<!-- 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-xl-960h-dementiabank
This model is a fine-tuned version of facebook/wav2vec2-large-960h on the None dataset. It achieves the following results on the evaluation set:
- Loss: 3483.2146
- Wer: 0.9860
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: 0.005
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 3
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
13934.5266 | 0.31 | 10 | 71265.4531 | 1.0 |
13443.6406 | 0.62 | 20 | 69977.6016 | 1.0 |
9336.9562 | 0.94 | 30 | 13763.1484 | 0.9843 |
2970.977 | 1.25 | 40 | 17587.7656 | 0.9860 |
1916.3354 | 1.56 | 50 | 4328.4521 | 1.0 |
1417.5775 | 1.88 | 60 | 4486.8071 | 0.9860 |
1841.7689 | 2.19 | 70 | 2988.0303 | 1.0 |
1355.0265 | 2.5 | 80 | 2972.6094 | 0.9860 |
1359.7979 | 2.81 | 90 | 3483.2146 | 0.9860 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.13.3
- Tokenizers 0.10.3