deberta-v3-large-ft-icar-a-v0.8

This model is a fine-tuned version of microsoft/deberta-v3-large on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9476
• Accuracy: 0.9280
• Precision: 0.9234
• Recall: 0.9186
• F1: 0.9202

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: 3e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 3
• optimizer: Use OptimizerNames.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: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
2.5424	1.0	871	0.7067	0.8025	0.6248	0.7323	0.6636
1.8986	2.0	1742	0.7439	0.8867	0.9022	0.8627	0.8742
1.5703	3.0	2613	0.6026	0.9081	0.8968	0.8992	0.8973
1.2576	4.0	3484	0.7480	0.9066	0.9079	0.8848	0.8943
1.0473	5.0	4355	0.6990	0.9204	0.9226	0.9029	0.9110
0.8209	6.0	5226	0.7821	0.9127	0.8990	0.9073	0.9019
0.6306	7.0	6097	0.8300	0.9219	0.9267	0.8975	0.9088
0.5627	8.0	6968	0.8917	0.9081	0.9080	0.8902	0.8964
0.4382	9.0	7839	0.7816	0.9234	0.9259	0.9085	0.9155
0.3512	10.0	8710	0.8973	0.9142	0.9140	0.8919	0.9003
0.2517	11.0	9581	0.8886	0.9173	0.9141	0.9050	0.9081
0.2207	12.0	10452	0.8375	0.9234	0.9129	0.9170	0.9129
0.1781	13.0	11323	1.1453	0.9005	0.8935	0.9103	0.8967
0.2042	14.0	12194	0.9084	0.9173	0.9101	0.9141	0.9110
0.1297	15.0	13065	1.2144	0.8974	0.8908	0.9072	0.8911
0.1313	16.0	13936	0.9528	0.9219	0.9196	0.9038	0.9102
0.1391	17.0	14807	0.9476	0.9280	0.9234	0.9186	0.9202
0.1283	18.0	15678	0.9934	0.9188	0.9141	0.9077	0.9097
0.1222	19.0	16549	1.0909	0.9096	0.8988	0.8995	0.8959
0.053	20.0	17420	1.0880	0.9173	0.9103	0.9014	0.9035

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.21.2