bert-base-uncased-issues-128

This model is a fine-tuned version of bert-base-uncased on the emotion dataset. It achieves the following results on the evaluation set:

• Loss: 2.3196

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: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 16

Training results

Training Loss	Epoch	Step	Validation Loss
2.6586	1.0	500	2.4203
2.4655	2.0	1000	2.3889
2.3769	3.0	1500	2.3279
2.3024	4.0	2000	2.3623
2.2178	5.0	2500	2.4500
2.1352	6.0	3000	2.3042
2.0793	7.0	3500	2.3053
2.0462	8.0	4000	2.2402
1.9886	9.0	4500	2.3407
1.9393	10.0	5000	2.2826
1.904	11.0	5500	2.3401
1.8742	12.0	6000	2.3276
1.8441	13.0	6500	2.2815
1.8082	14.0	7000	2.2739
1.8058	15.0	7500	2.3225
1.7951	16.0	8000	2.2933

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1