Update app.py

app.py

@@ -7,11 +7,9 @@ from llama_index.llms.llama_cpp import LlamaCPP
 from llama_index.embeddings.huggingface import HuggingFaceEmbedding
 from llama_index.core.node_parser import SemanticSplitterNodeParser
 
-import os
-
 app = FastAPI()
 
-# 🔹 Schéma d'entrée
+# 📥 Modèle de la requête JSON envoyée à /chunk
 class ChunkRequest(BaseModel):
     text: str
     source_id: Optional[str] = None
@@ -19,27 +17,29 @@ class ChunkRequest(BaseModel):
     source: Optional[str] = None
     type: Optional[str] = None
 
-# 🔹 Endpoint principal
-
 @app.post("/chunk")
 async def chunk_text(data: ChunkRequest):
+    # ✅ Chargement direct d’un modèle hébergé sur Hugging Face (pas de fichier local .gguf)
     llm = LlamaCPP(
-        model_path="/models/mistral-7b-instruct.gguf",
+        model_url="https://huggingface.co/leafspark/Mistral-7B-Instruct-v0.2-Q4_K_M-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
         temperature=0.1,
         max_new_tokens=512,
         context_window=2048,
         generate_kwargs={"top_p": 0.95},
-        model_kwargs={"n_gpu_layers": 1},
+        model_kwargs={"n_gpu_layers": 1},  # Laisse 1 si CPU
     )
 
+    # ✅ Embedding open-source via Hugging Face
     embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-small-en-v1.5")
 
+    # ✅ Configuration du service IA
     service_context = ServiceContext.from_defaults(
         llm=llm,
         embed_model=embed_model
     )
 
     try:
+        # ✅ Découpage sémantique intelligent
         parser = SemanticSplitterNodeParser.from_defaults(service_context=service_context)
         nodes = parser.get_nodes_from_documents([Document(text=data.text)])
 
@@ -53,8 +53,3 @@ async def chunk_text(data: ChunkRequest):
         }
     except Exception as e:
         return {"error": str(e)}
-
-
-
-
-