app.py

from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
    BitsAndBytesConfig,
    TextIteratorStreamer
)
import torch
import threading

app = FastAPI()

MODEL_NAME = "Qwen/Qwen2.5-Coder-7B"

# ---- Quantization config (CPU safe) ----
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float32,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4"
)

tokenizer = AutoTokenizer.from_pretrained(
    MODEL_NAME,
    trust_remote_code=True
)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    device_map="cpu",
    quantization_config=bnb_config,
    trust_remote_code=True
)

class Prompt(BaseModel):
    message: str

# -------------------------------------------------
# ✅ NORMAL CHAT (UNCHANGED)
# -------------------------------------------------
@app.post("/chat")
def chat(prompt: Prompt):
    inputs = tokenizer(prompt.message, return_tensors="pt")

    outputs = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.7,
        do_sample=True
    )

    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return {"response": response}


# -------------------------------------------------
# 🚀 STREAMING CHAT (CHATGPT-LIKE)
# -------------------------------------------------
@app.post("/chat-stream")
def chat_stream(prompt: Prompt):

    inputs = tokenizer(prompt.message, return_tensors="pt")

    streamer = TextIteratorStreamer(
        tokenizer,
        skip_special_tokens=True,
        skip_prompt=True
    )

    generation_kwargs = dict(
        **inputs,
        streamer=streamer,
        max_new_tokens=200,
        temperature=0.7,
        do_sample=True
    )

    # Run generation in background thread
    thread = threading.Thread(
        target=model.generate,
        kwargs=generation_kwargs
    )
    thread.start()

    def token_generator():
        for token in streamer:
            yield token

    return StreamingResponse(
        token_generator(),
        media_type="text/plain"
    )