V‑Blaze version 7.3.2 (February 2022)
V‑Blaze version 7.3.2-2, voci-webapi-3.0.3-2
New Features and Improvements
Improved the efficiency, accuracy, and logging of GPU process auto-sizing.
More efficient and complete use of GPU memory.
Enhanced the models.list configuration functionality to allow optional loading and sizing of acoustic models in CPU instead of GPU. Added a configurable gpumempad setting (default 40) for additional fine tuning. Refer to Language Model Configuration for more information.
Added
stream.tat.max
andstream.max
data to the /metrics, /stats, and /status outputs. Refer to metrics, stats, and status for more information.
V‑Blaze version 7.3.2
New Features and Improvements
Improved system resource usage by fixing potential stream hangs on systems configured for high concurrent streams and numerous language models.
Fixed failures related to non-standard sample rates.
Fixed possible stream hangs in edge case scenarios.
Improved time window consistency for data returned by /stats and /metrics API endpoints.
Added additional validation of audio headers when streaming audio from network sockets.
Miscellaneous improvements to logging and internal processes.