Starstruk Logo MarkStarstruk

v0.9.1

EricLBuehler/mistral.rsv0.9.1Aug 14, 2026by EricLBuehler

AI Summary

This release introduces significant performance improvements to the scheduler and concurrent serving, adds support for GGUF loading across all models, and introduces new features like LoRA kernels and Muse Glimmer model support.

Key Highlights

  • Major scheduler and concurrent serving performance improvements
  • First-class support for loading GGUFs for all supported models
  • New model support: Muse Glimmer
  • LoRA improvements: True LoRA kernels and dynamic adapter support
  • Core improvements: Support for lmhead/embedding quantization

New Features

  • Suggested commands for installer script
  • Optimized UQFF generation pass
  • True LoRA kernels and dynamic adapter support
  • LMHead/Embedding quantization support
  • GGUF loading support for all models

Full Release Notes

## What's Changed
* feat(install): show suggested commands for installer script by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2315
* feat(uqff): no dummy run for uqff generation pass by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2316
* fix(sampler): return natural-log probabilities from `logprobs`, not log10 by @dx2102 in https://github.com/EricLBuehler/mistral.rs/pull/2327
* fix(core): address recent correctness regressions by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2345
* chore(deps): bump serde_with from 3.16.1 to 3.21.0 by @dependabot[bot] in https://github.com/EricLBuehler/mistral.rs/pull/2349
* feat(lora): add true lora kernels and dynamic adapter support by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2351
* feat(core): support lmhead/embedding quantization by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2280
* fix(quant): misc fixes, warmup metal kernels by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2353
* fix(gemma4): improve audios ordering around text by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2355
* feat(core): major improvements to scheduler & concurrent serving performance by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2354
* feat(core): first-class support for loading GGUFs for all supported models by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2365
* feat(models): support Muse Glimmer by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2366

## New Contributors
* @dx2102 made their first contribution in https://github.com/EricLBuehler/mistral.rs/pull/2327

**Full Changelog**: https://github.com/EricLBuehler/mistral.rs/compare/v0.9.0...v0.9.1