app.py

from fastapi import FastAPI
from pydantic import BaseModel
from llama_cpp import Llama

app = FastAPI()

# Download and initialize the model when the server starts
llm = Llama.from_pretrained(
    repo_id="Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF", # Make sure this says 0.5B!
    filename="*q4_k_m.gguf", 
    n_ctx=2048, 
    n_threads=2, 
    n_batch=512  
)

class EvalRequest(BaseModel):
    task_description: str
    python_code: str

# --- ADDED HEALTH CHECK ROUTE HERE ---
@app.get("/")
async def health_check():
    return {"status": "Online", "message": "AI Code Evaluator is running! Send POST requests to /evaluate"}
# -------------------------------------

@app.post("/evaluate")
async def evaluate_code(request: EvalRequest):
    prompt = f"Task Context:\n{request.task_description}\n\nStudent Code:\n{request.python_code}"
    
    response = llm.create_chat_completion(
        messages=[
            {
                "role": "system", 
                "content": "You are a friendly Python grader. Output ONLY valid JSON."
            },
            {"role": "user", "content": prompt}
        ],
        max_tokens=250, 
        temperature=0.1,
        # THIS IS THE MAGIC LINE:
        response_format={"type": "json_object"} 
    )
        
    return {"evaluation": response['choices'][0]['message']['content']}