bert_uncased_L-2_H-512_A-8_stsb

This model is a fine-tuned version of google/bert_uncased_L-2_H-512_A-8 on the GLUE STSB dataset. It achieves the following results on the evaluation set:

• Loss: 0.6358
• Pearson: 0.8581
• Spearmanr: 0.8552
• Combined Score: 0.8566

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
3.6969	1.0	23	1.5140	0.7418	0.7179	0.7298
1.0774	2.0	46	0.8374	0.8224	0.8231	0.8228
0.7474	3.0	69	0.6987	0.8441	0.8408	0.8424
0.6216	4.0	92	0.6728	0.8521	0.8492	0.8507
0.5299	5.0	115	0.7252	0.8542	0.8526	0.8534
0.4661	6.0	138	0.6521	0.8564	0.8543	0.8554
0.4071	7.0	161	0.6358	0.8581	0.8552	0.8566
0.3604	8.0	184	0.6437	0.8567	0.8550	0.8559
0.3194	9.0	207	0.6857	0.8561	0.8530	0.8546
0.2981	10.0	230	0.6473	0.8575	0.8553	0.8564
0.2647	11.0	253	0.6541	0.8586	0.8552	0.8569
0.244	12.0	276	0.6888	0.8572	0.8546	0.8559

Framework versions

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