indicwav2vec-hindi-vaani-62066-normalized-alldata-1e-4-epochs-50-FT
This model is a fine-tuned version of ai4bharat/indicwav2vec-hindi on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.3807
- Global Wer: 13.3821
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: 0.0001
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 500
- num_epochs: 50
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Global Wer |
|---|---|---|---|---|
| 7.3382 | 1.0 | 540 | 0.5215 | 36.4671 |
| 1.0430 | 2.0 | 1080 | 0.3714 | 26.9527 |
| 0.8285 | 3.0 | 1620 | 0.3300 | 23.7900 |
| 0.7167 | 4.0 | 2160 | 0.3039 | 21.8976 |
| 0.6457 | 5.0 | 2700 | 0.2928 | 21.4377 |
| 0.5959 | 6.0 | 3240 | 0.2811 | 20.1900 |
| 0.5480 | 7.0 | 3780 | 0.2735 | 19.4587 |
| 0.5170 | 8.0 | 4320 | 0.2678 | 18.6558 |
| 0.4742 | 9.0 | 4860 | 0.2733 | 18.5238 |
| 0.4500 | 10.0 | 5400 | 0.2684 | 17.8302 |
| 0.4202 | 11.0 | 5940 | 0.2599 | 17.4796 |
| 0.3938 | 12.0 | 6480 | 0.2653 | 17.5362 |
| 0.3526 | 13.0 | 7020 | 0.2664 | 17.0084 |
| 0.3307 | 14.0 | 7560 | 0.2741 | 16.4694 |
| 0.3145 | 15.0 | 8100 | 0.2669 | 16.1716 |
| 0.2958 | 16.0 | 8640 | 0.2791 | 15.9115 |
| 0.2785 | 17.0 | 9180 | 0.2829 | 15.9228 |
| 0.2706 | 18.0 | 9720 | 0.2933 | 15.7305 |
| 0.2493 | 19.0 | 10260 | 0.2922 | 15.4328 |
| 0.2359 | 20.0 | 10800 | 0.2917 | 15.2443 |
| 0.2339 | 21.0 | 11340 | 0.2952 | 15.2857 |
| 0.2159 | 22.0 | 11880 | 0.3037 | 15.1689 |
| 0.2070 | 23.0 | 12420 | 0.3112 | 14.7655 |
| 0.1997 | 24.0 | 12960 | 0.3059 | 14.6298 |
| 0.1798 | 25.0 | 13500 | 0.3204 | 14.5695 |
| 0.1743 | 26.0 | 14040 | 0.3104 | 14.4527 |
| 0.1671 | 27.0 | 14580 | 0.3193 | 14.4602 |
| 0.1581 | 28.0 | 15120 | 0.3536 | 14.0116 |
| 0.1530 | 29.0 | 15660 | 0.3416 | 14.2604 |
| 0.1505 | 30.0 | 16200 | 0.3307 | 14.2076 |
| 0.1435 | 31.0 | 16740 | 0.3550 | 13.7515 |
| 0.1385 | 32.0 | 17280 | 0.3532 | 13.9551 |
| 0.1330 | 33.0 | 17820 | 0.3652 | 13.8872 |
| 0.1289 | 34.0 | 18360 | 0.3620 | 13.6686 |
| 0.1244 | 35.0 | 18900 | 0.3632 | 13.7364 |
| 0.1224 | 36.0 | 19440 | 0.3707 | 13.5404 |
| 0.1184 | 37.0 | 19980 | 0.3945 | 13.4160 |
| 0.1098 | 38.0 | 20520 | 0.3729 | 13.3293 |
| 0.1116 | 39.0 | 21060 | 0.3749 | 13.4952 |
| 0.1071 | 40.0 | 21600 | 0.3832 | 13.5216 |
| 0.1034 | 41.0 | 22140 | 0.3807 | 13.3821 |
Framework versions
- Transformers 5.13.0
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.22.2