Whisper Tiny ID - FLEURS-CV
This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5129
- Wer: 31.1298
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: 1e-05
- train_batch_size: 32
- 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: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.617 | 1.43 | 500 | 0.5956 | 40.1521 |
0.4062 | 2.86 | 1000 | 0.4991 | 33.2066 |
0.2467 | 4.29 | 1500 | 0.4755 | 31.6802 |
0.1904 | 5.71 | 2000 | 0.4681 | 30.5907 |
0.118 | 7.14 | 2500 | 0.4776 | 30.9368 |
0.0941 | 8.57 | 3000 | 0.4831 | 30.7297 |
0.0771 | 10.0 | 3500 | 0.4912 | 31.1014 |
0.0536 | 11.43 | 4000 | 0.5043 | 31.2319 |
0.0502 | 12.86 | 4500 | 0.5113 | 31.2404 |
0.0418 | 14.29 | 5000 | 0.5129 | 31.1298 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2
- Downloads last month
- 2
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
the model is not deployed on the HF Inference API.
Evaluation results
- WER on google/fleurstest set self-reported30.800
- CER on google/fleurstest set self-reported11.290
- WER on mozilla-foundation/common_voice_11_0test set self-reported32.490
- CER on mozilla-foundation/common_voice_11_0test set self-reported12.250