hBERTv1_new_pretrain_48_emb_com_stsb

This model is a fine-tuned version of gokuls/bert_12_layer_model_v1_complete_training_new_emb_compress_48 on the GLUE STSB dataset. It achieves the following results on the evaluation set:

• Loss: 1.9214
• Pearson: 0.4648
• Spearmanr: 0.4600
• Combined Score: 0.4624

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: 4e-05
• train_batch_size: 128
• eval_batch_size: 128
• seed: 10
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Pearson	Spearmanr	Combined Score
2.5817	1.0	45	2.6028	0.2027	0.1896	0.1962
2.1023	2.0	90	2.1596	0.2035	0.1938	0.1986
1.9567	3.0	135	2.3409	0.1855	0.1931	0.1893
1.7201	4.0	180	2.1790	0.2865	0.2934	0.2899
1.5153	5.0	225	2.1208	0.3381	0.3352	0.3367
1.2674	6.0	270	2.1224	0.3882	0.3898	0.3890
1.0115	7.0	315	2.2253	0.4304	0.4281	0.4293
0.7449	8.0	360	2.3235	0.4236	0.4323	0.4279
0.66	9.0	405	2.3617	0.4340	0.4351	0.4346
0.4678	10.0	450	2.0741	0.4300	0.4258	0.4279
0.4438	11.0	495	2.3816	0.4285	0.4294	0.4289
0.3192	12.0	540	2.1673	0.4580	0.4602	0.4591
0.2481	13.0	585	2.1544	0.4392	0.4357	0.4374
0.2296	14.0	630	2.0075	0.4603	0.4582	0.4593
0.1765	15.0	675	2.1395	0.4624	0.4617	0.4621
0.1533	16.0	720	2.2715	0.4512	0.4427	0.4469
0.1343	17.0	765	2.1726	0.4441	0.4417	0.4429
0.1373	18.0	810	2.0223	0.4532	0.4424	0.4478
0.1277	19.0	855	1.9992	0.4395	0.4299	0.4347
0.0968	20.0	900	2.1078	0.4620	0.4601	0.4610
0.084	21.0	945	2.0684	0.4627	0.4577	0.4602
0.0777	22.0	990	1.9214	0.4648	0.4600	0.4624
0.0572	23.0	1035	2.0636	0.4506	0.4422	0.4464
0.0615	24.0	1080	2.0404	0.4489	0.4388	0.4438
0.0516	25.0	1125	2.0599	0.4516	0.4435	0.4475
0.0501	26.0	1170	2.0359	0.4530	0.4489	0.4510
0.0515	27.0	1215	1.9571	0.4588	0.4508	0.4548

Framework versions

• Transformers 4.30.2
• Pytorch 1.14.0a0+410ce96
• Datasets 2.12.0
• Tokenizers 0.13.3