STS-Conventional-Fine-Tuning-pipe-roberta-base

This model is a fine-tuned version of FacebookAI/roberta-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.9623
• Accuracy: 0.7210

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: 3e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	113	0.7397	0.6760
No log	2.0	226	0.7139	0.7135
No log	3.0	339	0.7846	0.7041
No log	4.0	452	0.8368	0.6816
0.5292	5.0	565	1.0395	0.6873
0.5292	6.0	678	1.0863	0.7154
0.5292	7.0	791	1.2668	0.7097
0.5292	8.0	904	1.5312	0.7135
0.1806	9.0	1017	1.6574	0.7041
0.1806	10.0	1130	1.7880	0.7041
0.1806	11.0	1243	1.7902	0.7116
0.1806	12.0	1356	1.9168	0.7154
0.1806	13.0	1469	1.8637	0.7210
0.0717	14.0	1582	1.9391	0.7210
0.0717	15.0	1695	1.9623	0.7210

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2