PubMedBERT-Large-LitCovid-1.4

This model is a fine-tuned version of microsoft/BiomedNLP-BiomedBERT-large-uncased-abstract on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6105
• Hamming loss: 0.0623
• F1 micro: 0.6724
• F1 macro: 0.5303
• F1 weighted: 0.7292
• F1 samples: 0.6741
• Precision micro: 0.5423
• Precision macro: 0.4146
• Precision weighted: 0.6499
• Precision samples: 0.5845
• Recall micro: 0.8849
• Recall macro: 0.8178
• Recall weighted: 0.8849
• Recall samples: 0.9022
• Roc Auc: 0.9133
• Accuracy: 0.1313

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Hamming loss	F1 micro	F1 macro	F1 weighted	F1 samples	Precision micro	Precision macro	Precision weighted	Precision samples	Recall micro	Recall macro	Recall weighted	Recall samples	Roc Auc	Accuracy
0.589	1.0	1151	0.5719	0.1031	0.5554	0.4307	0.6704	0.5629	0.4034	0.3213	0.5843	0.4435	0.8909	0.8673	0.8909	0.9062	0.8941	0.0363
0.4668	2.0	2302	0.5438	0.0836	0.6082	0.4623	0.6974	0.6147	0.4599	0.3478	0.6098	0.5052	0.8976	0.8556	0.8976	0.9123	0.9077	0.0774
0.3791	3.0	3453	0.5510	0.0790	0.6225	0.4829	0.7070	0.6247	0.4754	0.3661	0.6205	0.5140	0.9012	0.8541	0.9012	0.9165	0.9119	0.0759
0.307	4.0	4605	0.5954	0.0635	0.6688	0.5235	0.7280	0.6689	0.5371	0.4078	0.6477	0.5767	0.8863	0.8212	0.8863	0.9036	0.9134	0.1229
0.2687	5.0	5755	0.6105	0.0623	0.6724	0.5303	0.7292	0.6741	0.5423	0.4146	0.6499	0.5845	0.8849	0.8178	0.8849	0.9022	0.9133	0.1313

Framework versions

• Transformers 4.28.0
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.13.3