convnext_manuscript_iiif

This model is a fine-tuned version of facebook/convnext-base-224-22k on the davanstrien/iiif_manuscripts_label_ge_50 dataset. It achieves the following results on the evaluation set:

• Loss: 5.5856
• F1: 0.0037

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: 0.0002
• train_batch_size: 64
• eval_batch_size: 64
• seed: 1337
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1
6.5753	1.0	2038	6.4121	0.0016
5.9865	2.0	4076	5.9466	0.0021
5.6521	3.0	6114	5.7645	0.0029
5.3123	4.0	8152	5.6890	0.0033
5.0337	5.0	10190	5.6692	0.0034
4.743	6.0	12228	5.5856	0.0037
4.4387	7.0	14266	5.5969	0.0042
4.1422	8.0	16304	5.6711	0.0043
3.8372	9.0	18342	5.6761	0.0044
3.5244	10.0	20380	5.8469	0.0042
3.2321	11.0	22418	5.8774	0.0045
2.9004	12.0	24456	6.1186	0.0047
2.5937	13.0	26494	6.2398	0.0046
2.2983	14.0	28532	6.3732	0.0049
2.0611	15.0	30570	6.5024	0.0045
1.8153	16.0	32608	6.6585	0.0047
1.6075	17.0	34646	6.8333	0.0043
1.4342	18.0	36684	6.9529	0.0044
1.2614	19.0	38722	7.1129	0.0046
1.1463	20.0	40760	7.1977	0.0039
1.0387	21.0	42798	7.2700	0.0044
0.9635	22.0	44836	7.3375	0.0040
0.8872	23.0	46874	7.4003	0.0039
0.8156	24.0	48912	7.4884	0.0039
0.7544	25.0	50950	7.4764	0.0039
0.6893	26.0	52988	7.5153	0.0042
0.6767	27.0	55026	7.5427	0.0043
0.6098	28.0	57064	7.5547	0.0042
0.5871	29.0	59102	7.5533	0.0041
0.5696	30.0	61140	7.5595	0.0041

Framework versions

• Transformers 4.18.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.3
• Tokenizers 0.11.6