deberta-v3-large-survey-topicality-rater-all

This model is a fine-tuned version of microsoft/deberta-v3-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3518
• Krippendorff: 0.9652
• Spearman: 0.9370
• Absolute Agreement: 0.8894
• Agreement Within One: 0.9620

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: 6e-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Krippendorff	Spearman	Absolute Agreement	Agreement Within One
No log	1.0	55	2.0618	-0.7525	-0.0087	0.0556	1.0
No log	2.0	110	2.0354	-0.7525	-0.0087	0.0556	1.0
No log	3.0	165	2.0026	-0.5886	-0.1318	0.0694	0.9583
No log	4.0	220	1.7124	-0.2309	nan	0.3889	0.7917
No log	5.0	275	1.6319	-0.2309	nan	0.3889	0.7917
No log	6.0	330	1.5901	-0.2309	nan	0.3889	0.7917
No log	7.0	385	1.5246	-0.2309	nan	0.3889	0.7917
No log	8.0	440	1.4971	-0.2309	nan	0.3889	0.7917
No log	9.0	495	1.4743	-0.2326	-0.0674	0.3889	0.7917
1.542	10.0	550	1.4066	-0.1165	0.2492	0.4722	0.8194
1.542	11.0	605	1.5851	-0.2309	nan	0.3889	0.7917
1.542	12.0	660	1.3221	0.3168	0.3679	0.5694	0.8611
1.542	13.0	715	1.3178	0.0006	0.3335	0.5833	0.8472
1.542	14.0	770	1.3719	0.0210	0.4729	0.5556	0.8472
1.542	15.0	825	1.3791	0.0031	0.4388	0.5417	0.8333
1.542	16.0	880	1.2066	0.3881	0.5206	0.6389	0.875
1.542	17.0	935	1.2682	0.3882	0.5709	0.5833	0.875
1.542	18.0	990	1.1614	0.3845	0.4319	0.6111	0.875
0.8142	19.0	1045	1.1435	0.3919	0.5426	0.6389	0.875
0.8142	20.0	1100	1.0325	0.3527	0.5206	0.6528	0.875

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1
• Datasets 2.10.1
• Tokenizers 0.12.1