roberta-large-finetuned-abbr-filtered-plod
This model is a fine-tuned version of the roberta-large on the PLODv2 filtered dataset. It is released with our LREC-COLING 2024 publication Using character-level models for efficient abbreviation and long-form detection. It achieves the following results on the test set:
Results on abbreviations:
- Precision: 0.9073
- Recall: 0.9348
- F1: 0.9208
Results on long forms:
- Precision: 0.8908
- Recall: 0.9318
- F1: 0.9108
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: 4
- eval_batch_size: 4
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 6
Training results
Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
---|---|---|---|---|---|---|---|
0.1169 | 0.25 | 7000 | 0.1114 | 0.9639 | 0.9581 | 0.9610 | 0.9575 |
0.1171 | 0.5 | 14000 | 0.1150 | 0.9655 | 0.9534 | 0.9594 | 0.9554 |
0.1202 | 0.75 | 21000 | 0.1058 | 0.9644 | 0.9578 | 0.9611 | 0.9575 |
0.1105 | 0.99 | 28000 | 0.1098 | 0.9664 | 0.9549 | 0.9606 | 0.9566 |
0.0935 | 1.24 | 35000 | 0.1270 | 0.9643 | 0.9570 | 0.9606 | 0.9570 |
0.0999 | 1.49 | 42000 | 0.1112 | 0.9626 | 0.9605 | 0.9615 | 0.9580 |
0.0948 | 1.74 | 49000 | 0.1114 | 0.9670 | 0.9606 | 0.9638 | 0.9603 |
0.1015 | 1.99 | 56000 | 0.1146 | 0.9680 | 0.9589 | 0.9634 | 0.9597 |
0.0816 | 2.24 | 63000 | 0.1244 | 0.9670 | 0.9607 | 0.9638 | 0.9603 |
0.0855 | 2.49 | 70000 | 0.1107 | 0.9675 | 0.9623 | 0.9649 | 0.9614 |
0.0814 | 2.73 | 77000 | 0.1047 | 0.9661 | 0.9630 | 0.9645 | 0.9611 |
0.0827 | 2.98 | 84000 | 0.1082 | 0.9665 | 0.9631 | 0.9648 | 0.9614 |
0.0655 | 3.23 | 91000 | 0.1485 | 0.9690 | 0.9615 | 0.9653 | 0.9618 |
0.0631 | 3.48 | 98000 | 0.1314 | 0.9683 | 0.9639 | 0.9661 | 0.9627 |
0.0667 | 3.73 | 105000 | 0.1164 | 0.9683 | 0.9643 | 0.9663 | 0.9629 |
0.0652 | 3.98 | 112000 | 0.1297 | 0.9681 | 0.9653 | 0.9667 | 0.9633 |
0.0485 | 4.23 | 119000 | 0.1441 | 0.9697 | 0.9645 | 0.9671 | 0.9636 |
0.0505 | 4.47 | 126000 | 0.1350 | 0.9700 | 0.9651 | 0.9675 | 0.9642 |
0.0498 | 4.72 | 133000 | 0.1243 | 0.9691 | 0.9657 | 0.9674 | 0.9640 |
0.0463 | 4.97 | 140000 | 0.1392 | 0.9699 | 0.9660 | 0.9679 | 0.9645 |
0.0371 | 5.22 | 147000 | 0.1527 | 0.9709 | 0.9658 | 0.9683 | 0.9649 |
0.0363 | 5.47 | 154000 | 0.1490 | 0.9703 | 0.9667 | 0.9685 | 0.9651 |
0.0341 | 5.72 | 161000 | 0.1538 | 0.9712 | 0.9666 | 0.9689 | 0.9656 |
0.0338 | 5.97 | 168000 | 0.1488 | 0.9705 | 0.9668 | 0.9687 | 0.9653 |
Framework versions
- Transformers 4.16.2
- Pytorch 1.11.0
- Datasets 2.1.0
- Tokenizers 0.10.3
- Downloads last month
- 16
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.