app.py

from fastapi import FastAPI, Request
from huggingface_hub import InferenceClient
import os

app = FastAPI()

HF_TOKEN = os.getenv("HF_TOKEN")  # đọc token từ Secrets

# client = InferenceClient(token=HF_TOKEN, model="google/gemma-3-12b-it", provider="featherless-ai")
client = InferenceClient(token=HF_TOKEN, model="meta-llama/Llama-3.2-3B-Instruct")
# client = InferenceClient(token=HF_TOKEN, model="Qwen/Qwen3-4B-Instruct-2507")
# client = InferenceClient(token=HF_TOKEN, model="swiss-ai/Apertus-70B-Instruct-2509")


@app.get("/")
def root():
    return {"message": "Gemma 3 API on CPU"}


@app.post("/generate")
async def generate(request: Request):
    body = await request.json()
    prompt = body.get("prompt", "")
    messages = [
        {"role": "user", "content": prompt}
    ]
    out = client.chat_completion(messages)
    print(out)
    response = out.choices[0].message.content
    return {"response": response}