Modelpublic

nehaMe123/trocr-medical-v1-optimized

sourceHugging Faceupdated 8mo agoView on Hugging Face

0likes24downloads

Model Card

trocr-medical-v1-optimized

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

—Loss: 0.0500
—Cer: 0.3118

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: 4e-05
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: linear
—num_epochs: 15
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
1.1384	1.0	176	0.8087	0.2478
0.2974	2.0	352	0.2246	0.2590
0.1963	3.0	528	0.1778	0.2712
0.1206	4.0	704	0.1506	0.2616
0.0706	5.0	880	0.1275	0.2052
0.0413	6.0	1056	0.1201	0.2565
0.0291	7.0	1232	0.0987	0.2397
0.0136	8.0	1408	0.0723	0.2641
0.01	9.0	1584	0.0643	0.3012
0.0026	10.0	1760	0.0630	0.2641
0.0049	11.0	1936	0.0618	0.3586
0.0008	12.0	2112	0.0486	0.3118
0.0008	13.0	2288	0.0426	0.3255
0.0004	14.0	2464	0.0487	0.3098
0.0013	15.0	2640	0.0500	0.3118

Framework versions

—Transformers 4.57.1
—Pytorch 2.8.0+cu126
—Datasets 4.4.2
—Tokenizers 0.22.1