<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
wav2vec2-large-xls-r-300m-zeroth
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the zeroth_korean_asr dataset. It achieves the following results on the evaluation set:
- Loss: 0.7052
- Wer: 0.4621
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: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
15.1763 | 1.61 | 400 | 4.6768 | 1.0 |
3.1779 | 3.21 | 800 | 1.6680 | 0.8752 |
1.052 | 4.82 | 1200 | 0.9580 | 0.7332 |
0.5412 | 6.42 | 1600 | 0.7752 | 0.5993 |
0.3281 | 8.03 | 2000 | 0.7158 | 0.5615 |
0.2312 | 9.64 | 2400 | 0.6975 | 0.5532 |
0.2001 | 11.24 | 2800 | 0.7489 | 0.5677 |
0.1587 | 12.85 | 3200 | 0.6954 | 0.5267 |
0.1321 | 14.46 | 3600 | 0.7329 | 0.5371 |
0.1178 | 16.06 | 4000 | 0.7534 | 0.5341 |
0.103 | 17.67 | 4400 | 0.7046 | 0.5066 |
0.0843 | 19.28 | 4800 | 0.7507 | 0.5028 |
0.079 | 20.88 | 5200 | 0.7137 | 0.4886 |
0.0647 | 22.49 | 5600 | 0.7170 | 0.4855 |
0.0565 | 24.1 | 6000 | 0.7124 | 0.4781 |
0.0487 | 25.7 | 6400 | 0.7043 | 0.4721 |
0.0433 | 27.31 | 6800 | 0.7128 | 0.4557 |
0.0379 | 28.91 | 7200 | 0.7052 | 0.4621 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.10.3