main.py

from typing import Optional
from pydantic import BaseModel, Field
from fastapi import FastAPI, HTTPException
from contextlib import asynccontextmanager
from starlette.responses import FileResponse

class ConvertRequest(BaseModel):
    hf_model_name: str
    hf_token: Optional[str] = Field(None, description="Hugging Face token for private models")
    hf_push_repo: Optional[str] = Field(None, description="Hugging Face repo to push the converted model")

@asynccontextmanager
async def lifespan(app:FastAPI):    
    yield

app = FastAPI(title="Huggingface Safetensor Model Converter to AWQ", version="0.1.0", lifespan=lifespan)

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

@app.post("/convert")
def convert(request: ConvertRequest)->FileResponse:
    raise HTTPException(status_code=501, detail="Not Implemented yet")
    
    #return FileResponse(file_location, media_type='application/octet-stream',filename=file_name)