TrOCR-SIN-Handwritten

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

• Loss: 2.4199
• Cer: 0.5981

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
1.1305	1.72	50	0.7056	2.0648
0.5468	3.45	100	0.7023	1.9332
0.0732	5.17	150	0.6624	2.2013
0.0995	6.9	200	0.6756	2.2015
0.1715	8.62	250	0.6602	2.0091
0.1086	10.34	300	0.6672	2.2239
0.0798	12.07	350	0.6839	2.0926
0.0194	13.79	400	0.6926	2.2650
0.0453	15.52	450	0.7223	2.4613
0.0579	17.24	500	0.6931	2.3523
0.0301	18.97	550	0.6215	2.1677
0.0215	20.69	600	0.6602	2.2610
0.0112	22.41	650	0.6415	2.2070
0.0048	24.14	700	0.6363	2.2219
0.0117	25.86	750	0.6609	2.2341
0.0034	27.59	800	0.6829	2.4177
0.0184	29.31	850	0.6826	2.3418
0.0108	31.03	900	0.6682	2.3715
0.0152	32.76	950	0.6542	2.2079
0.0019	34.48	1000	0.6365	2.2173
0.0009	36.21	1050	0.6574	2.4185
0.0013	37.93	1100	0.6520	2.3515
0.0005	39.66	1150	0.6081	2.2393
0.0003	41.38	1200	0.6632	2.7642
0.0003	43.1	1250	0.6313	2.2789
0.0002	44.83	1300	0.6360	2.3862
0.0004	46.55	1350	0.6228	2.3372
0.0001	48.28	1400	0.6542	2.3822
0.0003	50.0	1450	2.3609	0.6480
0.0015	51.72	1500	2.6877	0.6520
0.0352	53.45	1550	3.3146	0.6826
0.0085	55.17	1600	2.8430	0.6901
0.0004	56.9	1650	3.0267	0.6654
0.0006	58.62	1700	2.3768	0.6589
0.0013	60.34	1750	2.4242	0.6717
0.0033	62.07	1800	2.5133	0.6512
0.0002	63.79	1850	2.4018	0.6604
0.0006	65.52	1900	2.8438	0.6679
0.0006	67.24	1950	2.4073	0.6482
0.0001	68.97	2000	2.7375	0.6510
0.0002	70.69	2050	2.4250	0.6146
0.0001	72.41	2100	2.7045	0.6604
0.0001	74.14	2150	3.3714	0.6597
0.0001	75.86	2200	2.9585	0.6936
0.0	77.59	2250	2.5378	0.6131
0.0136	79.31	2300	2.6351	0.6146
0.0	81.03	2350	2.5288	0.6116
0.0001	82.76	2400	2.4601	0.6338
0.0002	84.48	2450	2.4646	0.6248
0.0	86.21	2500	2.8099	0.6542
0.0	87.93	2550	2.7981	0.6781
0.0	89.66	2600	2.4607	0.6200
0.0	91.38	2650	2.5363	0.6612
0.0096	93.1	2700	2.4875	0.6113
0.0	94.83	2750	2.4574	0.6268
0.0	96.55	2800	2.4199	0.5981
0.0	98.28	2850	2.4452	0.6181
0.0	100.0	2900	2.4369	0.6036

Framework versions

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