app.py

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

# ✅ Modules de LlamaIndex
from llama_index.core.settings import Settings
from llama_index.core import Document
from llama_index.llms.llama_cpp import LlamaCPP
from llama_index.core.node_parser import SemanticSplitterNodeParser

# ✅ Pour l'embedding LOCAL via transformers
from transformers import AutoTokenizer, AutoModel
import torch
import torch.nn.functional as F
import os

app = FastAPI()

# ✅ Configuration locale du cache HF pour Hugging Face
CACHE_DIR = "/data"
os.environ["HF_HOME"] = CACHE_DIR
os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
os.environ["HF_MODULES_CACHE"] = CACHE_DIR
os.environ["HF_HUB_CACHE"] = CACHE_DIR

# ✅ Configuration du modèle d’embedding local (ex: BGE / Nomic / GTE etc.)
MODEL_NAME = "BAAI/bge-small-en-v1.5"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, cache_dir=CACHE_DIR)
model = AutoModel.from_pretrained(MODEL_NAME, cache_dir=CACHE_DIR)

def get_embedding(text: str):
    with torch.no_grad():
        inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        outputs = model(**inputs)
        embeddings = outputs.last_hidden_state[:, 0]
        return F.normalize(embeddings, p=2, dim=1).squeeze().tolist()

# ✅ Données entrantes du POST
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):
    try:
        # ✅ Chargement du modèle LLM depuis Hugging Face en ligne (pas de .gguf local)
        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},
        )

        # ✅ Intégration manuelle de l'embedding local dans Settings
        class SimpleEmbedding:
            def get_text_embedding(self, text: str):
                return get_embedding(text)

        Settings.llm = llm
        Settings.embed_model = SimpleEmbedding()

        # ✅ Découpage sémantique intelligent
        parser = SemanticSplitterNodeParser.from_defaults()
        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)}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run("app:app", host="0.0.0.0", port=7860)