main.py

from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
import os

app = FastAPI()

# Enable CORS for your React/Lovable frontend
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# Configuration
REPO_ID = "devray11/Aevis-Medical-SLM"
MODEL_FILENAME = "DeepSeek-R1-Distill-Llama-8B.Q4_K_M.gguf"

print(f"🚀 Initializing Aevis Medical SLM...")

# Download model from Hugging Face Hub
try:
    model_path = hf_hub_download(repo_id=REPO_ID, filename=MODEL_FILENAME)
    print(f"✅ Model downloaded to: {model_path}")
except Exception as e:
    print(f"❌ Error downloading model: {e}")
    model_path = None

# Initialize Model with optimized CPU settings
# n_ctx=1024 (Saves RAM), n_threads=2 (Matches HF Free Tier)
llm = Llama(
    model_path=model_path,
    n_ctx=1024,
    n_threads=2,
    n_batch=512,
    verbose=True
) if model_path else None

class ChatRequest(BaseModel):
    prompt: str

@app.post("/generate")
async def generate(request: ChatRequest):
    if not llm:
        return {"error": "Model not loaded properly"}
        
    # Standard instruction format for fine-tuned medical models
    prompt = f"### Instruction:\n{request.prompt}\n\n### Response:\n"
    
    response = llm(
        prompt,
        max_tokens=512,
        stop=["###", "</s>"],
        echo=False
    )
    
    return {"response": response["choices"][0]["text"].strip()}

@app.get("/")
def home():
    return {"message": "Aevis Medical SLM API is Online", "model": REPO_ID}