Starstruk Logo MarkStarstruk

v0.30.8

parruda/swarmv0.30.8Jun 12, 2026by github-actions[bot]

AI Summary

Improvements to MLX inference stability, prompt caching logic, and support for recurrent models with enhanced kernel handling.

Key Highlights

  • Improved prompt caching by decoupling from context shift
  • More stable MLX inference with hardened layers
  • Speculative decoding snapshots for reliability
  • Improved recurrent model support with gated-delta kernels

New Features

  • Speculative decoding snapshots
  • Gated-delta kernels for recurrent models

Full Release Notes

## What's Changed
* Fixed `ollama launch` selecting the wrong provider in some cases
* Improved prompt caching by decoupling it from context shift for better KV cache reuse
* More stable MLX inference with hardened linear and embedding layers
* MLX runner now creates snapshots during prompt processing and speculative decoding for improved reliability
* Improved recurrent model support with per-boundary states from the gated-delta kernels

**Full Changelog**: https://github.com/ollama/ollama/compare/v0.30.7...v0.30.8