ECHR_test_Merged

This model is a fine-tuned version of nlpaueb/legal-bert-base-uncased on the lex_glue dataset. It achieves the following results on the evaluation set:

• Loss: 0.2162
• Macro-f1: 0.5607
• Micro-f1: 0.6726

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: 3e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Macro-f1	Micro-f1
0.2278	0.44	500	0.3196	0.2394	0.4569
0.1891	0.89	1000	0.2827	0.3255	0.5112
0.1803	1.33	1500	0.2603	0.3961	0.5698
0.1676	1.78	2000	0.2590	0.4251	0.6003
0.1635	2.22	2500	0.2489	0.4186	0.6030
0.1784	2.67	3000	0.2445	0.4627	0.6159
0.1556	3.11	3500	0.2398	0.4757	0.6170
0.151	3.56	4000	0.2489	0.4725	0.6163
0.1564	4.0	4500	0.2289	0.5019	0.6416
0.1544	4.44	5000	0.2406	0.5013	0.6408
0.1516	4.89	5500	0.2351	0.5145	0.6510
0.1487	5.33	6000	0.2354	0.5164	0.6394
0.1385	5.78	6500	0.2385	0.5205	0.6486
0.145	6.22	7000	0.2337	0.5197	0.6529
0.1332	6.67	7500	0.2294	0.5421	0.6526
0.1293	7.11	8000	0.2167	0.5576	0.6652
0.1475	7.56	8500	0.2218	0.5676	0.6649
0.1376	8.0	9000	0.2203	0.5565	0.6709
0.1408	8.44	9500	0.2178	0.5541	0.6716
0.133	8.89	10000	0.2212	0.5692	0.6640
0.1363	9.33	10500	0.2148	0.5642	0.6736
0.1344	9.78	11000	0.2162	0.5607	0.6726

Framework versions

• Transformers 4.19.4
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1