0.2.1

elyxlz/voxtral0.2.1Jul 16, 2026by zhu-han

AI Summary

This release focuses on performance improvements for ASR models, audio processing, and adding text normalization capabilities. It also introduces several community projects and bug fixes.

Key Highlights

  • Fixed silent FP32 attention, resulting in ~4x training speedup
  • Added VoiceClonePrompt save/load() for cross-session voice reuse
  • Fixed ASR model always loading on GPU 0; added asr_device option
  • Added opt-in text normalization for numbers, dates, and currency

New Features

  • Added audio.cpp to community projects
  • VoiceClonePrompt save()/load() methods
  • Opt-in text normalization support
  • Stale issue workflow

Full Release Notes

## What's Changed
* Add audio.cpp to community projects by @0xShug0 in https://github.com/k2-fsa/OmniVoice/pull/209
* Make flex_attention honor autocast (fixes silent fp32 attention, ~4x training slowdown) by @shuaills in https://github.com/k2-fsa/OmniVoice/pull/212
* Honor asr_model_name when the ASR model is loaded lazily by @HenryVarro666 in https://github.com/k2-fsa/OmniVoice/pull/221
* Add VoiceClonePrompt.save()/load() for cross-session voice reuse by @HenryVarro666 in https://github.com/k2-fsa/OmniVoice/pull/223
* Fix ASR model always loading on GPU 0; add asr_device option by @HenryVarro666 in https://github.com/k2-fsa/OmniVoice/pull/224
* Avoid decoding the full reference audio just to get its duration by @aradix85 in https://github.com/k2-fsa/OmniVoice/pull/220
* Add opt-in text normalization (numbers, dates, currency) via generate(normalize_text=True) by @HenryVarro666 in https://github.com/k2-fsa/OmniVoice/pull/227
* Add stale issue workflow by @zhu-han in https://github.com/k2-fsa/OmniVoice/pull/228

## New Contributors
* @0xShug0 made their first contribution in https://github.com/k2-fsa/OmniVoice/pull/209
* @shuaills made their first contribution in https://github.com/k2-fsa/OmniVoice/pull/212
* @HenryVarro666 made their first contribution in https://github.com/k2-fsa/OmniVoice/pull/221
* @aradix85 made their first contribution in https://github.com/k2-fsa/OmniVoice/pull/220

**Full Changelog**: https://github.com/k2-fsa/OmniVoice/compare/0.2.0...0.2.1