|
--- |
|
language: es |
|
tags: |
|
- biomedical |
|
- clinical |
|
- spanish |
|
- mdeberta-v3-base |
|
license: mit |
|
datasets: |
|
- "ehealth_kd" |
|
metrics: |
|
- f1 |
|
|
|
model-index: |
|
- name: IIC/mdeberta-v3-base-ehealth_kd |
|
results: |
|
- task: |
|
type: token-classification |
|
dataset: |
|
name: eHealth-KD |
|
type: ehealth_kd |
|
split: test |
|
metrics: |
|
- name: f1 |
|
type: f1 |
|
value: 0.844 |
|
pipeline_tag: token-classification |
|
|
|
--- |
|
|
|
# mdeberta-v3-base-ehealth_kd |
|
|
|
This model is a finetuned version of mdeberta-v3-base for the eHealth-KD dataset used in a benchmark in the paper TODO. The model has a F1 of 0.844 |
|
|
|
Please refer to the original publication for more information TODO LINK |
|
|
|
## Parameters used |
|
|
|
| parameter | Value | |
|
|-------------------------|:-----:| |
|
| batch size | 32 | |
|
| learning rate | 3e-05 | |
|
| classifier dropout | 0.2 | |
|
| warmup ratio | 0 | |
|
| warmup steps | 0 | |
|
| weight decay | 0 | |
|
| optimizer | AdamW | |
|
| epochs | 10 | |
|
| early stopping patience | 3 | |
|
|
|
|
|
## BibTeX entry and citation info |
|
|
|
```bibtex |
|
TODO |
|
``` |
|
|
|
|