metadata
license: cc-by-nc-4.0
metrics:
- accuracy
- f1
datasets:
- orai-nlp/basqueGLUE
language:
- eu
pipeline_tag: token-classification
tags:
- NERC
ElhBERTeu model finetuned on the NERC dataset from BasqueGlue.
In Domain and Out-of-domain tasks' train and dev datasets were merged for training. Reported performance results are tested on the merged tests dataset after training for 10 epochs (batch size 32, learning rate 3e-5).
Results on test set:
- accuracy = 0.9785551028520322
- f1 = 0.8665899340797322
- loss = 0.11313809949144367
- precision = 0.8650511802799248
- recall = 0.8681341719077568
Per class results:
| CLASS | precision | recall | F1 score | support |
|---|---|---|---|---|
| LOC | 89.43% | 87.95% | 88.68 | 1844 |
| MISC | 70.72% | 67.75% | 69.20 | 502 |
| ORG | 80.04% | 84.32% | 82.12 | 1082 |
| PER | 93.52% | 94.57% | 94.04 | 1359 |
| GLOBAL | 86.51% | 86.81% | 86.66 |
Tagset: ["O", "B-ORG", "B-PER", "I-PER", "I-ORG", "B-LOC", "I-LOC", "B-MISC", "I-MISC"]
Finetuning details: 10 epochs, batch size 32, learning rate 3e-5.