Spaces:
Running
Running
Commit ·
e4d59e7
1
Parent(s): 29b0b66
interface updates
Browse files- .env +5 -2
- __pycache__/config.cpython-310.pyc +0 -0
- __pycache__/pdf_processing.cpython-310.pyc +0 -0
- __pycache__/pinecone_utilsB.cpython-310.pyc +0 -0
- app.py +24 -11
- config.py +10 -1
- index_documents.py +0 -1
- initIndex.py +0 -28
- neo4j_initialize.py +102 -0
- pinecone_utilsB.py +0 -8
.env
CHANGED
|
@@ -4,6 +4,9 @@ LANGSMITH_TRACING=true
|
|
| 4 |
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
|
| 5 |
LANGSMITH_API_KEY="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"
|
| 6 |
LANGSMITH_PROJECT="rag_architecture"
|
| 7 |
-
#OPENAI_API_KEY="<your-openai-api-key>"
|
| 8 |
|
| 9 |
-
PINECONE_API_KEY="pcsk_4cofG5_Uk93QCMSKiPvf7btHrPtuhvK71HmcSwfp5g3hHMZTWfapyjs8tvDCYcQteB51Z"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
|
| 5 |
LANGSMITH_API_KEY="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"
|
| 6 |
LANGSMITH_PROJECT="rag_architecture"
|
|
|
|
| 7 |
|
| 8 |
+
PINECONE_API_KEY="pcsk_4cofG5_Uk93QCMSKiPvf7btHrPtuhvK71HmcSwfp5g3hHMZTWfapyjs8tvDCYcQteB51Z"
|
| 9 |
+
|
| 10 |
+
NEO4J_URI="neo4j+s://e50baf05.databases.neo4j.io"
|
| 11 |
+
NEO4J_USERNAME="neo4j"
|
| 12 |
+
NEO4J_PASSWORD="uu6scz4Hf9SwY6SlJgHxk58SHv1m3YNz_RwxAYQKaJc"
|
__pycache__/config.cpython-310.pyc
CHANGED
|
Binary files a/__pycache__/config.cpython-310.pyc and b/__pycache__/config.cpython-310.pyc differ
|
|
|
__pycache__/pdf_processing.cpython-310.pyc
CHANGED
|
Binary files a/__pycache__/pdf_processing.cpython-310.pyc and b/__pycache__/pdf_processing.cpython-310.pyc differ
|
|
|
__pycache__/pinecone_utilsB.cpython-310.pyc
CHANGED
|
Binary files a/__pycache__/pinecone_utilsB.cpython-310.pyc and b/__pycache__/pinecone_utilsB.cpython-310.pyc differ
|
|
|
app.py
CHANGED
|
@@ -1,10 +1,8 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
-
from pdf_processing import get_existing_pdf, load_and_preprocess_pdf, split_text
|
| 3 |
from graph_agentA import agent as agent_A
|
| 4 |
from graph_agentB import agent as agent_B
|
| 5 |
from config import *
|
| 6 |
from dotenv import load_dotenv
|
| 7 |
-
from initIndex import *
|
| 8 |
from pinecone_utilsB import *
|
| 9 |
|
| 10 |
|
|
@@ -76,9 +74,13 @@ def process_query_B(query):
|
|
| 76 |
def display_sidebar():
|
| 77 |
"""Affiche la barre latérale."""
|
| 78 |
with st.sidebar:
|
| 79 |
-
|
| 80 |
-
st.
|
| 81 |
-
st.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 82 |
|
| 83 |
def display_chat_history():
|
| 84 |
"""Affiche l'historique de chat."""
