distilbert_add_GLUE_Experiment_logit_kd_pretrain_qnli
This model is a fine-tuned version of gokuls/distilbert_add_pre-training-complete on the GLUE QNLI dataset. It achieves the following results on the evaluation set:
- Loss: 0.3579
- Accuracy: 0.6522
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
- distributed_type: multi-GPU
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 50
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.4059 | 1.0 | 410 | 0.4016 | 0.5585 |
| 0.3907 | 2.0 | 820 | 0.3735 | 0.6094 |
| 0.3715 | 3.0 | 1230 | 0.3602 | 0.6480 |
| 0.352 | 4.0 | 1640 | 0.3579 | 0.6522 |
| 0.3314 | 5.0 | 2050 | 0.3626 | 0.6670 |
| 0.309 | 6.0 | 2460 | 0.3650 | 0.6776 |
| 0.2865 | 7.0 | 2870 | 0.3799 | 0.6776 |
| 0.2679 | 8.0 | 3280 | 0.3817 | 0.6903 |
| 0.2525 | 9.0 | 3690 | 0.3942 | 0.6822 |
Framework versions
- Transformers 4.26.0
- Pytorch 1.14.0a0+410ce96
- Datasets 2.9.0
- Tokenizers 0.13.2