Gemma 4 26B A4B powers local ASCII chatbot
A GitHub-hosted single-file HTML chatbot demo built around Gemma 4 26B A4B Instruct running locally across an AMD RX 7900 XT and an RTX 3060 Ti. The page connects to LM Studio’s API and includes streaming output, Markdown rendering, model selection, six tuning sliders, message editing with history branching, regenerate, abort, and system prompt support. The repo frames it as a collection of generated single-file HTML outputs, with the showcased chatbot being the main artifact.
A strong proof-of-capability for local Gemma 4 workflows, this looks more like a showcase than a packaged product. The main draw is the combination of dual-GPU sharding, 32K context, and 50-65 t/s, while the practical chat UX features cover streaming, branching edits, regenerate, abort, and system prompts. The single-file HTML format makes the demo easy to inspect, share, and reuse, and the note about Claude fixing two DOM bugs suggests there were some edge-case UI issues along the way.
DISCOVERED
7d ago
2026-04-05
PUBLISHED
7d ago
2026-04-05
RELEVANCE
AUTHOR
Reaper_9382