deberta-v3-large-survey-topicality-rater-half-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: 2.3050
• Krippendorff: -0.4955
• Spearman: -0.1556
• Absolute Agreement: 0.0433
• Agreement Within One: 0.9459

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	52	1.9069	0.0782	0.2354	0.5556	0.8889
No log	2.0	104	1.9047	0.0782	0.2354	0.5556	0.8889
No log	3.0	156	1.9737	-0.0129	-0.1799	0.4722	0.8611
No log	4.0	208	2.2560	-0.2433	nan	0.2778	0.8333
No log	5.0	260	2.4201	-0.2433	nan	0.2778	0.8333
No log	6.0	312	2.3538	-0.2433	nan	0.2778	0.8333
No log	7.0	364	2.3367	-0.2433	nan	0.2778	0.8333
No log	8.0	416	2.3443	-0.2433	nan	0.2778	0.8333
No log	9.0	468	2.0372	-0.2469	-0.2354	0.25	0.8333
1.2964	10.0	520	2.2058	-0.2469	-0.2354	0.25	0.8333
1.2964	11.0	572	2.2627	-0.2433	nan	0.2778	0.8333
1.2964	12.0	624	2.1503	-0.1078	0.0520	0.2778	0.8333
1.2964	13.0	676	1.8854	-0.1076	0.2098	0.2778	0.8333
1.2964	14.0	728	2.1130	-0.1078	0.0520	0.2778	0.8333
1.2964	15.0	780	1.8515	0.0472	0.0751	0.2778	0.8611
1.2964	16.0	832	1.6744	-0.0472	0.0208	0.3611	0.8611
1.2964	17.0	884	2.1802	-0.1746	-0.0231	0.3889	0.8333
1.2964	18.0	936	1.7570	-0.1736	0.0034	0.3889	0.8333
1.2964	19.0	988	2.1582	-0.1023	-0.0053	0.3611	0.8333
0.6766	20.0	1040	2.1745	-0.0290	0.2050	0.3611	0.8333

Framework versions

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