trocr-base-handwritten
This model is a fine-tuned version of microsoft/trocr-base-handwritten on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.4825
- Cer: 0.1647
- Wer: 0.3687
- Penalized Wer: 0.3822
- Combined: 0.2735
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: 32
- seed: 2026
- optimizer: Use OptimizerNames.ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.05
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer | Wer | Penalized Wer | Combined |
|---|---|---|---|---|---|---|---|
| 8.3684 | 1.0 | 116 | 7.7234 | 0.5314 | 0.8450 | 0.8862 | 0.7088 |
| 6.1191 | 2.0 | 232 | 4.0425 | 0.4222 | 0.7276 | 0.7603 | 0.5912 |
| 3.5055 | 3.0 | 348 | 3.0009 | 0.3198 | 0.6157 | 0.6434 | 0.4816 |
| 2.7962 | 4.0 | 464 | 2.6273 | 0.2876 | 0.5739 | 0.6002 | 0.4439 |
| 2.5346 | 5.0 | 580 | 2.4117 | 0.2671 | 0.5368 | 0.5581 | 0.4126 |
| 2.3641 | 6.0 | 696 | 2.2578 | 0.2518 | 0.5180 | 0.5396 | 0.3957 |
| 2.2352 | 7.0 | 812 | 2.1378 | 0.2388 | 0.4889 | 0.5075 | 0.3731 |
| 2.1021 | 8.0 | 928 | 2.0344 | 0.2263 | 0.4705 | 0.4882 | 0.3572 |
| 1.9916 | 9.0 | 1044 | 1.9547 | 0.2186 | 0.4551 | 0.4724 | 0.3455 |
| 1.8868 | 10.0 | 1160 | 1.8899 | 0.2096 | 0.4447 | 0.4627 | 0.3362 |
| 1.8350 | 11.0 | 1276 | 1.8264 | 0.2030 | 0.4330 | 0.4504 | 0.3267 |
| 1.7901 | 12.0 | 1392 | 1.7776 | 0.1966 | 0.4228 | 0.4386 | 0.3176 |
| 1.7575 | 13.0 | 1508 | 1.7373 | 0.1918 | 0.4122 | 0.4282 | 0.3100 |
| 1.6409 | 14.0 | 1624 | 1.6968 | 0.1866 | 0.4094 | 0.4261 | 0.3064 |
| 1.6370 | 15.0 | 1740 | 1.6660 | 0.1835 | 0.4066 | 0.4223 | 0.3029 |
| 1.5953 | 16.0 | 1856 | 1.6408 | 0.1806 | 0.4003 | 0.4156 | 0.2981 |
| 1.5737 | 17.0 | 1972 | 1.6150 | 0.1787 | 0.3960 | 0.4111 | 0.2949 |
| 1.5381 | 18.0 | 2088 | 1.5917 | 0.1768 | 0.3925 | 0.4078 | 0.2923 |
| 1.5058 | 19.0 | 2204 | 1.5737 | 0.1740 | 0.3840 | 0.3977 | 0.2859 |
| 1.4760 | 20.0 | 2320 | 1.5564 | 0.1707 | 0.3819 | 0.3954 | 0.2831 |
| 1.4345 | 21.0 | 2436 | 1.5408 | 0.1718 | 0.3786 | 0.3924 | 0.2821 |
| 1.4309 | 22.0 | 2552 | 1.5288 | 0.1694 | 0.3793 | 0.3930 | 0.2812 |
| 1.4147 | 23.0 | 2668 | 1.5168 | 0.1681 | 0.3773 | 0.3922 | 0.2801 |
| 1.3955 | 24.0 | 2784 | 1.5072 | 0.1686 | 0.3743 | 0.3885 | 0.2785 |
| 1.4017 | 25.0 | 2900 | 1.5017 | 0.1661 | 0.3736 | 0.3876 | 0.2769 |
| 1.4114 | 26.0 | 3016 | 1.4926 | 0.1658 | 0.3708 | 0.3847 | 0.2753 |
| 1.4293 | 27.0 | 3132 | 1.4886 | 0.1639 | 0.3702 | 0.3838 | 0.2738 |
| 1.3707 | 28.0 | 3248 | 1.4852 | 0.1657 | 0.3713 | 0.3850 | 0.2753 |
| 1.3794 | 29.0 | 3364 | 1.4832 | 0.1638 | 0.3663 | 0.3797 | 0.2718 |
| 1.3586 | 30.0 | 3480 | 1.4825 | 0.1647 | 0.3687 | 0.3822 | 0.2735 |
Framework versions
- PEFT 0.19.1
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 5.0.0
- Tokenizers 0.22.2