Skip to content

v0.9.0 - reasoning channel for every model

Choose a tag to compare

@srvsngh99 srvsngh99 released this 17 Jun 07:25
· 189 commits to main since this release

KrillLM v0.9.0 generalizes the reasoning ("thinking") channel to any model that has one.

  • KRILL_ENABLE_THINKING=1 (or the per-call flag) now turns on the reasoning channel for any model whose chat template supports it - Qwen 3, the Gemma-4 channel fine-tunes (e.g. gemma-4-12b-coder), and any template that branches on enable_thinking. v0.8.0 only special-cased the Gemma-4 coder.
  • The engine renders the model's own chat template with enable_thinking set (via the same Jinja engine, re-encoded so ChatML/channel special tokens stay intact); the Gemma-4 channel template, which Swift's Jinja cannot parse, keeps its direct builder.
  • Safe no-op for models with no thinking channel, and behavior with thinking off is unchanged.

Install: brew tap srvsngh99/krillm && brew install krillm (or brew upgrade krillm)
Asset: krillm-0.9.0-arm64-apple-macos.tar.gz (Apple Silicon). Changelog: [0.9.0].