jethrowang's picture
End of training
15c4b46 verified
|
raw
history blame
2.71 kB
metadata
language:
  - zh
license: apache-2.0
base_model: openai/whisper-tiny
tags:
  - generated_from_trainer
datasets:
  - formospeech/hat_asr_aligned
model-index:
  - name: Whisper Tiny Hakka Condenser
    results: []

Whisper Tiny Hakka Condenser

This model is a fine-tuned version of openai/whisper-tiny on the HAT ASR Aligned dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2092
  • Cer: 12.0743

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: 64
  • eval_batch_size: 32
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 976
  • training_steps: 9760
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Cer
1.186 0.9980 488 1.2024 49.3007
0.3895 1.9959 976 0.4851 26.5471
0.2054 2.9939 1464 0.3152 17.7208
0.1232 3.9918 1952 0.2593 16.5314
0.0844 4.9898 2440 0.2355 14.3364
0.0557 5.9877 2928 0.2269 15.5339
0.0367 6.9857 3416 0.2197 13.5042
0.0265 7.9836 3904 0.2149 13.1851
0.0199 8.9816 4392 0.2107 13.2591
0.0131 9.9796 4880 0.2113 14.0000
0.0084 10.9775 5368 0.2118 14.3977
0.006 11.9755 5856 0.2103 14.0104
0.0046 12.9734 6344 0.2109 13.5192
0.0036 13.9714 6832 0.2086 13.4972
0.003 14.9693 7320 0.2074 13.3643
0.0027 15.9673 7808 0.2083 13.7666
0.0023 16.9652 8296 0.2095 12.4280
0.0021 17.9632 8784 0.2105 12.3968
0.0019 18.9611 9272 0.2095 12.4581
0.0018 19.9591 9760 0.2092 12.0743

Framework versions

  • Transformers 4.42.3
  • Pytorch 2.3.0+cu121
  • Datasets 2.20.0
  • Tokenizers 0.19.1