server.py

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

app = FastAPI()

# ✅ Fix: Set writable cache directory for Hugging Face models
os.environ["TRANSFORMERS_CACHE"] = "/tmp"
os.environ["HF_HOME"] = "/tmp"

# ✅ Ensure cache directory exists
if not os.path.exists("/tmp"):
    os.makedirs("/tmp")

# ✅ Load DeepSeek-Coder-V2-Base Model with `trust_remote_code=True`
model_name = "deepseek-ai/DeepSeek-Coder-V2-Base"
tokenizer = AutoTokenizer.from_pretrained(model_name, cache_dir="/tmp", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto", cache_dir="/tmp", trust_remote_code=True)

class CodeRequest(BaseModel):
    user_story: str

@app.post("/generate-code")
def generate_code(request: CodeRequest):
    """Generates structured AI-powered code based on user story"""
    prompt = f"Generate structured code for: {request.user_story}"

    inputs = tokenizer(prompt, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
    output = model.generate(**inputs, max_length=300)
    generated_code = tokenizer.decode(output[0], skip_special_tokens=True)

    return {"generated_code": generated_code}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)