Modelpublic

seoulsky-field/KoBART_jeju_dialect

sourceHugging Facemitupdated 3y agoView on Hugging Face

0likes5downloads

Model Card

KoBARTbasev2-trial2

This model is a fine-tuned version of gogamza/kobart-base-v2 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.1820

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: 0.0005
—trainbatchsize: 64
—evalbatchsize: 64
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_steps: 20
—num_epochs: 5
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.3889	0.11	50	0.5425
0.5339	0.22	100	0.4328
0.4609	0.32	150	0.4180
0.4631	0.43	200	0.4167
0.4065	0.54	250	0.3775
0.3898	0.65	300	0.3539
0.3637	0.76	350	0.3389
0.3347	0.87	400	0.3275
0.3428	0.97	450	0.3087
0.2871	1.08	500	0.3189
0.2843	1.19	550	0.3016
0.2685	1.3	600	0.2954
0.2603	1.41	650	0.2860
0.2636	1.52	700	0.2804
0.2586	1.62	750	0.2821
0.2485	1.73	800	0.2674
0.2483	1.84	850	0.2662
0.2322	1.95	900	0.2525
0.2052	2.06	950	0.2634
0.1838	2.16	1000	0.2472
0.1859	2.27	1050	0.2432
0.1887	2.38	1100	0.2392
0.1756	2.49	1150	0.2314
0.1697	2.6	1200	0.2332
0.1741	2.71	1250	0.2257
0.1665	2.81	1300	0.2204
0.1655	2.92	1350	0.2097
0.1539	3.03	1400	0.2141
0.126	3.14	1450	0.2129
0.1241	3.25	1500	0.2068
0.1266	3.35	1550	0.1999
0.1161	3.46	1600	0.1996
0.1183	3.57	1650	0.1943
0.1123	3.68	1700	0.1914
0.1096	3.79	1750	0.1881
0.1089	3.9	1800	0.1835
0.1096	4.0	1850	0.1803
0.0857	4.11	1900	0.1873
0.0833	4.22	1950	0.1857
0.0791	4.33	2000	0.1871
0.0825	4.44	2050	0.1852
0.0813	4.55	2100	0.1834
0.0806	4.65	2150	0.1830
0.0805	4.76	2200	0.1822
0.0786	4.87	2250	0.1820
0.0775	4.98	2300	0.1820

Framework versions

—Transformers 4.36.0
—Pytorch 2.0.1+cu117
—Datasets 2.15.0
—Tokenizers 0.15.0