app.py

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


app = FastAPI()

class ChatRequest(BaseModel):
    message: str

@app.get("/")
def root():
    
    # Load DeepSeek model (small one for local use)
    # Try bigger models if you have a GPU with >12GB VRAM
    model_name = "deepseek-ai/deepseek-coder-1.3b-instruct"

    print("Loading model... this may take a minute ⏳")
    global tokenizer
    global model
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
        device_map="auto"
    )
    print("Model loaded ✅")

    return {"status": "ok"}

@app.post("/chat")
def chat(request: ChatRequest):
    """Chat endpoint using DeepSeek model"""
    inputs = tokenizer(request.message, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=200)
    reply = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return {"reply": reply}