nb-whisper-base / README.md
pere's picture
Saving weights and logs of step 49999 - epoch 6
78981a6
|
raw
history blame
6.38 kB
metadata
language:
  - 'no'
license: apache-2.0
base_model: NbAiLab/nb-whisper-base-RC1
tags:
  - audio
  - asr
  - automatic-speech-recognition
  - hf-asr-leaderboard
model-index:
  - name: nb-whisper-base-v0.8-vad3
    results: []

nb-whisper-base-v0.8-vad3

This model is a fine-tuned version of NbAiLab/nb-whisper-base-RC1 on the NbAiLab/ncc_speech_styling_v2_vad3 dataset. It achieves the following results on the evaluation set:

  • step: 49999
  • validation_nst_loss: 0.4686
  • train_loss: 0.5185
  • validation_nst_wer: 4.7308
  • validation_nst_cer: 1.4896
  • validation_nst_exact_wer: 5.4603
  • validation_nst_exact_cer: 1.6035
  • validation_clean_stortinget_no_loss: 0.6454
  • validation_clean_stortinget_no_wer: 12.3235
  • validation_clean_stortinget_no_cer: 7.1963
  • validation_clean_stortinget_no_exact_wer: 15.7965
  • validation_clean_stortinget_no_exact_cer: 7.7657

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0001
  • lr_scheduler_type: linear
  • per_device_train_batch_size: 32
  • total_train_batch_size_per_node: 128
  • total_train_batch_size: 1024
  • total_optimization_steps: 50,000
  • starting_optimization_step: None
  • finishing_optimization_step: 50,000
  • num_train_dataset_workers: 32
  • num_hosts: 8
  • total_num_training_examples: 51,200,000
  • steps_per_epoch: 7455
  • num_beams: None
  • weight_decay: 0.01
  • adam_beta1: 0.9
  • adam_beta2: 0.98
  • adam_epsilon: 1e-06
  • dropout: True
  • bpe_dropout_probability: 0.2
  • activation_dropout_probability: 0.1

Training results

step validation_nst_loss train_loss validation_nst_wer validation_nst_cer validation_nst_exact_wer validation_nst_exact_cer validation_clean_stortinget_no_loss validation_clean_stortinget_no_wer validation_clean_stortinget_no_cer validation_clean_stortinget_no_exact_wer validation_clean_stortinget_no_exact_cer
0 0.4594 1.1239 4.7580 1.4524 5.4984 1.5715 0.5892 11.7267 6.8997 15.1963 7.4616
5000 0.4884 0.6674 5.5038 1.7824 6.2987 1.9002 0.6483 13.0246 7.4564 16.7240 8.0698
10000 0.4834 0.6476 5.5474 1.8085 6.2823 1.9149 0.6467 12.9252 7.4203 16.4844 8.0062
15000 0.4843 0.6350 5.6672 1.7963 6.4456 1.9222 0.6517 12.7144 7.3438 16.2709 7.9357
20000 0.4760 0.5842 5.1010 1.6146 5.8033 1.7244 0.6517 12.6338 7.3208 16.2780 7.9279
25000 0.4806 0.5751 5.1718 1.6323 5.8849 1.7354 0.6512 12.6078 7.2514 16.1713 7.8452
30000 0.4740 0.5723 4.7090 1.4878 5.4058 1.6035 0.6482 12.4917 7.2427 16.0337 7.8222
35000 0.4737 0.5598 4.8887 1.5241 5.7216 1.6548 0.6510 12.2572 7.0924 15.7917 7.6744
40000 0.4731 0.5471 4.8941 1.5297 5.6073 1.6438 0.6526 12.4064 7.1642 15.8842 7.7360
45000 0.4693 0.5217 4.7580 1.4645 5.4930 1.5834 0.6471 12.1696 7.0908 15.7134 7.6776
49999 0.4686 0.5185 4.7308 1.4896 5.4603 1.6035
49999 0.6454 0.5185 12.3235 7.1963 15.7965 7.7657

Framework versions

  • Transformers 4.34.1
  • Datasets 2.16.1
  • Tokenizers 0.14.1