g

model.py

@@ -1,3 +1,4 @@
+# model.py actualizado
 from langchain.chains import RetrievalQA
 from langchain.vectorstores import Chroma
 from langchain.embeddings import HuggingFaceEmbeddings
@@ -7,70 +8,78 @@ from huggingface_hub import hf_hub_download
 import os
 import shutil
 
-# Configuración esencial para Spaces
+# 1. Configuración esencial
 HF_TOKEN = os.getenv("HF_TOKEN")
 MISTRAL_API_KEY = os.getenv("MISTRAL_API_KEY")
+CHROMA_DIR = "/home/user/app/chroma_db"
 
-# 1. Configurar rutas específicas para Spaces
-CHROMA_DIR = "/home/user/app/chroma_db"  # Ruta dentro del espacio persistente
-os.makedirs(CHROMA_DIR, exist_ok=True)
 
+# 2. Limpieza inicial radical
+def full_clean():
+    # Eliminar todo rastro previo
+    shutil.rmtree(CHROMA_DIR, ignore_errors=True)
+    shutil.rmtree("/home/user/.cache/huggingface/hub", ignore_errors=True)
+    os.makedirs(CHROMA_DIR, exist_ok=True)
 
-# 2. Limpieza inicial de conflictos
-def clean_space():
-    paths_to_clean = [
-        "/home/user/.cache/huggingface/hub/datasets--VictorCarr02--Conversational-Agent-LawsEC",
-        CHROMA_DIR
-    ]
 
-    for path in paths_to_clean:
-        if os.path.exists(path):
-            shutil.rmtree(path, ignore_errors=True)
-    os.makedirs(CHROMA_DIR, exist_ok=True)
+full_clean()
 
 
-clean_space()
+# 3. Descargar y configurar ChromaDB
+def setup_chroma():
+    # Descargar archivo original
+    chroma_src = hf_hub_download(
+        repo_id="VictorCarr02/Conversational-Agent-LawsEC",
+        repo_type="dataset",
+        filename="chroma.sqlite3",
+        token=HF_TOKEN
+    )
 
-# 3. Descargar y mover chroma.sqlite3
-chroma_source = hf_hub_download(
-    repo_id="VictorCarr02/Conversational-Agent-LawsEC",
-    repo_type="dataset",
-    filename="chroma.sqlite3",
-    token=HF_TOKEN,
-    force_download=True
-)
+    # Configurar estructura requerida por Chroma
+    tenant_dir = os.path.join(CHROMA_DIR, "chroma.sqlite3")
+    os.makedirs(os.path.dirname(tenant_dir), exist_ok=True)
+    shutil.copy(chroma_src, tenant_dir)
 
-# Mover al directorio controlado
-shutil.move(chroma_source, os.path.join(CHROMA_DIR, "chroma.sqlite3"))
 
-# 4. Inicializar ChromaDB
-chroma_client = chromadb.PersistentClient(path=CHROMA_DIR)
-collection = chroma_client.get_or_create_collection("legal_docs")
+setup_chroma()
+
+# 4. Conexión explícita a ChromaDB
+chroma_client = chromadb.PersistentClient(
+    path=CHROMA_DIR,
+    tenant="default_tenant",
+    database="default_database"
+)
+
+# 5. Verificar/crear collection
+try:
+    collection = chroma_client.get_collection("legal_docs")
+except ValueError:
+    collection = chroma_client.create_collection("legal_docs")
 
-# 5. Configurar embeddings (optimizado para Spaces)
+# 6. Configurar embeddings
 embeddings = HuggingFaceEmbeddings(
     model_name="sentence-transformers/all-mpnet-base-v2",
     model_kwargs={"device": "cpu"}
 )
 
-# 6. Crear vector store
+# 7. Inicializar Chroma LangChain
 vector_store = Chroma(
     client=chroma_client,
     collection_name="legal_docs",
     embedding_function=embeddings
 )
 
-# 7. Configurar Mistral
+# 8. Configurar Mistral
 llm = ChatMistralAI(
     api_key=MISTRAL_API_KEY,
     model="mistral-large-latest",
     temperature=0.1
 )
 
-# 8. Cadena RAG final
+# 9. Cadena RAG final
 rag_chain = RetrievalQA.from_chain_type(
     llm=llm,
-    retriever=vector_store.as_retriever(search_kwargs={"k": 2}),
+    retriever=vector_store.as_retriever(search_kwargs={"k": 3}),
     chain_type="stuff",
     return_source_documents=True
 )

requirements.txt

@@ -3,4 +3,5 @@ langchain
 chromadb
 huggingface_hub
 langchain_mistralai
-langchain-community
+langchain-community
+sentence-transformers