Sans1807/APNR-Braincore-V2

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: 0.0856
• Cer: 0.0098
• Wer: 0.0573

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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.6623	1.0	217	0.4722	0.0574	0.2340
0.4122	2.0	434	0.3248	0.0378	0.1585
0.1077	3.0	651	0.0898	0.0132	0.0722
0.047	4.0	868	0.0848	0.0114	0.0614
0.0304	5.0	1085	0.0836	0.0122	0.0634
0.0224	6.0	1302	0.0891	0.0104	0.0566
0.0154	7.0	1519	0.0873	0.0107	0.0587
0.0137	8.0	1736	0.0852	0.0102	0.0560
0.0121	9.0	1953	0.0883	0.0107	0.0634
0.0095	10.0	2170	0.0829	0.0092	0.0526
0.0068	11.0	2387	0.0851	0.0091	0.0519
0.0075	12.0	2604	0.0831	0.0102	0.0600
0.0055	13.0	2821	0.0824	0.0098	0.0580
0.0048	14.0	3038	0.0821	0.0099	0.0587
0.0023	15.0	3255	0.0873	0.0096	0.0553
0.0018	16.0	3472	0.0835	0.0102	0.0593
0.0016	17.0	3689	0.0888	0.0100	0.0600
0.0034	18.0	3906	0.0853	0.0094	0.0553
0.001	19.0	4123	0.0857	0.0096	0.0566
0.0013	20.0	4340	0.0856	0.0098	0.0573

Framework versions

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