|
--- |
|
license: apache-2.0 |
|
tags: |
|
- generated_from_trainer |
|
datasets: |
|
- conll2002 |
|
metrics: |
|
- precision |
|
- recall |
|
- f1 |
|
- accuracy |
|
model-index: |
|
- name: roberta-base-bne-capitel-ner |
|
results: |
|
- task: |
|
name: Token Classification |
|
type: token-classification |
|
dataset: |
|
name: conll2002 |
|
type: conll2002 |
|
config: es |
|
split: validation |
|
args: es |
|
metrics: |
|
- name: Precision |
|
type: precision |
|
value: 0.8637694213015087 |
|
- name: Recall |
|
type: recall |
|
value: 0.8814338235294118 |
|
- name: F1 |
|
type: f1 |
|
value: 0.8725122256340272 |
|
- name: Accuracy |
|
type: accuracy |
|
value: 0.9780298635072827 |
|
--- |
|
|
|
<!-- 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. --> |
|
|
|
# roberta-base-bne-capitel-ner |
|
|
|
Este modelo es un finetuning de [BSC-LT/roberta-base-bne-capitel-ner](https://huggingface.co/BSC-LT/roberta-base-bne-capitel-ner) sobre el dataset conll2002. |
|
Este modelo logra los siguientes resultados sobre el conjunto de testeo: |
|
- Loss: 0.1137 |
|
- Precision: 0.8638 |
|
- Recall: 0.8814 |
|
- F1: 0.8725 |
|
- Accuracy: 0.9780 |
|
|
|
## Model description |
|
|
|
|
|
|
|
## Intended uses & limitations |
|
|
|
CoNLL2002 es el conjunto de datos español de la Tarea Compartida CoNLL-2002 (Tjong Kim Sang, 2002). El conjunto de datos está anotado con cuatro tipos de entidades nombradas (personas, ubicaciones, organizaciones y otras entidades diversas) formateadas en el formato estándar Beginning-Inside-Outside (BIO). El corpus consta de 8.324 sentencias de tren con 19.400 entidades nombradas, |
|
1.916 sentencias de desarrollo con 4.568 entidades nombradas y 1.518 sentencias de prueba con 3.644 entidades nombradas. |
|
|
|
## Training and evaluation data |
|
|
|
El modelo fue entrenado con una GPU 3080 TI de 10 Gz a 5 épocas y con un batch-seize de 8 y evaluado con F1-score por cada una de las épocas. |
|
|
|
## 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: 5 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy | |
|
|:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:| |
|
| 0.0041 | 1.0 | 1041 | 0.1137 | 0.8638 | 0.8814 | 0.8725 | 0.9780 | |
|
| 0.004 | 2.0 | 2082 | 0.1137 | 0.8638 | 0.8814 | 0.8725 | 0.9780 | |
|
| 0.0039 | 3.0 | 3123 | 0.1137 | 0.8638 | 0.8814 | 0.8725 | 0.9780 | |
|
| 0.003 | 4.0 | 4164 | 0.1137 | 0.8638 | 0.8814 | 0.8725 | 0.9780 | |
|
| 0.0032 | 5.0 | 5205 | 0.1137 | 0.8638 | 0.8814 | 0.8725 | 0.9780 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.30.0 |
|
- Pytorch 2.0.1+cu117 |
|
- Datasets 2.14.4 |
|
- Tokenizers 0.13.3 |
|
|