christophebourguignat
commited on
Commit
•
73f994d
1
Parent(s):
8781bb0
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import gradio as gr
|
3 |
+
from datasets import load_dataset
|
4 |
+
from langchain.document_loaders import DataFrameLoader
|
5 |
+
from langchain.text_splitter import CharacterTextSplitter
|
6 |
+
from langchain.vectorstores import FAISS
|
7 |
+
from langchain.embeddings import OpenAIEmbeddings
|
8 |
+
from langchain.chains import RetrievalQA
|
9 |
+
from langchain.chat_models import ChatOpenAI
|
10 |
+
|
11 |
+
def llm_response(insurer1, insurer2, question):
|
12 |
+
qa_chain1 = RetrievalQA.from_chain_type(llm1, retriever=db_dict[insurer1].as_retriever(search_kwargs={'k': 15}))
|
13 |
+
qa_chain2 = RetrievalQA.from_chain_type(llm2, retriever=db_dict[insurer2].as_retriever(search_kwargs={'k': 15}))
|
14 |
+
return qa_chain1({"query": question})['result'], qa_chain2({"query": question})['result']
|
15 |
+
|
16 |
+
examples = [
|
17 |
+
[None, None, "Est-il possible de choisir son avocat ?"],
|
18 |
+
[None, None, "Les problèmes de divorce sont-ils couverts ?"],
|
19 |
+
[None, None, "Que se passe-t-il en cas de désaccord sur les mesures à prendre pour régler le litige ?"],
|
20 |
+
[None, None, "Comment résilier le contrat ?"]
|
21 |
+
]
|
22 |
+
|
23 |
+
dataset_names = ["zelros/pj-ca", "zelros/pj-ce", "zelros/pj-da",
|
24 |
+
"zelros/pj-groupama", "zelros/pj-sg", "zelros/pj-lbp"]
|
25 |
+
|
26 |
+
insurers = ["Crédit Agricole","Caisse d'Epargne","Direct Assurance","Groupama","Société Générale","La Banque Postale"]
|
27 |
+
|
28 |
+
db_dict = {}
|
29 |
+
|
30 |
+
for i, name in enumerate(dataset_names):
|
31 |
+
dataset = load_dataset(name)
|
32 |
+
|
33 |
+
df = dataset['train'].to_pandas()
|
34 |
+
df['text'] = df["title"] + df["content"]
|
35 |
+
|
36 |
+
loader = DataFrameLoader(df, 'text')
|
37 |
+
documents = loader.load()
|
38 |
+
text_splitter = CharacterTextSplitter(chunk_size=5000, chunk_overlap=0)
|
39 |
+
texts = text_splitter.split_documents(documents)
|
40 |
+
embeddings = OpenAIEmbeddings()
|
41 |
+
|
42 |
+
db = FAISS.from_documents(texts, embeddings)
|
43 |
+
db_dict[insurers[i]] = db
|
44 |
+
|
45 |
+
llm1 = ChatOpenAI(model_name="gpt-4")
|
46 |
+
llm2 = ChatOpenAI(model_name="gpt-4")
|
47 |
+
|
48 |
+
demo = gr.Interface(llm_response,
|
49 |
+
inputs=[
|
50 |
+
gr.Dropdown(choices=insurers,
|
51 |
+
label="Assureur 1",
|
52 |
+
value="Société Générale",
|
53 |
+
info="De nouveaux assureurs seront disponibles prochainement !"),
|
54 |
+
gr.Dropdown(choices=insurers,
|
55 |
+
label="Assureur 2",
|
56 |
+
value="La Banque Postale"),
|
57 |
+
gr.Textbox(label="Question",
|
58 |
+
info="Quelques exemples ci-dessous :)")
|
59 |
+
],
|
60 |
+
outputs=[gr.Textbox(label="Réponse assureur 1"),
|
61 |
+
gr.Textbox(label="Réponse assureur 2")
|
62 |
+
],
|
63 |
+
title='Pour une Assurance plus accessible et compréhensible',
|
64 |
+
description='### <center>Comparez les assurances protection juridique',
|
65 |
+
examples=examples)
|
66 |
+
|
67 |
+
demo.launch()
|