bert_uncased_L-8_H-256_A-4_massive

This model is a fine-tuned version of google/bert_uncased_L-8_H-256_A-4 on the massive dataset. It achieves the following results on the evaluation set:

• Loss: 0.6529
• Accuracy: 0.8618

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: 64
• eval_batch_size: 64
• seed: 33
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.4012	1.0	180	2.6862	0.5268
2.3339	2.0	360	1.8775	0.6793
1.7107	3.0	540	1.4277	0.7526
1.3323	4.0	720	1.1676	0.7964
1.0741	5.0	900	1.0010	0.8111
0.8903	6.0	1080	0.8893	0.8269
0.757	7.0	1260	0.8285	0.8342
0.6586	8.0	1440	0.7609	0.8455
0.5814	9.0	1620	0.7194	0.8480
0.5163	10.0	1800	0.6987	0.8519
0.4728	11.0	1980	0.6760	0.8549
0.4371	12.0	2160	0.6659	0.8608
0.4053	13.0	2340	0.6592	0.8613
0.3867	14.0	2520	0.6523	0.8608
0.3784	15.0	2700	0.6529	0.8618

Framework versions

• Transformers 4.34.0
• Pytorch 1.14.0a0+410ce96
• Datasets 2.14.5
• Tokenizers 0.14.1