bert_uncased_L-4_H-512_A-8_stsb

This model is a fine-tuned version of google/bert_uncased_L-4_H-512_A-8 on the GLUE STSB dataset. It achieves the following results on the evaluation set:

• Loss: 0.5646
• Pearson: 0.8739
• Spearmanr: 0.8704
• Combined Score: 0.8721

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: 5e-05
• train_batch_size: 256
• eval_batch_size: 256
• seed: 10
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Pearson	Spearmanr	Combined Score
2.5878	1.0	23	0.9754	0.7823	0.7580	0.7702
0.797	2.0	46	0.7766	0.8466	0.8482	0.8474
0.5786	3.0	69	0.6314	0.8603	0.8587	0.8595
0.4961	4.0	92	0.6342	0.8643	0.8637	0.8640
0.3944	5.0	115	0.6018	0.8694	0.8683	0.8689
0.3362	6.0	138	0.6101	0.8659	0.8657	0.8658
0.2932	7.0	161	0.6056	0.8678	0.8666	0.8672
0.2495	8.0	184	0.6255	0.8679	0.8672	0.8675
0.2268	9.0	207	0.5970	0.8699	0.8685	0.8692
0.2037	10.0	230	0.6517	0.8691	0.8672	0.8682
0.191	11.0	253	0.6017	0.8709	0.8677	0.8693
0.1678	12.0	276	0.6097	0.8704	0.8685	0.8695
0.1546	13.0	299	0.6052	0.8713	0.8701	0.8707
0.1486	14.0	322	0.5914	0.8714	0.8689	0.8701
0.1372	15.0	345	0.6175	0.8738	0.8702	0.8720
0.131	16.0	368	0.5826	0.8727	0.8702	0.8715
0.1216	17.0	391	0.5779	0.8717	0.8686	0.8702
0.1145	18.0	414	0.5646	0.8739	0.8704	0.8721
0.1158	19.0	437	0.5811	0.8738	0.8711	0.8724
0.109	20.0	460	0.5896	0.8763	0.8720	0.8742
0.105	21.0	483	0.5863	0.8737	0.8705	0.8721
0.0995	22.0	506	0.5758	0.8741	0.8701	0.8721
0.0971	23.0	529	0.5781	0.8748	0.8713	0.8731

Framework versions

• Transformers 4.46.3
• Pytorch 2.2.1+cu118
• Datasets 2.17.0
• Tokenizers 0.20.3