twitter-roberta-base-efl-hateval

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-2021-124m on the HatEval dataset. It achieves the following results on the evaluation set:

• Accuracy: 0.7913
• F1: 0.7899
• Loss: 0.3683

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: 1e-06
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss
0.5392	1.0	211	0.7	0.6999	0.4048
0.3725	2.0	422	0.759	0.7584	0.3489
0.3158	3.0	633	0.7613	0.7570	0.3287
0.289	4.0	844	0.769	0.7684	0.3307
0.2716	5.0	1055	0.7767	0.7750	0.3241
0.2575	6.0	1266	0.7787	0.7782	0.3272
0.2441	7.0	1477	0.7783	0.7776	0.3258
0.2363	8.0	1688	0.7777	0.7773	0.3316
0.2262	9.0	1899	0.7843	0.7815	0.3150
0.2191	10.0	2110	0.7813	0.7802	0.3241
0.2112	11.0	2321	0.7867	0.7860	0.3276
0.2047	12.0	2532	0.7897	0.7886	0.3266
0.1973	13.0	2743	0.7893	0.7884	0.3299
0.1897	14.0	2954	0.792	0.7907	0.3301
0.1862	15.0	3165	0.794	0.7925	0.3283
0.1802	16.0	3376	0.7907	0.7903	0.3465
0.1764	17.0	3587	0.7937	0.7922	0.3393
0.1693	18.0	3798	0.7903	0.7893	0.3494
0.1666	19.0	4009	0.7943	0.7930	0.3486
0.1631	20.0	4220	0.7927	0.7917	0.3516
0.1609	21.0	4431	0.7907	0.7893	0.3537
0.1581	22.0	4642	0.7913	0.7902	0.3586
0.1548	23.0	4853	0.789	0.7884	0.3698
0.1535	24.0	5064	0.7893	0.7880	0.3622
0.1522	25.0	5275	0.7923	0.7909	0.3625
0.15	26.0	5486	0.7913	0.7899	0.3632
0.1479	27.0	5697	0.792	0.7909	0.3677
0.1441	28.0	5908	0.792	0.7909	0.3715
0.145	29.0	6119	0.792	0.7906	0.3681
0.1432	30.0	6330	0.7913	0.7899	0.3683

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 2.0.0
• Tokenizers 0.11.6