Code

Whisper encoder + No 30 second padding#5

Code

Merged

farzadab merged 9 commits intomainfixie-ai/ultravox:mainfrom farzad-whisper-prfixie-ai/ultravox:farzad-whisper-prCopy head branch name to clipboardJun 4, 2024

+175-11

Commits

Commits on Jun 4, 2024

• enable whisper model (no need for max_padding)

  farzadab

  committed
  e2ee6da

  Copy full SHA for e2ee6da

• bugfix: boolq_in didn't have audio_transcript

  farzadab

  committed
  8b46222

  Copy full SHA for 8b46222

• bugfix: move model to device before merging lora weights

  farzadab

  committed
  ab67298

  Copy full SHA for ab67298

• formatting

  farzadab

  committed
  1a52c9e

  Copy full SHA for 1a52c9e

• rename modified_whisper -> whisper_model_modified

  farzadab

  committed
  79c9dbd

  Copy full SHA for 79c9dbd

• boolq_in: use _get_transcribe_sample

  farzadab

  committed
  988b33a

  Copy full SHA for 988b33a

• add comment for right padding audio_values

  farzadab

  committed
  a6f10ab

  Copy full SHA for a6f10ab

• improve whisper comments

  farzadab

  committed
  333ce1c

  Copy full SHA for 333ce1c

• reducing exposure of configurable audio_padding

  farzadab

  committed
  60b330f

  Copy full SHA for 60b330f