BERT_pretraining_h_100_wo_deepspeed

This model is a fine-tuned version of bert-large-uncased on the gokuls/wiki_book_corpus_complete_processed_bert_dataset dataset. It achieves the following results on the evaluation set:

• Loss: 5.7778
• Accuracy: 0.1539

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: 1e-05
• train_batch_size: 208
• eval_batch_size: 208
• seed: 10
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100000
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
6.8769	0.36	10000	6.7582	0.1101
6.4647	0.71	20000	6.4764	0.1314
6.3679	1.07	30000	6.3218	0.1407
6.252	1.42	40000	6.2139	0.1454
6.2132	1.78	50000	6.1398	0.1478
6.0407	2.13	60000	6.0774	0.1502
6.0694	2.49	70000	6.0303	0.1516
5.9996	2.84	80000	5.9893	0.1521
5.9166	3.2	90000	5.9553	0.1526
5.8915	3.55	100000	5.9261	0.1530
5.8924	3.91	110000	5.8996	0.1534
5.8972	4.26	120000	5.8814	0.1533
5.8454	4.62	130000	5.8626	0.1532
5.8104	4.97	140000	5.8494	0.1534
5.8461	5.33	150000	5.8378	0.1534
5.8476	5.68	160000	5.8246	0.1536
5.7255	6.04	170000	5.8155	0.1532
5.8431	6.39	180000	5.8068	0.1537
5.7526	6.75	190000	5.7981	0.1537
5.7826	7.1	200000	5.7886	0.1537

Framework versions

• Transformers 4.37.1
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1