Add native transformers GraniteSpeechNarForCTC format

#6
by eustlb HF Staff - opened
IBM Granite org

Converts this checkpoint from the remote-code implementation to the native GraniteSpeechNarForCTC model now available in πŸ€— Transformers.

Changes

  • Native config.json (model_type: granite_speech_nar, no auto_map) and re-keyed model.safetensors.
  • Processor/feature-extractor config in processor_config.json (native SequenceFeatureExtractor layout).
  • Remove the remote-code modules (*_granite_speech_nar.py, __init__.py) and the stale preprocessor_config.json (superseded by the nested feature-extractor config).

Produced by transformers/src/transformers/models/granite_speech_nar/convert_granite_speech_nar_to_hf.py.

Follow-ups for maintainers: model.sig needs re-signing (weights changed), and the README usage snippet can drop trust_remote_code.

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment