app.py

import os
import glob
from huggingface_hub import snapshot_download
from ctransformers import AutoModelForCausalLM
import gradio as gr

# Use home directory for cache (always writable in Spaces)
MODEL_CACHE_DIR = os.environ.get(
    "MODEL_DIR",
    os.path.join(os.path.expanduser("~"), "model_cache")
)
os.makedirs(MODEL_CACHE_DIR, exist_ok=True)

MODEL_REPO = "TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF"
MODEL_FILE = "tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"

# Ensure model is downloaded
def ensure_model():
    local_paths = glob.glob(os.path.join(MODEL_CACHE_DIR, "**", MODEL_FILE), recursive=True)
    if local_paths:
        return local_paths[0]
    print("Downloading model...")
    repo_path = snapshot_download(repo_id=MODEL_REPO, cache_dir=MODEL_CACHE_DIR)
    matches = glob.glob(os.path.join(repo_path, "**", MODEL_FILE), recursive=True)
    if matches:
        return matches[0]
    raise FileNotFoundError(f"{MODEL_FILE} not found in {MODEL_REPO}")

model_path = ensure_model()

# Load model
model = AutoModelForCausalLM.from_pretrained(
    MODEL_REPO,
    model_file=model_path,
    model_type="llama",
    max_new_tokens=256,
    temperature=0.7
)

# Custom responses
custom_data = {
    "hi": "hello Good morning, I am chatbot Deepika",
    "about you": "I am Deepika AI Chatbot, how may I help you?",
    "i love you": "I love you too",
    "age": "45"
}

def chatbot(msg):
    if msg in custom_data:
        return custom_data[msg]
    return model(f"### Instruction:\n{msg}\n\n### Response:\n")

if __name__ == "__main__":
    port = int(os.environ.get("PORT", 7860))
    gr.Interface(
        fn=chatbot,
        inputs="text",
        outputs="text",
        title="Custom Chatbot - TinyLlama"
    ).launch(server_name="0.0.0.0", server_port=port)