bert_uncased_L-4_H-128_A-2_stsb

This model is a fine-tuned version of google/bert_uncased_L-4_H-128_A-2 on the GLUE STSB dataset. It achieves the following results on the evaluation set:

• Loss: 0.7195
• Pearson: 0.8255
• Spearmanr: 0.8275
• Combined Score: 0.8265

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
6.5937	1.0	23	4.2841	0.1711	0.2041	0.1876
4.7932	2.0	46	3.4947	0.2599	0.2547	0.2573
3.9815	3.0	69	2.9842	0.1082	0.1081	0.1081
3.3791	4.0	92	2.6163	0.0598	0.0576	0.0587
2.8814	5.0	115	2.3940	nan	nan	nan
2.5786	6.0	138	2.2811	nan	nan	nan
2.3519	7.0	161	2.2501	nan	nan	nan
2.2133	8.0	184	2.2792	0.3315	0.3276	0.3295
2.0732	9.0	207	1.7190	0.6363	0.5993	0.6178
1.688	10.0	230	1.2774	0.7001	0.7069	0.7035
1.3697	11.0	253	1.2016	0.6980	0.7378	0.7179
1.197	12.0	276	1.1024	0.7279	0.7207	0.7243
1.0723	13.0	299	0.9007	0.7777	0.7922	0.7849
0.9748	14.0	322	0.8908	0.7814	0.7975	0.7895
0.923	15.0	345	0.8621	0.7905	0.8112	0.8008
0.8518	16.0	368	0.8973	0.7831	0.7829	0.7830
0.7957	17.0	391	0.7932	0.8057	0.8142	0.8100
0.7691	18.0	414	0.8001	0.8066	0.8182	0.8124
0.7204	19.0	437	0.7710	0.8134	0.8204	0.8169
0.6959	20.0	460	0.7704	0.8145	0.8228	0.8187
0.6458	21.0	483	0.7613	0.8175	0.8231	0.8203
0.6427	22.0	506	0.7614	0.8188	0.8233	0.8210
0.6307	23.0	529	0.7195	0.8255	0.8275	0.8265
0.5979	24.0	552	0.7540	0.8207	0.8243	0.8225
0.5774	25.0	575	0.7823	0.8183	0.8226	0.8204
0.5542	26.0	598	0.7305	0.8253	0.8278	0.8266
0.548	27.0	621	0.7279	0.8261	0.8278	0.8270
0.5295	28.0	644	0.7397	0.8254	0.8267	0.8260

Framework versions

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