataset), total=len(req)):
|
126 |
+
name = future_to_dataset[future]
|
127 |
+
|
128 |
+
try:
|
129 |
+
dataset = future.result()
|
130 |
+
|
131 |
+
if dataset:
|
132 |
+
datasets_list.append(dataset)
|
133 |
+
|
134 |
+
except Exception as exc:
|
135 |
+
logging.error(f"Error processing dataset {name}: {exc}")
|
136 |
+
|
137 |
+
return datasets_list
|
138 |
+
|
139 |
+
|
140 |
+
req = [
|
141 |
+
"louisbrulenaudet/code-artisanat",
|
142 |
+
"louisbrulenaudet/code-action-sociale-familles",
|
143 |
+
# ...
|
144 |
+
]
|
145 |
+
|
146 |
+
datasets_list = load_datasets(
|
147 |
+
req=req,
|
148 |
+
streaming=True
|
149 |
+
)
|
150 |
+
|
151 |
+
dataset = datasets.concatenate_datasets(
|
152 |
+
datasets_list
|
153 |
+
)
|
154 |
+
```
|
155 |
+
|
156 |
## Dataset generation
|
157 |
|
158 |
This JSON file is a list of dictionaries, each dictionary contains the following fields:
|