HBERTv1_48_L4_H512_A8_massive

This model is a fine-tuned version of gokuls/HBERTv1_48_L4_H512_A8 on the massive dataset. It achieves the following results on the evaluation set:

• Loss: 0.7982
• Accuracy: 0.8628

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
2.2635	1.0	180	1.0540	0.7378
0.852	2.0	360	0.7276	0.8067
0.5577	3.0	540	0.6679	0.8224
0.3964	4.0	720	0.6441	0.8352
0.28	5.0	900	0.6340	0.8441
0.2055	6.0	1080	0.6703	0.8470
0.1508	7.0	1260	0.6997	0.8465
0.1185	8.0	1440	0.7411	0.8480
0.0889	9.0	1620	0.7219	0.8519
0.0652	10.0	1800	0.7566	0.8613
0.0477	11.0	1980	0.7694	0.8564
0.0331	12.0	2160	0.7993	0.8578
0.0247	13.0	2340	0.7982	0.8628
0.0174	14.0	2520	0.8348	0.8554
0.0141	15.0	2700	0.8415	0.8583

Framework versions

• Transformers 4.34.0
• Pytorch 1.14.0a0+410ce96
• Datasets 2.14.5
• Tokenizers 0.14.0