Modelpublic

domenicrosati/deberta-v3-large-survey-fluency-rater-all

sourceHugging Facemitupdated 3y agoView on Hugging Face

0likes4downloads

Model Card

deberta-v3-large-survey-fluency-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: 0.7178
—Krippendorff: 0.9045
—Spearman: 0.8801
—Absolute Agreement: 0.7396
—Agreement Within One: 0.9608

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 20
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Krippendorff	Spearman	Absolute Agreement	Agreement Within One
No log	1.0	55	2.0214	-0.6902	0.2469	0.0139	1.0
No log	2.0	110	2.0118	-0.6398	0.1978	0.0278	1.0
No log	3.0	165	2.0147	-0.6585	-0.0351	0.0278	0.9583
No log	4.0	220	2.0186	-0.6499	-0.1548	0.0139	0.9306
No log	5.0	275	1.7283	-0.2880	nan	0.2917	0.7222
No log	6.0	330	1.7079	-0.2454	0.1527	0.3056	0.7222
No log	7.0	385	1.6519	-0.0919	-0.0671	0.3889	0.7639
No log	8.0	440	1.6403	0.1140	-0.0254	0.4028	0.7778
No log	9.0	495	1.6711	-0.1939	-0.0213	0.3333	0.75
1.5874	10.0	550	1.6714	-0.0236	0.1527	0.3056	0.7361
1.5874	11.0	605	1.5761	-0.0843	-0.1552	0.4028	0.7639
1.5874	12.0	660	1.4782	0.2760	0.1495	0.4444	0.7917
1.5874	13.0	715	1.4679	0.1322	-0.0231	0.4028	0.7778
1.5874	14.0	770	1.4314	0.1322	-0.0231	0.4028	0.7778
1.5874	15.0	825	1.3915	0.3094	0.2541	0.4583	0.7917
1.5874	16.0	880	1.3383	0.3143	0.3726	0.4583	0.7917
1.5874	17.0	935	1.2967	0.1861	0.3025	0.4583	0.7778
1.5874	18.0	990	1.3368	0.1848	0.4152	0.4722	0.8056
0.9758	19.0	1045	1.2081	0.2376	0.4096	0.4861	0.8056
0.9758	20.0	1100	1.1952	0.2625	0.4872	0.5	0.8056

Framework versions

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