|
|
@@ -96,20 +98,31 @@ def main():
|
|
| 96 |
if not check_indexes_ready():
|
| 97 |
return
|
| 98 |
|
| 99 |
-
st.title("
|
| 100 |
-
|
| 101 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
display_sidebar()
|
| 103 |
display_chat_history()
|
| 104 |
|
| 105 |
query = st.chat_input("Posez votre question ici:")
|
| 106 |
if query:
|
| 107 |
-
if
|
| 108 |
process_query_B(query)
|
| 109 |
else:
|
| 110 |
process_query_A(query)
|
| 111 |
st.rerun()
|
| 112 |
|
| 113 |
if __name__ == "__main__":
|
| 114 |
-
main()
|
| 115 |
-
|
|
|
|
| 1 |
import streamlit as st
|
|
|
|
| 2 |
from graph_agentA import agent as agent_A
|
| 3 |
from graph_agentB import agent as agent_B
|
| 4 |
from config import *
|
| 5 |
from dotenv import load_dotenv
|
|
|
|
| 6 |
from pinecone_utilsB import *
|
| 7 |
|
| 8 |
|
|
|
|
| 74 |
def display_sidebar():
|
| 75 |
"""Affiche la barre latérale."""
|
| 76 |
with st.sidebar:
|
| 77 |
+
lien_ressource = "https://www.fnac.com/livre-numerique/a21290809/Gaspard-Boreal-La-Confession-muette"
|
| 78 |
+
#st.title("📄 La confession muette")
|
| 79 |
+
#st.write("Posez vos questions sur le document.")
|
| 80 |
+
st.image(agent_B.get_graph().draw_mermaid_png(), caption="Workflow Graph")
|
| 81 |
+
st.markdown("Document de référence 📄 : \nLa confession muette ()2025")
|
| 82 |
+
st.markdown("Avec l'aimable autorisation de Gaspard Boréal: \n[Récit d'origine]({})".format(lien_ressource))
|
| 83 |
+
|
| 84 |
|
| 85 |
def display_chat_history():
|
| 86 |
"""Affiche l'historique de chat."""
|
|
|
|
| 98 |
if not check_indexes_ready():
|
| 99 |
return
|
| 100 |
|
| 101 |
+
st.title("RAG architectures")
|
| 102 |
+
st.markdown(
|
| 103 |
+
"""
|
| 104 |
+
<style>
|
| 105 |
+
div[data-testid="stSelectbox"] {
|
| 106 |
+
width: 200px !important;
|
| 107 |
+
}
|
| 108 |
+
</style>
|
| 109 |
+
""",
|
| 110 |
+
unsafe_allow_html=True
|
| 111 |
+
)
|
| 112 |
+
architecture = st.selectbox(
|
| 113 |
+
"Sélectionnez une architecture :",
|
| 114 |
+
["Basic", "Intermédiaire", "Avancée"]
|
| 115 |
+
)
|
| 116 |
display_sidebar()
|
| 117 |
display_chat_history()
|
| 118 |
|
| 119 |
query = st.chat_input("Posez votre question ici:")
|
| 120 |
if query:
|
| 121 |
+
if architecture == "Intermédiaire":
|
| 122 |
process_query_B(query)
|
| 123 |
else:
|
| 124 |
process_query_A(query)
|
| 125 |
st.rerun()
|
| 126 |
|
| 127 |
if __name__ == "__main__":
|
| 128 |
+
main()
|
|
|
config.py
CHANGED
|
@@ -4,7 +4,7 @@ from dotenv import load_dotenv
|
|
| 4 |
from pinecone import Pinecone, ServerlessSpec, Index
|
| 5 |
from langsmith import Client
|
| 6 |
from langchain_mistralai.chat_models import ChatMistralAI
|
| 7 |
-
from
|
| 8 |
|
| 9 |
# Charger les variables d'environnement
|
| 10 |
