app.py

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

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/deepseek-vl-1.3b-chat")
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/deepseek-vl-1.3b-chat")

# Initialize FastAPI app
app = FastAPI()

# Define input schema
class RequestBody(BaseModel):
    prompt: str

# Define the model inference function
@app.post("/predict")
async def predict(request: RequestBody):
    inputs = tokenizer(request.prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=100)
    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return {"response": result}

# For testing locally (not needed for Hugging Face)
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)