proseph/ctrlv-wav2vec2-tokenizer

1---2license: apache-2.03tags:4- generated_from_trainer5model-index:6- name: ctrlv-wav2vec2-tokenizer7  results: []8---9 10<!-- This model card has been generated automatically according to the information the Trainer had access to. You11should probably proofread and complete it, then remove this comment. -->12 13# ctrlv-wav2vec2-tokenizer14 15This model is a fine-tuned version of [facebook/wav2vec2-base](https://huggingface.co/facebook/wav2vec2-base) on the None dataset.16It achieves the following results on the evaluation set:17- Loss: 0.396718- Wer: 0.313819 20## Model description21 22More information needed23 24## Intended uses & limitations25 26More information needed27 28## Training and evaluation data29 30More information needed31 32## Training procedure33 34### Training hyperparameters35 36The following hyperparameters were used during training:37- learning_rate: 0.000138- train_batch_size: 3239- eval_batch_size: 840- seed: 4241- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-0842- lr_scheduler_type: linear43- lr_scheduler_warmup_steps: 100044- num_epochs: 3045- mixed_precision_training: Native AMP46 47### Training results48 49| Training Loss | Epoch | Step | Validation Loss | Wer    |50|:-------------:|:-----:|:----:|:---------------:|:------:|51| 3.4359        | 3.45  | 500  | 1.3595          | 0.9159 |52| 0.5692        | 6.9   | 1000 | 0.4332          | 0.4036 |53| 0.2198        | 10.34 | 1500 | 0.4074          | 0.3678 |54| 0.1314        | 13.79 | 2000 | 0.3480          | 0.3409 |55| 0.0929        | 17.24 | 2500 | 0.3714          | 0.3346 |56| 0.0692        | 20.69 | 3000 | 0.3977          | 0.3224 |57| 0.0542        | 24.14 | 3500 | 0.4068          | 0.3187 |58| 0.0422        | 27.59 | 4000 | 0.3967          | 0.3138 |59 60 61### Framework versions62 63- Transformers 4.11.364- Pytorch 1.10.0+cu11165- Datasets 1.18.366- Tokenizers 0.10.367