load_dotenv()
|
|
@@ -30,6 +30,15 @@ client = Client(
|
|
| 30 |
api_key=langsmith_api_key,
|
| 31 |
)
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
# Mistral AI configuration
|
| 34 |
mistral_api_key = os.getenv("MISTRAL_API_KEY")
|
| 35 |
llm = ChatMistralAI(
|
|
|
|
| 4 |
from pinecone import Pinecone, ServerlessSpec, Index
|
| 5 |
from langsmith import Client
|
| 6 |
from langchain_mistralai.chat_models import ChatMistralAI
|
| 7 |
+
from neo4j import GraphDatabase
|
| 8 |
|
| 9 |
# Charger les variables d'environnement
|
| 10 |
load_dotenv()
|
|
|
|
| 30 |
api_key=langsmith_api_key,
|
| 31 |
)
|
| 32 |
|
| 33 |
+
# Initialize Neo4j connection
|
| 34 |
+
neo4j_uri = os.getenv("NEO4J_URI")
|
| 35 |
+
neo4j_username = os.getenv("NEO4J_USERNAME")
|
| 36 |
+
neo4j_password = os.getenv("NEO4J_PASSWORD")
|
| 37 |
+
neo4j_driver = GraphDatabase.driver(
|
| 38 |
+
neo4j_uri,
|
| 39 |
+
auth=(neo4j_username, neo4j_password)
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
# Mistral AI configuration
|
| 43 |
mistral_api_key = os.getenv("MISTRAL_API_KEY")
|
| 44 |
llm = ChatMistralAI(
|
index_documents.py
CHANGED
|
@@ -1,5 +1,4 @@
|
|
| 1 |
# index_documents.py
|
| 2 |
-
from config import sparse_index, dense_index
|
| 3 |
from pinecone_utilsA import index_pdf as index_pdf_A
|
| 4 |
from pinecone_utilsB import *
|
| 5 |
from pdf_processing import get_existing_pdf, load_and_preprocess_pdf, split_text
|
|
|
|
| 1 |
# index_documents.py
|
|
|
|
| 2 |
from pinecone_utilsA import index_pdf as index_pdf_A
|
| 3 |
from pinecone_utilsB import *
|
| 4 |
from pdf_processing import get_existing_pdf, load_and_preprocess_pdf, split_text
|
initIndex.py
DELETED
|
@@ -1,28 +0,0 @@
|
|
| 1 |
-
from config import *
|
| 2 |
-
import os
|
| 3 |
-
from dotenv import load_dotenv
|
| 4 |
-
from pinecone import Pinecone, ServerlessSpec
|
| 5 |
-
from pdf_processing import get_existing_pdf, load_and_preprocess_pdf, split_text
|
| 6 |
-
from pinecone_utilsA import index_pdf as index_pdf_A
|
| 7 |
-
from pinecone_utilsB import *
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
#pdf_path = get_existing_pdf()
|
| 11 |
-
|
| 12 |
-
def index_pdf(pdf_path, use_architecture_B=False):
|
| 13 |
-
"""Indexe un PDF dans Pinecone."""
|
| 14 |
-
if not pdf_path:
|
| 15 |
-
print("Aucun fichier PDF trouvé.")
|
| 16 |
-
return
|
| 17 |
-
|
| 18 |
-
text = load_and_preprocess_pdf(pdf_path)
|
| 19 |
-
texts = split_text(text)
|
| 20 |
-
|
| 21 |
-
if use_architecture_B:
|
| 22 |
-
print("Indexation pour l'architecture B en cours...")
|
| 23 |
-
index_pdf_B(texts)
|
| 24 |
-
print("Indexation pour l'architecture B terminée.")
|
| 25 |
-
else:
|
| 26 |
-
print("Indexation pour l'architecture A en cours...")
