app.py

from fastapi import FastAPI
from pydantic import BaseModel
from typing import Optional

from llama_index.core import Document, ServiceContext
from llama_index.llms.llama_cpp import LlamaCPP
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core.node_parser import SemanticSplitterNodeParser

app = FastAPI()

# 📥 Modèle de la requête JSON envoyée à /chunk
class ChunkRequest(BaseModel):
    text: str
    source_id: Optional[str] = None
    titre: Optional[str] = None
    source: Optional[str] = None
    type: Optional[str] = None

@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_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},  # 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)])

        return {
            "chunks": [node.text for node in nodes],
            "metadatas": [node.metadata for node in nodes],
            "source_id": data.source_id,
            "titre": data.titre,
            "source": data.source,
            "type": data.type
        }
    except Exception as e:
        return {"error": str(e)}