Added
-
Timeout configured to allow model to be unloaded when idle
MODEL_IDLE_TIMEOUT=300
Defaults to
0. After no activity for this period (in seconds), unload the model until it is requested again. Setting
0disables the timeout, keeping the model loaded indefinitely. -
Added detection confidence to langauge detection endpoint
-
Set mel generation to adjust n_dims automatically to match the loaded model
-
Refactor classes, Add comments, implement abstract methods, and add factory method for engine selection
Changed
- Upgraded
- SYSTRAN/faster-whisper to v1.1.0
- uvicorn to v0.34.0
- tqdm to v4.67.1
- python-multipart to v0.0.20
- fastapi to v0.115.6
- pytest to v8.3.4
- ruff to v0.8.3
- black to v24.10.0
- mkdocs to v1.6.1
- mkdocs-material to v9.5.49
- pymdown-extensions to v10.12
What's Changed
- Correctly load n_dims for openai_whisper model by @aidancrowther in #259
- Add confidence percentage to language detection by @aidancrowther in #260
- Fix typo in Endpoints Documentation by @MatthewGlenn in #239
- Fix incorrect spelling of output param in docs by @kaveet in #243
- Add vram flushing support by @aidancrowther in #256
- Refactor classes and upgrade faster whisper to v1.1.0 by @ahmetoner in #261
New Contributors
- @aidancrowther made their first contribution in #259
- @MatthewGlenn made their first contribution in #239
- @kaveet made their first contribution in #243
Full Changelog: v1.6.0...v1.7.0