<!-- 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-xlsr-korean-demo-colab
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4534
- Wer: 0.3272
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: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- 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 |
---|---|---|---|---|
17.4809 | 0.65 | 400 | 4.6145 | 1.0 |
4.4863 | 1.29 | 800 | 4.3819 | 1.0 |
4.2921 | 1.94 | 1200 | 4.1163 | 0.9970 |
2.7971 | 2.59 | 1600 | 1.5376 | 0.8379 |
1.5061 | 3.24 | 2000 | 1.0354 | 0.7299 |
1.1123 | 3.88 | 2400 | 0.7909 | 0.6418 |
0.9037 | 4.53 | 2800 | 0.6345 | 0.5698 |
0.779 | 5.18 | 3200 | 0.5909 | 0.5571 |
0.6834 | 5.83 | 3600 | 0.5339 | 0.5063 |
0.6287 | 6.47 | 4000 | 0.5326 | 0.4954 |
0.5518 | 7.12 | 4400 | 0.4930 | 0.4607 |
0.5315 | 7.77 | 4800 | 0.4577 | 0.4451 |
0.4867 | 8.41 | 5200 | 0.4547 | 0.4382 |
0.4543 | 9.06 | 5600 | 0.4581 | 0.4371 |
0.4089 | 9.71 | 6000 | 0.4387 | 0.4258 |
0.3893 | 10.36 | 6400 | 0.4300 | 0.4100 |
0.3751 | 11.0 | 6800 | 0.4265 | 0.4137 |
0.3333 | 11.65 | 7200 | 0.4294 | 0.4011 |
0.3039 | 12.3 | 7600 | 0.4187 | 0.3912 |
0.2974 | 12.94 | 8000 | 0.4079 | 0.3805 |
0.2658 | 13.59 | 8400 | 0.4273 | 0.3864 |
0.2676 | 14.24 | 8800 | 0.4103 | 0.3734 |
0.2466 | 14.89 | 9200 | 0.4122 | 0.3701 |
0.2282 | 15.53 | 9600 | 0.4176 | 0.3650 |
0.2186 | 16.18 | 10000 | 0.4199 | 0.3632 |
0.2132 | 16.83 | 10400 | 0.4159 | 0.3671 |
0.1962 | 17.48 | 10800 | 0.4321 | 0.3641 |
0.1922 | 18.12 | 11200 | 0.4300 | 0.3535 |
0.1827 | 18.77 | 11600 | 0.4244 | 0.3596 |
0.1709 | 19.42 | 12000 | 0.4191 | 0.3518 |
0.157 | 20.06 | 12400 | 0.4308 | 0.3496 |
0.147 | 20.71 | 12800 | 0.4360 | 0.3457 |
0.1502 | 21.36 | 13200 | 0.4329 | 0.3431 |
0.1448 | 22.01 | 13600 | 0.4334 | 0.3432 |
0.1407 | 22.65 | 14000 | 0.4392 | 0.3440 |
0.1342 | 23.3 | 14400 | 0.4418 | 0.3399 |
0.1325 | 23.95 | 14800 | 0.4360 | 0.3383 |
0.1183 | 24.6 | 15200 | 0.4521 | 0.3359 |
0.1174 | 25.24 | 15600 | 0.4426 | 0.3322 |
0.1137 | 25.89 | 16000 | 0.4438 | 0.3356 |
0.1129 | 26.54 | 16400 | 0.4547 | 0.3347 |
0.1077 | 27.18 | 16800 | 0.4482 | 0.3300 |
0.0999 | 27.83 | 17200 | 0.4491 | 0.3281 |
0.0978 | 28.48 | 17600 | 0.4533 | 0.3281 |
0.0997 | 29.13 | 18000 | 0.4542 | 0.3283 |
0.0908 | 29.77 | 18400 | 0.4534 | 0.3272 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.12.0+cu113
- Datasets 1.18.3
- Tokenizers 0.12.1