TrOCR-SIN

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

• Cer: 0.2343
• Loss: 0.6859

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
• training_steps: 97000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
2.3833	0.36	1000	0.8052	2.3856
2.0754	0.71	2000	0.8113	2.2862
1.9838	1.07	3000	0.8163	2.3398
2.0476	1.42	4000	0.8057	2.1957
2.1071	1.78	5000	0.8085	2.2221
1.8743	2.13	6000	0.8142	2.3726
1.8685	2.49	7000	0.7860	2.2151
1.5893	2.84	8000	0.7558	1.9693
1.3116	3.2	9000	0.7187	1.9843
1.3257	3.55	10000	0.6980	1.9958
1.1866	3.91	11000	0.6662	1.7693
1.0506	4.27	12000	0.6439	1.7593
1.0177	4.62	13000	0.6157	1.6142
0.849	4.98	14000	0.5923	1.5052
0.9062	5.33	15000	0.5733	1.6439
0.9613	5.69	16000	0.5635	1.2713
0.698	6.04	17000	0.5348	1.3989
0.5992	6.4	18000	0.5197	1.5645
0.7429	6.75	19000	0.5132	1.3758
0.5958	7.11	20000	0.4961	1.4102
0.5933	7.47	21000	0.4845	1.2843
0.5802	7.82	22000	0.4760	1.2866
0.5026	8.18	23000	0.4733	1.3028
0.528	8.53	24000	0.4634	1.3796
0.5591	8.89	25000	0.4611	1.2754
0.5399	9.24	26000	0.4645	1.3143
0.5875	9.6	27000	0.4383	1.0949
0.5281	9.95	28000	0.4252	1.0851
0.4801	10.31	29000	0.4065	1.1674
0.4978	10.66	30000	0.3869	1.0382
0.2993	11.02	31000	0.3862	1.0100
0.3392	11.38	32000	0.3657	0.9267
0.4248	11.73	33000	0.3800	0.8588
0.2666	12.09	34000	0.3458	0.9895
0.3525	12.44	35000	0.3649	0.8927
0.259	12.8	36000	0.3272	0.9232
0.2105	13.15	37000	0.3358	0.7679
0.2125	13.51	38000	0.3291	0.8509
0.2744	13.86	39000	0.3367	0.7735
0.1858	14.22	40000	0.3005	0.7237
0.1762	14.58	41000	0.3238	0.7320
0.2107	14.93	42000	0.3035	0.8229
0.1403	15.29	43000	0.2981	0.8188
0.124	15.64	44000	0.3082	0.8104
0.1398	16.0	45000	0.2967	0.8586
0.1207	16.35	46000	0.2838	0.9125
0.1422	16.71	47000	0.3029	0.9329
0.0779	17.06	48000	0.3022	0.7960
0.1103	17.42	49000	0.2900	0.8678
0.1011	17.77	50000	0.2931	0.7747
0.0883	18.13	51000	0.2722	0.7624
0.0468	18.49	52000	0.2826	0.7573
0.0782	18.84	53000	0.2745	0.8906
0.0558	19.2	54000	0.2756	0.7796
0.0792	19.55	55000	0.2799	0.8554
0.063	19.91	56000	0.2916	0.8130
0.0464	20.26	57000	0.2889	0.9519
0.058	20.62	58000	0.2719	0.7782
0.062	20.97	59000	0.2697	0.8140
0.038	21.33	60000	0.2876	0.7488
0.0436	21.69	61000	0.2776	0.7391
0.0363	22.04	62000	0.2730	0.8416
0.0406	22.4	63000	0.2852	0.8974
0.0268	22.75	64000	0.2818	0.9051
0.0143	23.11	65000	0.2733	0.8073
0.0274	23.46	66000	0.2694	0.9573
0.0233	23.82	67000	0.2705	0.8856
0.0177	24.17	68000	0.2701	0.8605
0.0237	24.53	69000	0.2683	0.7962
0.0247	24.88	70000	0.2717	0.8272
0.0135	25.24	71000	0.2737	0.8667
0.0169	25.6	72000	0.2739	0.8405
0.0173	25.95	73000	0.2685	0.7505
0.0168	26.31	74000	0.2682	0.9736
0.0179	26.66	75000	0.2644	0.8753
0.0114	27.02	76000	0.2749	0.8917
0.0121	27.37	77000	0.2733	0.9144
0.0145	27.73	78000	0.2637	0.8889
0.0131	28.08	79000	0.2693	0.9278
0.0078	28.44	80000	0.2669	0.9077
0.0129	28.79	81000	0.2665	0.9218
0.0215	29.15	82000	0.2509	0.7342
0.0291	29.51	83000	0.2573	0.7706
0.0233	29.86	84000	0.2516	0.7602
0.0305	30.22	85000	0.2839	1.0254
0.0424	30.57	86000	0.2725	0.8747
0.0346	30.93	87000	0.2725	0.8864
0.0212	31.28	88000	0.2746	0.8550
0.0266	31.64	89000	0.2834	0.8797
0.0255	31.99	90000	0.2687	0.7178
0.019	32.35	91000	0.2744	0.8784
0.0151	32.71	92000	0.2494	0.6553
0.0243	33.06	93000	0.2531	0.7540
0.04	33.42	94000	0.2526	0.8605
0.0307	33.77	95000	0.2597	0.8507
0.0258	34.13	96000	0.2714	0.7760
0.0245	34.48	97000	0.2343	0.6859

Framework versions

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