model.py

import os
import requests
from llama_cpp import Llama

HF_TOKEN = os.getenv("HF_TOKEN")
MODEL_REPO = "TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF"
MODEL_FILENAME = "tinyllama-1.1b-chat-v1.0.Q2_K.gguf"
MODEL_PATH = f"./models/{MODEL_FILENAME}"

# Manual download with fallback
if not os.path.exists(MODEL_PATH):
    print("📦 Downloading GGUF model manually from Hugging Face...")

    url = f"https://huggingface.co/{MODEL_REPO}/resolve/main/{MODEL_FILENAME}"
    headers = {"Authorization": f"Bearer {HF_TOKEN}"}

    os.makedirs("./models", exist_ok=True)
    with requests.get(url, headers=headers, stream=True) as r:
        r.raise_for_status()
        with open(MODEL_PATH, "wb") as f:
            for chunk in r.iter_content(chunk_size=8192):
                f.write(chunk)

    print(f"✅ Model downloaded to {MODEL_PATH}")

# Load with llama-cpp
llm = Llama(
    model_path="./models/tinyllama-1.1b-chat-v1.0.Q2_K.gguf",
    n_ctx=128,          # Limit context for smaller RAM/CPU
    n_batch=32,
    n_threads=2,        # Use 2 threads (you can try 1 if needed)
    n_gpu_layers=0,     # CPU-only
    chat_format="llama-2"
)

def generate_structure(prompt: str) -> str:
    output = llm.create_completion(
        prompt=prompt,
        temperature=0.7,
        max_tokens=512,
    )
    return output["choices"][0]["text"].strip()