# app.py
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

# Modèle HF Flan-T5
MODEL_NAME = "google/flan-t5-base"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
model.eval()

app = FastAPI(title="Flan-T5 Service")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

class ImproveRequest(BaseModel):
    text: str

@app.get("/")
def health():
    return {"status": "ok", "model": MODEL_NAME}

@app.post("/improve")
def improve_text(req: ImproveRequest):
    try:
        inputs = tokenizer(req.text, return_tensors="pt")
        with torch.no_grad():
            outputs = model.generate(**inputs, max_new_tokens=512)
        improved = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return JSONResponse({"success": True, "improved_text": improved})
    except Exception as e:
        return JSONResponse({"success": False, "error": str(e)}, status_code=500)