Upload Qwen2.5 0.5B GGUF model

Use faster Qwen CPU model

Use downloaded model cache path

Use Qwen2.5 1.5B as default model

Use direct system prompt

Format prompts for SmolLM chat

Install llama-cpp-python at build time

Fix Gradio interface compatibility and lazy model load

Upload models/SmolLM2-360M-Instruct-Q4_K_M.gguf with huggingface_hub

Avoid build-time llama-cpp-python install; install at runtime if missing

Set default public SmolLM GGUF repo for auto-download in Space

Add automatic GGUF model download support for Hugging Face Space

Add Hugging Face Space app and CPU-friendly GGUF setup

initial commit