|
| 27 |
-
index_pdf_A(texts)
|
| 28 |
-
print("Indexation pour l'architecture A terminée.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
neo4j_initialize.py
ADDED
|
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from langchain_community.graphs import Neo4jGraph
|
| 2 |
+
from config import *
|
| 3 |
+
from neo4j_graphrag.experimental.components.text_splitters.fixed_size_splitter import FixedSizeSplitter
|
| 4 |
+
from neo4j_graphrag.experimental.pipeline.kg_builder import SimpleKGPipeline
|
| 5 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
| 6 |
+
from pdf_processing import get_existing_pdf, load_and_preprocess_pdf, split_text
|
| 7 |
+
from neo4j_graphrag.experimental.pipeline.kg_builder import SimpleKGPipeline
|
| 8 |
+
from neo4j import GraphDatabase
|
| 9 |
+
|
| 10 |
+
try:
|
| 11 |
+
with neo4j_driver.session() as session:
|
| 12 |
+
result = session.run("RETURN 1 AS test")
|
| 13 |
+
print("Connexion à Neo4j réussie :", result.single()["test"])
|
| 14 |
+
except Exception as e:
|
| 15 |
+
print("Erreur de connexion à Neo4j :", e)
|
| 16 |
+
|
| 17 |
+
graph = Neo4jGraph()
|
| 18 |
+
|
| 19 |
+
def generate_graph():
|
| 20 |
+
|
| 21 |
+
basic_node_labels = ["Personnage", "Objet", "Lieu", "Événement", "PériodeTemporelle"]
|
| 22 |
+
story_node_labels = ["Protagoniste", "Antagoniste", "PersonnageSecondaire", "CréatureMythique",
|
| 23 |
+
"FigureHistorique", "Narrateur"]
|
| 24 |
+
literary_node_labels = ["Thème", "Motif", "Symbole"]
|
| 25 |
+
|
| 26 |
+
node_labels = basic_node_labels + story_node_labels + literary_node_labels
|
| 27 |
+
|
| 28 |
+
rel_types = ["CONNAÎT", "SITUE_DANS", "FAIT_PARTIE_DE", "SE_PRODUIT_PENDANT", "IMPLIQUE",
|
| 29 |
+
"S'OPPOSE_À", "CRÉÉ_PAR", "INSPIRÉ_PAR", "REPRÉSENTE", "TRANSFORME"]
|
| 30 |
+
|
| 31 |
+
embeddings = HuggingFaceEmbeddings(model_name="intfloat/multilingual-e5-large")
|
| 32 |
+
|
| 33 |
+
prompt_template = '''
|
| 34 |
+
Vous êtes un expert en analyse de texte chargé d'extraire des informations à partir d'un récit et de les structurer sous forme de graphe de propriétés pour faciliter la compréhension et l'analyse du texte.
|
| 35 |
+
|
| 36 |
+
Extrayez les entités (nœuds) et spécifiez leur type à partir du texte d'entrée suivant.
|
| 37 |
+
Extrayez également les relations entre ces nœuds. La direction de la relation va du nœud de départ au nœud d'arrivée.
|
| 38 |
+
|
| 39 |
+
Retournez le résultat au format JSON en utilisant le modèle suivant :
|
| 40 |
+
{{
|
| 41 |
+
"nodes": [
|
| 42 |
+
{{
|
| 43 |
+
"id": "0",
|
| 44 |
+
"label": "type d'entité",
|
| 45 |
+
"properties": {{
|
| 46 |
+
"name": "nom de l'entité"
|
| 47 |
+
}}
|
| 48 |
+
}}
|
| 49 |
+
],
|
| 50 |
+
"relationships": [
|
| 51 |
+
{{
|
| 52 |
+
"type": "TYPE_DE_RELATION",
|
| 53 |
+
"start_node_id": "0",
|
| 54 |
+
"end_node_id": "1",
|
| 55 |
+
"properties": {{
|
| 56 |
+
"details": "Description de la relation"
|
| 57 |
+
}}
|
| 58 |
+
}}
|
| 59 |
+
]
|
| 60 |
+
}}
|
| 61 |
+
|
| 62 |
+
- Utilisez uniquement les informations du texte d'entrée. N'ajoutez aucune information supplémentaire.
|
| 63 |
+
- Si le texte d'entrée est vide, retournez un JSON vide.
|
| 64 |
+
- Créez autant de nœuds et de relations que nécessaire pour offrir un contexte riche et détaillé.
|
| 65 |
+
- Un assistant de connaissance basé sur l'IA doit pouvoir lire ce graphe et comprendre immédiatement le contexte pour poser des questions détaillées.
|
| 66 |
+
Utilisez uniquement les nœuds et relations suivants (s'ils sont fournis) :
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
Attribuez un identifiant unique (chaîne de caractères) à chaque nœud et réutilisez-le pour définir les relations.
|
| 70 |
+
Respectez les types de nœuds source et cible pour les relations, ainsi que la direction des relations.
|
| 71 |
+
|
| 72 |
+
Ne retournez aucune information supplémentaire autre que le JSON.
