Skip to content

v0.105.1

Choose a tag to compare

View all tags
@basnijholt basnijholt released this 08 Sep 13:08
v0.105.1
f4d931f
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Bug fixes

  • Release unused CUDA cache after each transcription in the transformers ASR backend, keeping model weights loaded for the next request. Long recordings no longer leave their peak temporary GPU allocation reserved throughout the idle timeout. In a Qwen3-ASR test, idle process VRAM dropped from 10.89 GiB to 4.13 GiB. (#642)
  • Clear temporary tensor references held by failed-request tracebacks, including chained and grouped exceptions, before releasing CUDA cache. CPU and MPS behavior is unchanged. (#642)

Full changelog: v0.105.0...v0.105.1

Assets 4
Loading