<!-- 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-53-Total_2e-4_2
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: 0.2733
- Wer: 0.2116
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: 0.0002
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
5.2741 | 0.1 | 200 | 2.9070 | 0.9707 |
2.034 | 0.2 | 400 | 0.7240 | 0.6798 |
1.0037 | 0.3 | 600 | 0.5651 | 0.5368 |
0.8834 | 0.4 | 800 | 0.4709 | 0.4669 |
0.7973 | 0.5 | 1000 | 0.4305 | 0.4261 |
0.7489 | 0.6 | 1200 | 0.4017 | 0.3763 |
0.7507 | 0.7 | 1400 | 0.3662 | 0.3481 |
0.7108 | 0.8 | 1600 | 0.3604 | 0.3513 |
0.7151 | 0.9 | 1800 | 0.3563 | 0.3406 |
0.6755 | 1.0 | 2000 | 0.3365 | 0.3210 |
0.6038 | 1.1 | 2200 | 0.3394 | 0.3053 |
0.6109 | 1.2 | 2400 | 0.3179 | 0.2844 |
0.5999 | 1.3 | 2600 | 0.3166 | 0.2773 |
0.6291 | 1.4 | 2800 | 0.3134 | 0.2733 |
0.626 | 1.5 | 3000 | 0.3060 | 0.2690 |
0.6188 | 1.6 | 3200 | 0.3038 | 0.2644 |
0.5757 | 1.7 | 3400 | 0.3015 | 0.2566 |
0.5943 | 1.8 | 3600 | 0.2925 | 0.2494 |
0.6043 | 1.9 | 3800 | 0.2858 | 0.2491 |
0.5874 | 2.0 | 4000 | 0.2874 | 0.2452 |
0.5263 | 2.1 | 4200 | 0.2800 | 0.2364 |
0.5282 | 2.2 | 4400 | 0.2848 | 0.2387 |
0.4953 | 2.3 | 4600 | 0.2793 | 0.2360 |
0.5428 | 2.4 | 4800 | 0.2863 | 0.2414 |
0.5618 | 2.5 | 5000 | 0.2788 | 0.2350 |
0.5395 | 2.6 | 5200 | 0.2765 | 0.2325 |
0.5178 | 2.7 | 5400 | 0.2787 | 0.2351 |
0.5264 | 2.8 | 5600 | 0.2755 | 0.2312 |
0.5222 | 2.9 | 5800 | 0.2692 | 0.2258 |
0.5184 | 3.0 | 6000 | 0.2681 | 0.2242 |
0.4826 | 3.1 | 6200 | 0.2736 | 0.2224 |
0.479 | 3.2 | 6400 | 0.2896 | 0.2353 |
0.4938 | 3.3 | 6600 | 0.2744 | 0.2252 |
0.4772 | 3.4 | 6800 | 0.2735 | 0.2242 |
0.4831 | 3.5 | 7000 | 0.2721 | 0.2225 |
0.4869 | 3.6 | 7200 | 0.2710 | 0.2194 |
0.4515 | 3.7 | 7400 | 0.2692 | 0.2196 |
0.4732 | 3.8 | 7600 | 0.2729 | 0.2269 |
0.4683 | 3.9 | 7800 | 0.2713 | 0.2211 |
0.4674 | 4.0 | 8000 | 0.2642 | 0.2116 |
0.4239 | 4.1 | 8200 | 0.2773 | 0.2176 |
0.4306 | 4.2 | 8400 | 0.2779 | 0.2191 |
0.441 | 4.3 | 8600 | 0.2758 | 0.2136 |
0.4343 | 4.4 | 8800 | 0.2797 | 0.2203 |
0.4059 | 4.5 | 9000 | 0.2763 | 0.2159 |
0.4399 | 4.6 | 9200 | 0.2755 | 0.2123 |
0.4131 | 4.7 | 9400 | 0.2741 | 0.2124 |
0.4331 | 4.8 | 9600 | 0.2728 | 0.2101 |
0.4288 | 4.9 | 9800 | 0.2730 | 0.2110 |
0.4341 | 5.0 | 10000 | 0.2733 | 0.2116 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.10.3