cellate1.0-tapt_freeze_llrd_ww_mask-LR_2e-05

This model is a fine-tuned version of microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext on the Unlabelled Fulltext article collection of CeLLaTe. It achieves the following results on the evaluation set:

• Loss: 1.3267
• Accuracy: 0.7214

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: 16
• eval_batch_size: 16
• seed: 3407
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-06 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.06
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.3798	1.0	6	1.3859	0.7176
1.3841	2.0	12	1.3744	0.7155
1.3566	3.0	18	1.4371	0.7130
1.3652	4.0	24	1.3793	0.7193
1.3869	5.0	30	1.4522	0.7055
1.3302	6.0	36	1.4605	0.7042
1.3277	7.0	42	1.4848	0.7005
1.3264	8.0	48	1.3649	0.7277
1.3271	9.0	54	1.3329	0.7348
1.3242	10.0	60	1.4326	0.7109
1.3164	11.0	66	1.3092	0.7252
1.3218	12.0	72	1.4003	0.7114
1.3153	13.0	78	1.2768	0.7323
1.3749	14.0	84	1.3195	0.7256
1.3106	15.0	90	1.3585	0.7067
1.3176	16.0	96	1.3711	0.7264
1.4064	16.7273	100	1.3267	0.7214

Framework versions

• Transformers 4.48.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.2
• Tokenizers 0.21.0