Nougat_CH_20k

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

• Loss: 0.1668

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 6
• total_train_batch_size: 48
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss
10.4291	1.0	334	1.5553
4.8127	2.0	668	0.8458
2.9138	3.0	1002	0.4572
1.8249	4.0	1336	0.4366
1.1195	5.0	1670	0.2577
1.0455	6.0	2004	0.1916
0.9011	7.0	2338	0.2527
0.5274	8.0	2672	0.2161
0.4131	9.0	3006	0.1683
0.2793	10.0	3340	0.1742
0.2865	11.0	3674	0.1788
0.2515	12.0	4008	0.1852
0.1991	13.0	4342	0.1589
0.1869	14.0	4676	0.1569
0.1761	15.0	5010	0.2064
0.1546	16.0	5344	0.1736
0.1553	17.0	5678	0.1631
0.171	18.0	6012	0.1642
0.1502	19.0	6346	0.1668

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 4.1.1
• Tokenizers 0.21.0