Code

Finetuning bugs#100

Code

Merged

saattrupdan merged 55 commits intomainalexandrainst/coral:mainfrom fix/whisper-finetuningalexandrainst/coral:fix/whisper-finetuningCopy head branch name to clipboardOct 23, 2024

+1,321-1,176

Commits

Commits on Oct 22, 2024

• fix: Convert processor outputs to numpy arrays

  saattrupdan

  committed
  c31a750

  Copy full SHA for c31a750

• chore: Revert commit

  saattrupdan

  committed
  2bb52f1

  Copy full SHA for 2bb52f1

• debug

  saattrupdan

  committed
  d677a96

  Copy full SHA for d677a96

• debug

  saattrupdan

  committed
  5673c60

  Copy full SHA for 5673c60

• docs: Casing

  saattrupdan

  committed
  f533a1e

  Copy full SHA for f533a1e

• debug: Temp logging

  saattrupdan

  committed
  777e270

  Copy full SHA for 777e270

• debug

  saattrupdan

  committed
  bd6aa4a

  Copy full SHA for bd6aa4a

• docs: Remove temp comments

  saattrupdan

  committed
  dbeb536

  Copy full SHA for dbeb536

• docs: Remove temp comment

  saattrupdan

  committed
  c284fab

  Copy full SHA for c284fab

• chore: Logging

  saattrupdan

  committed
  034e277

  Copy full SHA for 034e277

• debug

  saattrupdan

  committed
  c9d7e65

  Copy full SHA for c9d7e65

• chore: Revert

  saattrupdan

  committed
  71bad8c

  Copy full SHA for 71bad8c

• fix: Always process dataset, as it is necessary for training

  saattrupdan

  committed
  215f451

  Copy full SHA for 215f451

• debug

  saattrupdan

  committed
  a52e06d

  Copy full SHA for a52e06d

• debug

  saattrupdan

  committed
  91cd695

  Copy full SHA for 91cd695

• debug

  saattrupdan

  committed
  7d3af9b

  Copy full SHA for 7d3af9b

• fix: Input features to numpy array in data collator

  saattrupdan

  committed
  9085910

  Copy full SHA for 9085910

• fix: Process dataset after interleave_datasets

  saattrupdan

  committed
  d5611fd

  Copy full SHA for d5611fd

• fix: Cast audio to sampling rate before interleave_datasets

  saattrupdan

  committed
  99ef2ac

  Copy full SHA for 99ef2ac

• chore: Update make recipe

  saattrupdan

  committed
  fb38d03

  Copy full SHA for fb38d03

• chore: Make recipe

  saattrupdan

  committed
  b6432e4

  Copy full SHA for b6432e4

• chore: Logging

  saattrupdan

  committed
  26d3cfb

  Copy full SHA for 26d3cfb

• docs: Temp comment

  saattrupdan

  committed
  8e4df84

  Copy full SHA for 8e4df84

• chore: Use AutoModelForSpeechSeq2Seq for initiating Whisper

  saattrupdan

  committed
  b2658df

  Copy full SHA for b2658df

• chore: Use AutoProcessor

  saattrupdan

  committed
  60daf3f

  Copy full SHA for 60daf3f

• debug: Try using default_data_collator

  saattrupdan

  committed
  ee08ffb

  Copy full SHA for ee08ffb

• chore: Revert

  saattrupdan

  committed
  ff46961

  Copy full SHA for ff46961

• chore: Temporarily only use one dataset

  saattrupdan

  committed
  9be064a

  Copy full SHA for 9be064a

• chore: Remove temp block

  saattrupdan

  committed
  b8a2d99

  Copy full SHA for b8a2d99

• chore: Revert

  saattrupdan

  committed
  3541bae

  Copy full SHA for 3541bae

• chore: use pip3 to install pipx

  saattrupdan

  committed
  4fdc302

  Copy full SHA for 4fdc302

• chore: revert

  saattrupdan

  committed
  8c7cbe2

  Copy full SHA for 8c7cbe2

• fix: Use apt to install pipx on Linux

  saattrupdan

  committed
  75efbff

  Copy full SHA for 75efbff

Commits on Oct 23, 2024

• debug

  saattrupdan

  committed
  c88926b

  Copy full SHA for c88926b

• fix: Use hf_config.max_length when setting tokenizer model_max_length for Whisper

  saattrupdan

  committed
  be2ffba

  Copy full SHA for be2ffba

• chore: Update lock file

  saattrupdan

  committed
  dcd610f

  Copy full SHA for dcd610f

• chore: Update pre-commit

  saattrupdan

  committed
  2c1ed86

  Copy full SHA for 2c1ed86

• fix: Use tokenizer model_max_length when padding to max_length

  saattrupdan

  committed
  1bcac9e

  Copy full SHA for 1bcac9e

• chore: Re-enable hook

  saattrupdan

  committed
  a4a558f

  Copy full SHA for a4a558f

• chore: Update lock file

  saattrupdan

  committed
  e216bea

  Copy full SHA for e216bea

• fix: Set split_batches=True

  saattrupdan

  committed
  26f241a

  Copy full SHA for 26f241a

• Merge branch 'fix/whisper-finetuning' of github.com:alexandrainst/coral into fix/whisper-finetuning

  saattrupdan

  committed
  dd24672

  Copy full SHA for dd24672

• fix: Temporarily disable specaugment

  saattrupdan

  committed
  6338756

  Copy full SHA for 6338756

• fix: Set feature_size and num_mel_bins to 160k

  saattrupdan

  committed
  d40f6ea

  Copy full SHA for d40f6ea

• fix: Set ignore_mismatched_sizes=True

  saattrupdan

  committed
  8b4a2b5

  Copy full SHA for 8b4a2b5

• chore: Revert

  saattrupdan

  committed
  6dd8fb7

  Copy full SHA for 6dd8fb7

• fix: Disable split_batches again

  saattrupdan

  committed
  6a026a6

  Copy full SHA for 6a026a6

• fix: Do not force padding to be max_length in multigpu setups

  saattrupdan

  committed
  1917e07

  Copy full SHA for 1917e07

• fix: Set dispatch_batches=False

  saattrupdan

  committed
  38cdcb2

  Copy full SHA for 38cdcb2

• docs: Add TODO comment regarding the commenting of max length change in multigpu

  saattrupdan

  committed
  d4e9559

  Copy full SHA for d4e9559

• Merge branch 'main' of github.com:alexandrainst/coral into fix/whisper-finetuning

  saattrupdan

  committed
  c477ce9

  Copy full SHA for c477ce9

• chore: Update gradio

  saattrupdan

  committed
  3b5e084

  Copy full SHA for 3b5e084

• fix: Set dispatch_batches=False for wav2vec2 models as well

  saattrupdan

  committed
  9b68b86

  Copy full SHA for 9b68b86

• Merge branch 'fix/whisper-finetuning' of github.com:alexandrainst/coral into fix/whisper-finetuning

  saattrupdan

  committed
  70899f6

  Copy full SHA for 70899f6

• fix: Remove cast_to_sampling_rate from process_dataset call in validation module

  saattrupdan

  committed
  dff2c7a

  Copy full SHA for dff2c7a