Modelpublic

domenicrosati/deberta-v3-large-survey-new_fact_related_passage-rater

sourceHugging Facemitupdated 3y agoView on Hugging Face

0likes17downloads

Model Card

deberta-v3-large-survey-newfactrelated_passage-rater

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: 0.2316
—Krippendorff: 0.9171
—Spearman: 0.9146
—Absolute Agreement: 0.9372
—Agreement Within One: 0.9874

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	50	1.9025	-0.0055	nan	0.0833	0.6389
No log	2.0	100	1.9012	0.1665	0.2932	0.1528	0.8611
No log	3.0	150	1.9013	-0.5318	nan	0.1806	1.0
No log	4.0	200	1.9030	-0.5318	nan	0.1806	1.0
No log	5.0	250	1.9633	-0.3932	-0.1362	0.1528	0.8889
No log	6.0	300	2.1312	-0.3716	-0.3587	0.0556	0.7361
No log	7.0	350	2.1081	-0.2882	-0.1942	0.1111	0.8056
No log	8.0	400	2.0798	-0.2532	-0.1339	0.1806	0.875
No log	9.0	450	2.0693	-0.2269	-0.1385	0.25	0.5694
1.6529	10.0	500	2.0939	-0.2456	-0.2515	0.1667	0.6667
1.6529	11.0	550	2.1227	-0.0758	0.0820	0.2639	0.6111
1.6529	12.0	600	2.0465	-0.1802	0.0413	0.3056	0.5694
1.6529	13.0	650	2.0901	-0.1164	-0.0485	0.2639	0.625
1.6529	14.0	700	2.1685	0.2526	0.3009	0.2778	0.9028
1.6529	15.0	750	2.1592	0.2603	0.2661	0.3194	0.7083
1.6529	16.0	800	2.3096	0.2225	0.2235	0.2917	0.7778
1.6529	17.0	850	2.2947	0.2268	0.2128	0.3056	0.7222
1.6529	18.0	900	2.3396	0.0808	0.1413	0.3056	0.6528
1.6529	19.0	950	2.6040	-0.0037	0.1140	0.3056	0.6389
0.7915	20.0	1000	2.5205	0.2986	0.2466	0.1944	0.8194

Framework versions

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