<!-- 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-ascend
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the ascend dataset. It achieves the following results on the evaluation set:
- Loss: 4.9059
- Wer: 0.6633
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 |
---|---|---|---|---|
22.8456 | 1.32 | 400 | 5.4265 | 1.0 |
5.1649 | 2.65 | 800 | 5.0265 | 1.0 |
4.8311 | 3.97 | 1200 | 4.7929 | 1.0 |
4.5407 | 5.3 | 1600 | 4.5517 | 1.0 |
4.0548 | 6.62 | 2000 | 3.9723 | 0.9219 |
3.2573 | 7.95 | 2400 | 3.2263 | 0.7997 |
2.4774 | 9.27 | 2800 | 2.9549 | 0.7244 |
1.9606 | 10.6 | 3200 | 2.8674 | 0.7015 |
1.5414 | 11.92 | 3600 | 3.0565 | 0.6604 |
1.1045 | 13.24 | 4000 | 3.0344 | 0.6614 |
0.7768 | 14.57 | 4400 | 3.0164 | 0.6811 |
0.5191 | 15.89 | 4800 | 3.2068 | 0.6432 |
0.3 | 17.22 | 5200 | 3.4341 | 0.6638 |
0.1713 | 18.54 | 5600 | 3.7193 | 0.6611 |
0.0987 | 19.87 | 6000 | 4.0413 | 0.6570 |
0.05 | 21.19 | 6400 | 4.5981 | 0.6769 |
0.0321 | 22.52 | 6800 | 4.6013 | 0.6766 |
0.0198 | 23.84 | 7200 | 4.3965 | 0.6714 |
0.0091 | 25.17 | 7600 | 4.6236 | 0.6621 |
0.006 | 26.49 | 8000 | 4.7604 | 0.6606 |
0.0044 | 27.81 | 8400 | 4.8233 | 0.6612 |
0.0027 | 29.14 | 8800 | 4.9059 | 0.6633 |
Framework versions
- Transformers 4.23.1
- Pytorch 1.12.1+cu113
- Datasets 2.5.2
- Tokenizers 0.13.1