levit-192_finetuned_on_unlabelled_IA_with_snorkel_labels

This model is a fine-tuned version of facebook/levit-192 on the None dataset. It achieves the following results on the evaluation set:

• Loss: nan
• Precision: 0.9836
• Recall: 0.9822
• F1: 0.9829
• Accuracy: 0.9873

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: 128
• eval_batch_size: 256
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	253	nan	0.9743	0.9791	0.9766	0.9826
0.0557	2.0	506	nan	0.9829	0.9801	0.9815	0.9863
0.0557	3.0	759	nan	0.9836	0.9822	0.9829	0.9873
0.0543	4.0	1012	nan	0.9839	0.9775	0.9807	0.9858
0.0543	5.0	1265	nan	0.9616	0.9727	0.9670	0.9752
0.0457	6.0	1518	nan	0.9563	0.9699	0.9629	0.9720
0.0457	7.0	1771	nan	0.9822	0.9808	0.9815	0.9863
0.0418	8.0	2024	nan	0.9735	0.9769	0.9752	0.9815
0.0418	9.0	2277	nan	0.9832	0.9811	0.9822	0.9868
0.0396	10.0	2530	nan	0.9843	0.9815	0.9829	0.9873

Framework versions

• Transformers 4.22.2
• Pytorch 1.12.1+cu113
• Datasets 2.5.2
• Tokenizers 0.12.1