app.py

from fastapi import FastAPI
from pydantic import BaseModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

app = FastAPI()

# Charger le modèle et le tokenizer
#model_name = "mistralai/Mistral-7B-Instruct-v0.1"  # Modèle Mistral 7B
#model_name = "HuggingFaceH4/zephyr-3b"
#model_name = "serkanarslan/mistral-7b-mini-ft"
model_name = "microsoft/phi-2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"  # Utilise le GPU si dispo
)

# Définir le format des requêtes
class ChatRequest(BaseModel):
    message: str

@app.post("/chat")
async def chat(request: ChatRequest):
    inputs = tokenizer(request.message, return_tensors="pt").to("cuda")
    output = model.generate(**inputs, max_length=100)
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    return {"response": response}