CeLLaTe-tapt-pubmedbert-tokenizer-adapted-wwmask

This model is a fine-tuned version of bioformers/bioformer-16l on the Mardiyyah/TAPT_CeLLaTe2.0 dataset. It achieves the following results on the evaluation set:

Loss: 1.4695
Accuracy: 0.7180
Perplexity: 4.3469

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: 3e-05
train_batch_size: 32
eval_batch_size: 32
seed: 3407
gradient_accumulation_steps: 2
total_train_batch_size: 64
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: 100
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Perplexity
1.983	1.0	15	1.7321	0.7043	5.6526
1.953	2.0	30	1.6524	0.7021	5.2195
1.902	3.0	45	1.6351	0.7119	5.1301
1.8591	4.0	60	1.6204	0.7119	5.0553
1.831	5.0	75	1.5531	0.7137	4.7261
1.7986	6.0	90	1.5859	0.7055	4.8838
1.7585	7.0	105	1.5395	0.7145	4.6625
1.7148	8.0	120	1.4938	0.7194	4.4540
1.7015	9.0	135	1.5612	0.7073	4.7647
1.6811	10.0	150	1.4989	0.7162	4.4766
1.6711	11.0	165	1.5252	0.7091	4.5960
1.6506	12.0	180	1.5499	0.7093	4.7109
1.6307	13.0	195	1.4886	0.7187	4.4308
1.6393	14.0	210	1.4708	0.7176	4.3529
1.5838	15.0	225	1.4932	0.7175	4.4513
1.5804	16.0	240	1.5226	0.7088	4.5840
1.5629	17.0	255	1.4859	0.7175	4.4188
1.5368	18.0	270	1.4150	0.7257	4.1167
1.5255	19.0	285	1.5382	0.7117	4.6561
1.5222	20.0	300	1.5028	0.7142	4.4945
1.5103	21.0	315	1.4642	0.7175	4.3239
1.5249	22.0	330	1.4608	0.7147	4.3096
1.4988	23.0	345	1.4419	0.7232	4.2288