CeLLaTe-tapt-pubmedbert-tokenizer-original-spanmask

This model is a fine-tuned version of microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext on the Mardiyyah/TAPT_CeLLaTe2.0 dataset. It achieves the following results on the evaluation set:

• Loss: 2.1721
• Accuracy: 0.5857
• Perplexity: 8.7763

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
2.7396	1.0	14	2.3642	0.5703	10.6359
2.5077	2.0	28	2.2759	0.5742	9.7369
2.4329	3.0	42	2.1662	0.5936	8.7251
2.3886	4.0	56	2.1748	0.5859	8.8000
2.378	5.0	70	2.1150	0.5914	8.2897
2.3618	6.0	84	2.1638	0.5821	8.7040
2.322	7.0	98	2.1467	0.5910	8.5563
2.299	8.0	112	2.1630	0.5827	8.6973
2.3081	9.0	126	2.0821	0.5989	8.0216
2.2939	10.0	140	2.1986	0.5835	9.0125
2.2826	11.0	154	2.1773	0.5870	8.8226
2.2581	12.0	168	2.1368	0.5880	8.4727
2.2392	13.0	182	2.1805	0.5857	8.8505
2.2305	14.0	196	2.1175	0.5906	8.3106

Framework versions

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