Whisper Small Odia 10k steps
This model is a fine-tuned version of openai/whisper-small on the Ranjit/or_in_dataset dataset. It achieves the following results on the evaluation set:
- Loss: 0.2806
- Wer: 19.7720
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-06
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.0384 | 0.49 | 1000 | 0.1349 | 40.3740 |
0.0175 | 0.98 | 2000 | 0.1601 | 22.6468 |
0.0091 | 1.46 | 3000 | 0.1817 | 23.1515 |
0.0082 | 1.95 | 4000 | 0.2125 | 23.9139 |
0.0048 | 2.44 | 5000 | 0.2110 | 20.2522 |
0.0037 | 2.93 | 6000 | 0.2270 | 21.4855 |
0.0017 | 3.42 | 7000 | 0.2534 | 20.2399 |
0.0018 | 3.9 | 8000 | 0.2706 | 20.7277 |
0.0005 | 4.39 | 9000 | 0.2806 | 19.7720 |
0.0006 | 4.88 | 10000 | 0.2929 | 20.9747 |
- Downloads last month
- 3
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Space using Ranjit/Whisper_Small_Odia_10k_steps 1
Evaluation results
- Wer on Ranjit/or_in_datasettest set self-reported19.772