deberta-v3-large-survey-main_passage_consistency-rater-all-gpt4

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: 0.2267
• Krippendorff: 0.9473
• Spearman: 0.9280
• Absolute Agreement: 0.9286
• Agreement Within One: 0.9896

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	1.9958	-0.2750	-0.1177	0.2361	0.9583
No log	2.0	110	1.9825	-0.3122	-0.1408	0.2222	0.9583
No log	3.0	165	1.8124	-0.2303	nan	0.375	0.8194
No log	4.0	220	2.2006	-0.2303	nan	0.375	0.8194
No log	5.0	275	2.0086	-0.2303	nan	0.375	0.8194
No log	6.0	330	1.9755	-0.2303	nan	0.375	0.8194
No log	7.0	385	1.7607	-0.2303	nan	0.375	0.8194
No log	8.0	440	1.7119	-0.2303	nan	0.375	0.8194
No log	9.0	495	1.5061	-0.2303	nan	0.375	0.8194
1.1298	10.0	550	1.5980	-0.2303	nan	0.375	0.8194
1.1298	11.0	605	1.3121	0.1846	0.2912	0.375	0.8611
1.1298	12.0	660	1.3874	0.1465	0.4674	0.4167	0.8472
1.1298	13.0	715	1.2683	0.2376	0.3666	0.5139	0.9028
1.1298	14.0	770	1.2002	0.1714	0.3920	0.5278	0.9028
1.1298	15.0	825	1.2179	0.5302	0.4983	0.5417	0.9583
1.1298	16.0	880	1.2864	0.3380	0.5633	0.5556	0.8889
1.1298	17.0	935	1.1488	0.5258	0.5874	0.6111	0.9167
1.1298	18.0	990	1.1360	0.5455	0.5015	0.5556	0.9444
0.3876	19.0	1045	1.0601	0.5423	0.4674	0.5833	0.9306
0.3876	20.0	1100	1.0164	0.5400	0.4590	0.5694	0.9444

Framework versions

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