slplab's picture
Duplicate from slplab/whisper-large_v2-asd_v1
922184b
metadata
license: apache-2.0
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: whisper-large_v2-asd_v1
    results: []
datasets:
  - slplab/asd_apac
language:
  - ko
pipeline_tag: automatic-speech-recognition
duplicated_from: slplab/whisper-large_v2-asd_v1

whisper-large_v2-asd_v1

This model is a fine-tuned version of openai/whisper-large-v2 on slplab/asd_apac dataset. It achieves the following results on the validation set:

  • Loss: 0.8553
  • Wer: 78.4722

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: 1e-05
  • train_batch_size: 16
  • eval_batch_size: 32
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss Wer
0.0617 10.53 100 0.6858 81.9444
0.004 21.05 200 0.7322 79.8611
0.0005 31.58 300 0.7923 80.5556
0.0003 42.11 400 0.8131 79.1667
0.0002 52.63 500 0.8263 76.7361
0.0002 63.16 600 0.8365 77.4306
0.0002 73.68 700 0.8451 78.4722
0.0002 84.21 800 0.8503 78.4722
0.0002 94.74 900 0.8541 78.4722
0.0002 105.26 1000 0.8553 78.4722

Test results

  • Loss: 0.6359
  • Wer: 36.6876

Framework versions

  • Transformers 4.31.0.dev0
  • Pytorch 2.0.1+cu117
  • Datasets 2.13.1
  • Tokenizers 0.13.3