94 2 weeks ago

A modality-stripped variant of google/gemma-4-E4B-it for text-only classification, entity extraction, and structured-memory extraction. The vision encoder (150M params) and audio encoder (300M params) are removed; the text path is unchanged.

tools thinking
0f73cb1f6222 · 137B
<start_of_turn>user
{{ .System }}{{ if .System }}
{{ end }}{{ .Prompt }}<end_of_turn>
<start_of_turn>model
{{ .Response }}<end_of_turn>