Upload js/model.js with huggingface_hub

Upload index.html with huggingface_hub

Upload js/app.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload index.html with huggingface_hub

Upload js/app.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Upload js/model.js with huggingface_hub

Pin MODEL_REVISION to specific commit hash for CDN cache busting

Rebuild with transformers.js v4.2.0 and properly exported ONNX model

Fix WebGPU hang: use q4 decoder (now int32-patched) for all backends

Add shader compilation progress indicator

Update MODEL_REVISION to main to include WebGPU-compatible model files

Use WebGPU-compatible decoder (int32) for GPU mode

Add fp16→fp32 fallback for vision encoder WebGPU compatibility

Update MODEL_REVISION to latest decoder rebuild (asymmetric q4, no accuracy_level)

Update revision for rebuilt vision_encoder_fp16

Update revision for fixed embed_tokens input shape

Switch back to our model with fixed GatherBlockQuantized embed_tokens

TEST: Use onnx-community model to verify infrastructure works

Add CPU/GPU toggle, token streaming, fix chat input enable

Fix: enable chat input after model load, add WebGPU detection + inference logging

Update revision to fix external data issue

Pin model revision to bypass cache (fix duplicate node error)

Upload js/model.js with huggingface_hub

Upload folder using huggingface_hub

initial commit