| license: apache-2.0 | |
| tags: | |
| - generated_from_trainer | |
| datasets: | |
| - cvbn | |
| model-index: | |
| - name: wav2vec2_murad | |
| results: [] | |
| <!-- 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_murad | |
| This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the cvbn dataset. | |
| It achieves the following results on the evaluation set: | |
| - eval_loss: 0.2006 | |
| - eval_wer: 0.2084 | |
| - eval_runtime: 556.4634 | |
| - eval_samples_per_second: 8.985 | |
| - eval_steps_per_second: 0.562 | |
| - epoch: 12.32 | |
| - step: 28800 | |
| ## 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: 7.5e-05 | |
| - train_batch_size: 16 | |
| - eval_batch_size: 16 | |
| - 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: 30 | |
| - mixed_precision_training: Native AMP | |
| ### Framework versions | |
| - Transformers 4.21.1 | |
| - Pytorch 1.11.0+cu102 | |
| - Datasets 2.4.0 | |
| - Tokenizers 0.12.1 | |