roberta-base-stsb

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

• Loss: 0.4221
• Pearson: 0.9116
• Spearmanr: 0.9092
• Combined Score: 0.9104

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.06
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Pearson	Spearmanr	Combined Score
1.6552	1.39	500	0.5265	0.8925	0.8925	0.8925
0.3579	2.78	1000	0.4626	0.9022	0.8991	0.9007
0.2198	4.17	1500	0.4396	0.9054	0.9042	0.9048
0.1585	5.56	2000	0.4537	0.9069	0.9052	0.9060
0.1139	6.94	2500	0.4975	0.9091	0.9065	0.9078
0.0868	8.33	3000	0.4221	0.9116	0.9092	0.9104
0.073	9.72	3500	0.4311	0.9096	0.9077	0.9086

Framework versions

• Transformers 4.21.3
• Pytorch 1.7.1
• Datasets 1.18.3
• Tokenizers 0.11.6