Update Qwen2-VL: now text+vision backbone quantized (fixed qkv + NaN bug)

Fix: show image upload for all VLMs (Gemma4 + Qwen2-VL), add Gemma4 examples

Show full traceback on error to diagnose Gemma4 failure

Fix: apply chat template in generate_text (fixes Qwen3 garbage output)

Fix: actually bundle updated inference.py with Hub download

Fix: add HF Hub download support to load_ternary_model

Upgrade to Gradio 5 (Python 3.13 compatible)

Fix Python 3.13 audioop: add pyaudioop

Bundle ternary_quant package directly (private repo fix)

Deploy ternary-quant Gradio demo

initial commit