Sans1807/APNR-Braincore-V1

microsoft/trocr-base-str

This model is a fine-tuned version of microsoft/trocr-base-str on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1131
• Cer: 0.0102
• Wer: 0.0601

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.8737	1.0	173	0.5194	0.0668	0.2809
0.3674	2.0	346	0.1408	0.0183	0.1015
0.0907	3.0	519	0.1151	0.0150	0.0821
0.0584	4.0	692	0.0987	0.0140	0.0778
0.0374	5.0	865	0.1041	0.0113	0.0711
0.0237	6.0	1038	0.0988	0.0102	0.0626
0.0172	7.0	1211	0.0981	0.0109	0.0643
0.0177	8.0	1384	0.1019	0.0109	0.0643
0.0134	9.0	1557	0.1080	0.0106	0.0643
0.009	10.0	1730	0.1023	0.0094	0.0567
0.0092	11.0	1903	0.1021	0.0090	0.0558
0.0085	12.0	2076	0.1057	0.0105	0.0635
0.006	13.0	2249	0.1055	0.0107	0.0618
0.006	14.0	2422	0.1066	0.0105	0.0618
0.0046	15.0	2595	0.1110	0.0103	0.0618
0.0038	16.0	2768	0.1093	0.0101	0.0601
0.0028	17.0	2941	0.1095	0.0099	0.0575
0.0026	18.0	3114	0.1095	0.0099	0.0575
0.0021	19.0	3287	0.1088	0.0097	0.0584
0.0017	20.0	3460	0.1127	0.0105	0.0601
0.0014	21.0	3633	0.1127	0.0099	0.0575
0.0015	22.0	3806	0.1129	0.0101	0.0592
0.0008	23.0	3979	0.1132	0.0101	0.0592
0.0008	24.0	4152	0.1132	0.0102	0.0601
0.0006	25.0	4325	0.1131	0.0102	0.0601

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 2.17.0
• Tokenizers 0.19.1