TrOCR-SIN-DeiT-Handwritten-Beam10

This model is a fine-tuned version of kavg/TrOCR-SIN-DeiT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.2754
• Cer: 0.5246

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: 5e-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
• training_steps: 2400
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
0.9957	1.75	100	0.6176	1.6796
0.0678	3.51	200	0.5996	1.7777
0.1315	5.26	300	0.6794	2.1444
0.0668	7.02	400	0.6363	2.0162
0.0656	8.77	500	0.6046	1.9573
0.0612	10.53	600	0.6330	1.9388
0.0454	12.28	700	0.6679	3.0649
0.004	14.04	800	0.5814	2.0252
0.0034	15.79	900	0.5492	2.0399
0.0336	17.54	1000	0.6041	2.9769
0.0135	19.3	1100	0.5742	1.9405
0.0012	21.05	1200	0.5959	2.5722
0.0143	22.81	1300	0.5527	2.0862
0.0018	24.56	1400	0.5764	2.4146
0.0064	26.32	1500	0.5647	2.0710
0.0006	28.07	1600	0.5472	2.1849
0.0004	29.82	1700	0.5547	2.4497
0.0001	31.58	1800	0.5430	2.0830
0.0215	33.33	1900	0.5560	2.5979
0.0	35.09	2000	0.5525	2.4792
0.0	36.84	2100	0.5428	2.4779
0.0	38.6	2200	0.5438	2.7873
0.0	40.35	2300	0.5552	2.9236
0.0	42.11	2400	0.5246	2.2754

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.1