WhisperKitPro: match whisperkitpro-cli 3.x flags and expose Sortformer model options - #103
Conversation
…r model options whisperkitpro-cli 3.x selects the diarization backend with --diarizer and has no --engine or --fast-load flags, so orchestration runs with diarization enabled failed on released CLI builds with "Unknown option". Emit --diarizer and stop emitting --fast-load (the `fast_load` field is kept as a documented no-op so existing configs keep loading). Expose the Sortformer options the CLI already supports: `sortformer_model_version` and `sortformer_model_variant` on the engine and orchestration configs, and `speaker_models_path` on the orchestration config. All are optional and only emitted when set. Also add unit tests for the generated CLI arguments.
61e3a9a to
255fe8c
Compare
| ) | ||
| sortformer_model_version: str | None = Field( | ||
| None, | ||
| description="Sortformer model version (e.g. `v2-1` or `v3-preview`). Only applicable when `engine` is `sortformer`.", |
There was a problem hiding this comment.
We selected nemotron-3-diarization for the model version
There was a problem hiding this comment.
Done in 510d48d: renamed to nemotron-3-diarization (and the variant to the published 684_74MB). Verified with a one-sample CallHome run through openbench-cli using the published model.
arda-argmax
left a comment
arda-argmax
left a comment
There was a problem hiding this comment.
We just need to rename the model version to nemotron-3-diarization. The published model is available here: https://huggingface.co/argmaxinc/speakerkit-pro/tree/main/sortformer/nemotron-3-diarization
Aside from that, LGTM!
| ) | ||
| sortformer_model_version: str | None = Field( | ||
| None, | ||
| description="Sortformer model version (e.g. `v2-1` or `v3-preview`). Only applicable when `engine` is `sortformer`.", |
There was a problem hiding this comment.
Done in 510d48d: renamed to nemotron-3-diarization (and the variant to the published 684_74MB). Verified with a one-sample CallHome run through openbench-cli using the published model.
The Sortformer v3 model was published as `nemotron-3-diarization` with the `684_74MB` variant, so use those names in the field descriptions and tests instead of the local preview names.
2 participants
Summary
whisperkitpro-cli 3.x selects the diarizer with
--diarizerand has no--engineor--fast-loadflags, so orchestration runs with diarization failed with "Unknown option" on released CLI builds.--diarizerinstead of--engine; stop emitting--fast-load(fast_loadkept as a no-op for config compatibility)sortformer_model_version,sortformer_model_variant,speaker_models_path