seq2seq_model_printed

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

• Loss: 1.5578
• Cer: 0.2138

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: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 35
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
4.6751	1.0	244	3.5585	1.0652
3.3624	2.0	488	2.7206	0.7373
2.9125	3.0	732	1.9523	0.5418
2.5734	4.0	976	2.0470	0.7149
2.4143	5.0	1220	1.4577	0.3585
2.2459	6.0	1464	1.5741	0.4501
2.1514	7.0	1708	1.2625	0.3218
2.0268	8.0	1952	1.4816	0.3625
1.9513	9.0	2196	1.6085	0.3259
1.7768	10.0	2440	1.4458	0.4033
1.7156	11.0	2684	1.4845	0.3055
1.5976	12.0	2928	1.6491	0.3503
1.4664	13.0	3172	1.3220	0.3381
1.3276	14.0	3416	1.4486	0.3503
1.2354	15.0	3660	1.6394	0.3177
1.1072	16.0	3904	1.5189	0.3035
0.9209	17.0	4148	1.3820	0.2485
0.7356	18.0	4392	1.4799	0.2607
0.6336	19.0	4636	1.5075	0.2220
0.5035	20.0	4880	1.5413	0.2179
0.406	21.0	5124	1.5602	0.2464
0.3294	22.0	5368	1.4495	0.2159
0.2515	23.0	5612	1.5809	0.2240
0.2207	24.0	5856	1.5188	0.2281
0.1689	25.0	6100	1.5153	0.2118
0.1426	26.0	6344	1.5616	0.2118
0.1142	27.0	6588	1.7044	0.2179
0.0785	28.0	6832	1.6267	0.2281
0.0751	29.0	7076	1.6769	0.2159
0.0507	30.0	7320	1.7316	0.2342
0.0388	31.0	7564	1.5750	0.2220
0.0264	32.0	7808	1.7028	0.2159
0.021	33.0	8052	1.6861	0.2322
0.0195	34.0	8296	1.7154	0.2077
0.0167	35.0	8540	1.5578	0.2138

Framework versions

• Transformers 4.42.4
• Pytorch 2.1.0a0+b5021ba
• Datasets 2.20.0
• Tokenizers 0.19.1