BERT-tiny-Massive-intent

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

• Loss: 0.6740
• Accuracy: 0.8475

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
3.6104	1.0	720	3.0911	0.3601
2.8025	2.0	1440	2.3800	0.5165
2.2292	3.0	2160	1.9134	0.5991
1.818	4.0	2880	1.5810	0.6744
1.5171	5.0	3600	1.3522	0.7108
1.2876	6.0	4320	1.1686	0.7442
1.1049	7.0	5040	1.0355	0.7683
0.9623	8.0	5760	0.9466	0.7885
0.8424	9.0	6480	0.8718	0.7875
0.7473	10.0	7200	0.8107	0.8028
0.6735	11.0	7920	0.7710	0.8180
0.6085	12.0	8640	0.7404	0.8210
0.5536	13.0	9360	0.7180	0.8229
0.5026	14.0	10080	0.6980	0.8318
0.4652	15.0	10800	0.6970	0.8337
0.4234	16.0	11520	0.6822	0.8372
0.3987	17.0	12240	0.6691	0.8436
0.3707	18.0	12960	0.6679	0.8455
0.3433	19.0	13680	0.6740	0.8475
0.3206	20.0	14400	0.6760	0.8451
0.308	21.0	15120	0.6704	0.8436
0.2813	22.0	15840	0.6701	0.8416

Framework versions

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