bert-tiny-emotion-KD-distilBERT

This model is a fine-tuned version of google/bert_uncased_L-2_H-128_A-2 on the emotion dataset. It achieves the following results on the evaluation set:

• Loss: 0.5444
• Accuracy: 0.913

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: 16
• eval_batch_size: 16
• seed: 33
• 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
4.2533	1.0	1000	2.8358	0.7675
2.3274	2.0	2000	1.5893	0.8675
1.3974	3.0	3000	1.0286	0.891
0.9035	4.0	4000	0.7534	0.8955
0.6619	5.0	5000	0.6350	0.905
0.5482	6.0	6000	0.6180	0.899
0.4937	7.0	7000	0.5448	0.91
0.4013	8.0	8000	0.5493	0.906
0.3839	9.0	9000	0.5481	0.9095
0.3281	10.0	10000	0.5528	0.9115
0.3098	11.0	11000	0.5864	0.9095
0.2762	12.0	12000	0.5566	0.9095
0.2467	13.0	13000	0.5444	0.913
0.2286	14.0	14000	0.5306	0.912
0.2215	15.0	15000	0.5312	0.9115
0.2038	16.0	16000	0.5242	0.912

Framework versions

• Transformers 4.22.1
• Pytorch 1.12.1+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1