deberta-v3-large-survey-main_passage_old_facts-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: 0.7151
• Krippendorff: 0.8837
• Spearman: 0.9031
• Absolute Agreement: 0.7615
• Agreement Within One: 0.9320

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	1.9126	-0.4378	0.1864	0.1528	1.0
No log	2.0	110	1.9170	-0.5514	nan	0.1389	1.0
No log	3.0	165	1.9480	-0.5514	nan	0.1389	1.0
No log	4.0	220	2.1324	-0.5514	nan	0.1389	1.0
No log	5.0	275	2.0791	-0.5514	nan	0.1389	1.0
No log	6.0	330	2.0516	-0.5514	nan	0.1389	1.0
No log	7.0	385	2.0073	-0.4410	-0.0554	0.1667	0.9861
No log	8.0	440	2.0510	0.0412	0.2417	0.1806	0.9306
No log	9.0	495	1.8581	0.3416	0.3139	0.2083	0.75
1.6735	10.0	550	1.9075	0.3730	0.3806	0.2083	0.8056
1.6735	11.0	605	1.8024	0.5377	0.5418	0.2222	0.7917
1.6735	12.0	660	1.8039	0.5034	0.4816	0.2361	0.7917
1.6735	13.0	715	1.7669	0.5795	0.5456	0.3611	0.875
1.6735	14.0	770	1.7351	0.5956	0.5678	0.375	0.8194
1.6735	15.0	825	1.6817	0.5321	0.4698	0.3333	0.8056
1.6735	16.0	880	1.5989	0.5717	0.5629	0.3889	0.8056
1.6735	17.0	935	1.6268	0.5621	0.5687	0.375	0.7917
1.6735	18.0	990	1.5200	0.5958	0.5848	0.3889	0.8472
1.0164	19.0	1045	1.5787	0.5202	0.5297	0.4444	0.75
1.0164	20.0	1100	1.4346	0.5484	0.5457	0.4722	0.7778

Framework versions

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