roberta_finetuned_astronomicalNER

This model is a fine-tuned version of xlm-roberta-large-finetuned-conll03-english for NER on astronomical objects. The dataset comes from the Shared Task DEAL: Detecting Entities in the Astrophysics Literature

The model achieves the following results on the evaluation set:

• Loss: 0.1416
• Precision: 0.7659
• Recall: 0.7986
• F1: 0.7819
• Accuracy: 0.9640

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	176	0.1571	0.7362	0.7788	0.7569	0.9593
No log	2.0	352	0.1416	0.7529	0.7831	0.7677	0.9624
0.1109	3.0	528	0.1416	0.7659	0.7986	0.7819	0.9640

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1