bert-base-uncased-issues-128

This model is a fine-tuned version of bert-base-uncased on the GitHub issues dataset. The model is used in Chapter 9: Dealing with Few to No Labels in the NLP with Transformers book. You can find the full code in the accompanying Github repository.

It achieves the following results on the evaluation set:

• Loss: 1.2520

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
2.0949	1.0	291	1.7072
1.649	2.0	582	1.4409
1.4835	3.0	873	1.4099
1.3938	4.0	1164	1.3858
1.3326	5.0	1455	1.2004
1.2949	6.0	1746	1.2955
1.2451	7.0	2037	1.2682
1.1992	8.0	2328	1.1938
1.1784	9.0	2619	1.1686
1.1397	10.0	2910	1.2050
1.1293	11.0	3201	1.2058
1.1006	12.0	3492	1.1680
1.0835	13.0	3783	1.2414
1.0757	14.0	4074	1.1522
1.062	15.0	4365	1.1176
1.0535	16.0	4656	1.2520

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu102
• Datasets 1.13.0
• Tokenizers 0.10.3