Rebuild: settings+new-chat visible in chat header

Rebuild: fix browser cache persistence

Rebuild: settings panel + updated examples + Vexorium style

Rebuild: header Vexorium logo, right Nemotron image

Rebuild dist/: Nemotron × Vexorium logo

Rebuild dist/ with Vexorium brand style

Rebuild dist/ with grpo-v21 model ID

add back test-webgpu.html to dist

revert to 9b50a208 + modelId only, thinking enabled

deploy: fine-tuned model with thinking disabled (Q4 thinking broken on WebGPU)

fix: adapter.requestAdapterInfo() not available in all browsers

fix: use explicit ESM path for transformers.js CDN import

fix: move test page to dist/ so static Space can serve it

revert: back to 9b50a208 (reference model, known working)

fix: handle Q4 model missing </think> on WebGPU

test: 9b50a208 code + only modelId changed to fine-tuned model

revert: back to 9b50a208 (reference model) for testing

fix: MessageBubble falls back to reasoning when content is empty — eliminates 'No response'

debug: add FINAL console.log to trace parser state — must find root cause

fix: when model skips </think>, move reasoning to content and clear reasoning

fix: extract last question from reasoning when model stops after </think> without content

fix: remove cache_dir override — use default browser Cache API for ONNX persistence

fix: handle missing </think> — fallback reasoning to content, never show 'No response'

fix: explicit eos_token_id=[2,11] at generation time — works even with cached model config

fix: revert to reference streamer approach + fix eos_token_id in model config

fix: filter special tokens and trailing whitespace from stream output

test: float-only patch — base q4 weights + fine-tuned layernorms/biases

test: use reference base model to verify pipeline works

fix: remove ort-web override — model ONNX now has single external data file

fix: transformers.js 4.0.0-next.8 + ort-web 1.16.3 — fixes both nemotron_h and MountedFiles errors

fix: pin ort-web to 1.16.3 to fix Module.MountedFiles error with q4 ONNX external data

Revert to dtype=q4 — model_q4.onnx now available in our repo

Fix: use dtype=fp32 to load model.onnx (not model_q4.onnx which doesn't exist)

Rebuild dist with bobber/lex-interviewer-nemotron-4b-grpo-v12 model ID

Add mobile warning (4B model too large for phone VRAM), update landing page branding

Enable browser Cache API for ONNX files, better download progress

Add settings panel with editable system prompt, update examples

Customize for Lex Fridman interviewer: system prompt, examples, branding

Upload 88 files