louisbrulenaudet commited on
Commit
5ffd666
1 Parent(s): 95843a0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +106 -0
README.md CHANGED
@@ -1,3 +1,109 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
+ language:
4
+ - fr
5
+ multilinguality:
6
+ - monolingual
7
+ tags:
8
+ - finetuning
9
+ - legal
10
+ - tax
11
+ - llm
12
+ - fiscal
13
+ - cgi
14
+ - Code Général des Impôts
15
+ source_datasets:
16
+ - original
17
+ pretty_name: Code Général des Impôts (CGI)
18
+ task_categories:
19
+ - text-generation
20
+ - table-question-answering
21
+ - summarization
22
+ - conversational
23
+ size_categories:
24
+ - 1K<n<10K
25
  ---
26
+ # Code Général des Impôts, non-instruct (11-12-2023)
27
+
28
+ This project focuses on fine-tuning pre-trained language models to create efficient and accurate models for tax practice.
29
+
30
+ Fine-tuning is the process of adapting a pre-trained model to perform specific tasks or cater to particular domains. It involves adjusting the model's parameters through a further round of training on task-specific or domain-specific data. While conventional fine-tuning strategies involve supervised learning with labeled data, instruction-based fine-tuning introduces a more structured and interpretable approach.
31
+
32
+ Instruction-based fine-tuning leverages the power of human-provided instructions to guide the model's behavior. These instructions can be in the form of text prompts, prompts with explicit task descriptions, or a combination of both. This approach allows for a more controlled and context-aware interaction with the LLM, making it adaptable to a multitude of specialized tasks.
33
+
34
+ Instruction-based fine-tuning significantly enhances the performance of LLMs in the following ways:
35
+
36
+ - Task-Specific Adaptation: LLMs, when fine-tuned with specific instructions, exhibit remarkable adaptability to diverse tasks. They can switch seamlessly between translation, summarization, and question-answering, guided by the provided instructions.
37
+ - Reduced Ambiguity: Traditional LLMs might generate ambiguous or contextually inappropriate responses. Instruction-based fine-tuning allows for a clearer and more context-aware generation, reducing the likelihood of nonsensical outputs.
38
+ - Efficient Knowledge Transfer: Instructions can encapsulate domain-specific knowledge, enabling LLMs to benefit from expert guidance. This knowledge transfer is particularly valuable in fields like tax practice, law, medicine, and more.
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
+ ## Dataset generation
43
+
44
+ This JSON file is a list of dictionaries, each dictionary contains the following fields:
45
+
46
+ - `instruction`: `string`, presenting the instruction linked to the element.
47
+ - `input`: `string`, signifying the input details for the element.
48
+ - `output`: `string`, indicating the output information for the element.
49
+
50
+ We used the following list of instructions for generating the dataset:
51
+ ```python
52
+ instructions = [
53
+ "Compose l'intégralité de l'article sous forme écrite.",
54
+ "Écris la totalité du contenu de l'article.",
55
+ "Formule la totalité du texte présent dans l'article.",
56
+ "Produis l'intégralité de l'article en écriture.",
57
+ "Développe l'article dans son ensemble par écrit.",
58
+ "Génère l'ensemble du texte contenu dans l'article.",
59
+ "Formule le contenu intégral de l'article en entier.",
60
+ "Rédige la totalité du texte de l'article en entier.",
61
+ "Compose l'intégralité du contenu textuel de l'article.",
62
+ "Rédige l'ensemble du texte qui constitue l'article.",
63
+ "Formule l'article entier dans son contenu écrit.",
64
+ "Composez l'intégralité de l'article sous forme écrite.",
65
+ "Écrivez la totalité du contenu de l'article.",
66
+ "Formulez la totalité du texte présent dans l'article.",
67
+ "Développez l'article dans son ensemble par écrit.",
68
+ "Générez l'ensemble du texte contenu dans l'article.",
69
+ "Formulez le contenu intégral de l'article en entier.",
70
+ "Rédigez la totalité du texte de l'article en entier.",
71
+ "Composez l'intégralité du contenu textuel de l'article.",
72
+ "Écrivez l'article dans son intégralité en termes de texte.",
73
+ "Rédigez l'ensemble du texte qui constitue l'article.",
74
+ "Formulez l'article entier dans son contenu écrit.",
75
+ "Composer l'intégralité de l'article sous forme écrite.",
76
+ "Écrire la totalité du contenu de l'article.",
77
+ "Formuler la totalité du texte présent dans l'article.",
78
+ "Produire l'intégralité de l'article en écriture.",
79
+ "Développer l'article dans son ensemble par écrit.",
80
+ "Générer l'ensemble du texte contenu dans l'article.",
81
+ "Formuler le contenu intégral de l'article en entier.",
82
+ "Rédiger la totalité du texte de l'article en entier.",
83
+ "Composer l'intégralité du contenu textuel de l'article.",
84
+ "Rédiger l'ensemble du texte qui constitue l'article.",
85
+ "Formuler l'article entier dans son contenu écrit.",
86
+ "Quelles sont les dispositions de l'article ?",
87
+ "Quelles dispositions sont incluses dans l'article ?",
88
+ "Quelles sont les dispositions énoncées dans l'article ?",
89
+ "Quel est le texte intégral de l'article ?",
90
+ "Quelle est la lettre de l'article ?"
91
+ ]
92
+ ```
93
+
94
+ ## Citing this project
95
+
96
+ If you use this code in your research, please use the following BibTeX entry.
97
+
98
+ ```BibTeX
99
+ @misc{louisbrulenaudet2023,
100
+ author = {Louis Brulé Naudet},
101
+ title = {Code Général des Impôts, non-instruct (11-12-2023)},
102
+ howpublished = {\url{https://huggingface.co/datasets/louisbrulenaudet/cgi}},
103
+ year = {2023}
104
+ }
105
+ ```
106
+
107
+ ## Feedback
108
+
109
+ If you have any feedback, please reach out at [louisbrulenaudet@icloud.com](mailto:louisbrulenaudet@icloud.com).