<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
xlm-roberta-large-NER-AstrID
This model is a fine-tuned version of xlm-roberta-large-finetuned-conll03-english on the WIESP2022-NER dataset. It achieves the following results on a sample evaluation set (from the dev set):
- Loss: 0.1950
- Precision: 0.8124
- Recall: 0.8140
- F1: 0.8132
- Accuracy: 0.9473
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
- 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 | 220 | 0.3204 | 0.6974 | 0.7064 | 0.7019 | 0.9229 |
| No log | 2.0 | 440 | 0.2166 | 0.8083 | 0.8072 | 0.8078 | 0.9462 |
| 0.3277 | 3.0 | 660 | 0.1950 | 0.8124 | 0.8140 | 0.8132 | 0.9473 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.12.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1
