fix: remove phi models

fix: remove smollm model option

fix: remove gemma and smollm models

fix: relabel gemma and add space metadata

fix: bust cache for model picker

fix: remove banner and improve mobile layout

feat: add phi mini 4k and 128k models

fix: use english gemma 2 fallback

fix: switch gemma slot to gemma 2

fix: load model chooser and gemma export

feat: add model switcher for local spaces

Add .gitignore; remove tracked .DS_Store

Revert to eeb49e0: Tighter intro bubble + Clear restores default

Revert to e212a6f: restore model switching (Llama ↔ MedGemma multimodal)

Revert to Llama-only: remove .model-picker block, fix Files-tab link

Revert to Llama-only: remove MedGemma model + picker code

MedGemma v2: point at kellogcheung/medgemma-1.5-4b-it-ONNX-v2 (fused-op text-only export)

MedGemma: q4 decoder (q4f16 was producing noise on this fine-tune)

MedGemma: use q4f16 for embed_tokens and vision_encoder too

Skip WASM fallback for multimodal; dump full WebGPU error fields

Add #fresh URL hash to force re-download (busts HF max-age=31536000 cache)

Bump transformers.js 3.8.1 -> 4.2.0 (adds Gemma3ImageProcessor)

MedGemma: use AutoProcessor + AutoModelForImageTextToText

Add model switching: Llama 3.2 1B (text) ↔ MedGemma 1.5 4B IT (multimodal)

Add .gitignore; untrack accidentally-committed .DS_Store

Tighter intro bubble + Clear restores default

Wrap AI assistant greeting in paragraph tags for improved formatting

Remove paragraph tags from AI assistant greeting in chat bubble

Remove disclaimer about professional advice from chat bubble

Remove redundant tips section from index.html

Chat scrolls inside its panel + adjustable generation params

Better file-extract UX + auto-resume after Safari tab eviction

Update placeholder text in chat input and remove redundant tips in index.html

Update title in README.md to correct project name to "Local Llama"

Update title and emoji in README.md to reflect project name change

Update header title from "MedGemma Local" to "Local Llama"

Update title and remove logo from header in index.html

Clean up UI text for Llama 3.2 baseline

Switch baseline to Llama-3.2-1B-Instruct-q4f16

Try gemma-3-1b-it-ONNX-GQA variant (browser-targeted)

Upgrade transformers.js 3.5.2 -> 3.8.1 for Gemma 3 fixes

Smoke test: switch to gemma-3-270m-it (1B fails ONNX parse)

Switch quantization q4 -> q4f16 (q4 variant fails to load)

Auto-fallback WebGPU -> WASM; surface ONNX numeric error codes

Build MedGemma Local: transformers.js static Space

initial commit