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
Texte d'entrée :
|
| 77 |
+
|
| 78 |
+
'''
|
| 79 |
+
|
| 80 |
+
kg_builder_pdf = SimpleKGPipeline(
|
| 81 |
+
llm=llm,
|
| 82 |
+
driver=neo4j_driver,
|
| 83 |
+
text_splitter=FixedSizeSplitter(chunk_size=1024, chunk_overlap=200),
|
| 84 |
+
embedder=embeddings,
|
| 85 |
+
entities=node_labels,
|
| 86 |
+
relations=rel_types,
|
| 87 |
+
prompt_template=prompt_template,
|
| 88 |
+
from_pdf=True
|
| 89 |
+
)
|
| 90 |
+
|
| 91 |
+
# Charger et prétraiter les PDF
|
| 92 |
+
pdf_files = get_existing_pdf()
|
| 93 |
+
texts = []
|
| 94 |
+
for pdf_file in pdf_files:
|
| 95 |
+
text = load_and_preprocess_pdf(pdf_file)
|
| 96 |
+
texts.extend(split_text(text))
|
| 97 |
+
|
| 98 |
+
results = kg_builder_pdf.run_async(texts)
|
| 99 |
+
graph.add_graph_documents(results)
|
| 100 |
+
|
| 101 |
+
if __name__ == "__main__":
|
| 102 |
+
generate_graph()
|
pinecone_utilsB.py
CHANGED
|
@@ -1,17 +1,12 @@
|
|
| 1 |
from sentence_transformers import SentenceTransformer
|
| 2 |
from pinecone_text.sparse import BM25Encoder
|
| 3 |
-
from langchain.retrievers import PineconeHybridSearchRetriever
|
| 4 |
from langchain.embeddings import HuggingFaceEmbeddings
|
| 5 |
-
from langchain_pinecone import PineconeVectorStore
|
| 6 |
-
from langchain.schema import Document
|
| 7 |
-
import pinecone
|
| 8 |
import streamlit as st
|
| 9 |
from config import sparse_index as indexB
|
| 10 |
import nltk
|
| 11 |
import zlib
|
| 12 |
import base64
|
| 13 |
import json
|
| 14 |
-
import os
|
| 15 |
import hashlib
|
| 16 |
import uuid
|
| 17 |
|
|
@@ -120,9 +115,6 @@ def index_pdf_B(texts):
|
|
| 120 |
|
| 121 |
def hybrid_search(query):
|
| 122 |
"""Récupère les documents pertinents en combinant les résultats de Pinecone et BM25."""
|
| 123 |
-
#if not is_initialized():
|
| 124 |
-
#st.warning("L'index BM25 n'est pas encore prêt. Veuillez patienter pendant l'indexation...")
|
| 125 |
-
#return []
|
| 126 |
|
| 127 |
try:
|
| 128 |
# Générer le vecteur dense pour la requête
|
|
|
|
| 1 |
from sentence_transformers import SentenceTransformer
|
| 2 |
from pinecone_text.sparse import BM25Encoder
|
|
|
|
| 3 |
from langchain.embeddings import HuggingFaceEmbeddings
|
|
|
|
|
|
|
|
|
|
| 4 |
import streamlit as st
|
| 5 |
from config import sparse_index as indexB
|
| 6 |
import nltk
|
| 7 |
import zlib
|
| 8 |
import base64
|
| 9 |
import json
|
|
|
|
| 10 |
import hashlib
|
| 11 |
import uuid
|
| 12 |
|
|
|
|
| 115 |
|
| 116 |
def hybrid_search(query):
|
| 117 |
"""Récupère les documents pertinents en combinant les résultats de Pinecone et BM25."""
|
|
|
|
|
|
|
|
|
|
| 118 |
|
| 119 |
try:
|
| 120 |
# Générer le vecteur dense pour la requête
|