Refactor: Tách code thành cấu trúc module chuyên nghiệp

Add audio autoplay event handler and fix layout issues

Fix CSS syntax error and add JavaScript for audio autoplay

Fix UI layout issues and ensure audio autoplay works correctly

Professional UI redesign: modern glassmorphism, improved typography, animations, better spacing

Improve Colab notebook: add environment check, better error handling, troubleshooting guide

Fix STT and TTS: improve error handling, validate audio files, fix return values

Redesign UI: Modern chatbot conversation interface with bubbles, avatars, and better styling

Add processing time display, improve STT accuracy, enhance TTS

Add Talk-to-Talk mode: automatic voice conversation loop

Final TTS improvements: better error handling

Improve TTS: add error handling, ensure auto-read works

Enable auto TTS: AI automatically reads all responses out loud

Fix: use full 40GB GPU RAM, optimize memory allocation

Optimize for Colab: disable torch.compile on T4, simplify generation, use full 40GB GPU RAM

Update Colab notebook: add faster-whisper, flash-attn, optimum dependencies

Add SDPA to load_kwargs for faster attention

Ultra-fast LLM optimization: SDPA, torch.compile reduce-overhead, pre-warm, half precision inputs, inference_mode

Upgrade to faster-whisper + optimize for max speed: torch.compile, flash attention, full GPU utilization

Optimize for full performance mode - remove all limits, max speed STT/LLM

Update print message to show correct model name

Update to use bank-1.7 model (Qwen3-VLTO-1.7B-Instruct)

Fix Qwen3VL generation config to prevent gibberish output - add repetition penalty and better temperature control

Fix global variables issue with @GPU decorator - separate internal load function

Add @GPU decorator for Hugging Face Spaces and fix torch_dtype deprecation

Fix Qwen3VL model loading - use Qwen3VLForConditionalGeneration

Update app to use bank-model-2b with voice chat features

initial commit

Duplicate from gradio-templates/chatbot