Clean slate - fresh start

Trigger rebuild: update description

Fix: replace broken demo fallback, add catch-all error handling, move model back to CPU after GPU inference

Load model at deploy time on CPU, move to GPU on generate

Fix: remove broken chat template, add demo fallback, wrap model load in try-except

Increase GPU duration to 300s, fix case-sensitive DOCTYPE check

User changes

Remove build.md from repo (local only)

Add build.md project documentation

Remove all fallbacks - AI only, show error if model fails

Redesign: left preview, right sidebar with lang dropdown + copy, chat box at bottom, AI generates all 5 formats

Re-enable model with @spaces.GPU for ZeroGPU support

Skip model (CPU too slow), use instant built-in generator for fast results

Fix: no retry (saves 4min), better fallback content per component type, fix color picker via iframe DOM access, fix syntax

Fix: use MiniCPM5-1B, fix prompts to generate real code, validate output has actual HTML/React code

Switch to Qwen2.5-1.5B-Instruct for better HF Spaces compatibility

Add demo mode fallback, detailed error handling, real login form variants

Fix: css/theme back in gr.Blocks(), single visible output, yield for instant loading

Add .gitignore for __pycache__

Fix: move css/theme to launch(), remove auto_localization, fix invalid escape sequence

Fix: remove queue=True (removed in Gradio 5+), inject JS directly in card HTML, add demo.queue()

Replace deprecated head= with gr.HTML sanitize_html=False injection

Upgrade Gradio from 4.0.0 to 6.17.3

Fix colorTo to valid HF color (pink)

Add ComponentForge app with MiniCPM3-4B, 3 variant cards, live color preview

initial commit