Whisper_Medium_Hi_Vyapar_V3
This model is a fine-tuned version of openai/whisper-medium on the Vyapar_Calling_Data_500_hours dataset. It achieves the following results on the evaluation set:
- Loss: 0.5455
- Wer: 31.9144
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: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Use adafactor and the args are: No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 20000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.816 | 0.0259 | 1000 | 0.8171 | 43.2510 |
| 0.7255 | 0.0519 | 2000 | 0.7386 | 51.1530 |
| 0.7224 | 0.0778 | 3000 | 0.6966 | 39.1227 |
| 0.7127 | 0.1038 | 4000 | 0.6729 | 37.4067 |
| 0.6472 | 0.1297 | 5000 | 0.6562 | 36.8165 |
| 0.6547 | 0.1557 | 6000 | 0.6404 | 36.5441 |
| 0.6207 | 0.1816 | 7000 | 0.6248 | 36.8320 |
| 0.6445 | 0.2076 | 8000 | 0.6162 | 35.6743 |
| 0.6371 | 0.2335 | 9000 | 0.6057 | 34.4885 |
| 0.5909 | 0.2594 | 10000 | 0.5983 | 33.8377 |
| 0.6115 | 0.2854 | 11000 | 0.5895 | 34.3792 |
| 0.6298 | 0.3113 | 12000 | 0.5834 | 33.9399 |
| 0.6091 | 0.3373 | 13000 | 0.5757 | 33.0228 |
| 0.5421 | 0.3632 | 14000 | 0.5702 | 32.8652 |
| 0.6022 | 0.3892 | 15000 | 0.5654 | 33.0867 |
| 0.5559 | 0.4151 | 16000 | 0.5596 | 32.6704 |
| 0.5314 | 0.4410 | 17000 | 0.5544 | 32.3119 |
| 0.5714 | 0.4670 | 18000 | 0.5503 | 32.2174 |
| 0.5171 | 0.4929 | 19000 | 0.5471 | 31.9557 |
| 0.5268 | 0.5189 | 20000 | 0.5455 | 31.9144 |
Framework versions
- Transformers 4.48.0
- Pytorch 2.5.1+cu124
- Datasets 3.2.0
- Tokenizers 0.21.0