David
Update README.md
6c3ff31
---
language: es
tags:
- zero-shot-classification
- nli
- pytorch
datasets:
- xnli
pipeline_tag: zero-shot-classification
license: apache-2.0
widget:
- text: "El autor se perfila, a los 50 años de su muerte, como uno de los grandes de su siglo"
candidate_labels: "cultura, sociedad, economia, salud, deportes"
---
# Zero-shot SELECTRA: A zero-shot classifier based on SELECTRA
*Zero-shot SELECTRA* is a [SELECTRA model](https://huggingface.co/Recognai/selectra_small) fine-tuned on the Spanish portion of the [XNLI dataset](https://huggingface.co/datasets/xnli). You can use it with Hugging Face's [Zero-shot pipeline](https://huggingface.co/transformers/master/main_classes/pipelines.html#transformers.ZeroShotClassificationPipeline) to make [zero-shot classifications](https://joeddav.github.io/blog/2020/05/29/ZSL.html).
In comparison to our previous zero-shot classifier [based on BETO](https://huggingface.co/Recognai/bert-base-spanish-wwm-cased-xnli), zero-shot SELECTRA is **much more lightweight**. As shown in the *Metrics* section, the *small* version (5 times fewer parameters) performs slightly worse, while the *medium* version (3 times fewer parameters) **outperforms** the BETO based zero-shot classifier.
## Usage
```python
from transformers import pipeline
classifier = pipeline("zero-shot-classification",
model="Recognai/zeroshot_selectra_medium")
classifier(
"El autor se perfila, a los 50 años de su muerte, como uno de los grandes de su siglo",
candidate_labels=["cultura", "sociedad", "economia", "salud", "deportes"],
hypothesis_template="Este ejemplo es {}."
)
"""Output
{'sequence': 'El autor se perfila, a los 50 años de su muerte, como uno de los grandes de su siglo',
'labels': ['sociedad', 'cultura', 'economia', 'salud', 'deportes'],
'scores': [0.6450043320655823,
0.16710571944713593,
0.08507631719112396,
0.0759836807847023,
0.026829993352293968]}
"""
```
The `hypothesis_template` parameter is important and should be in Spanish. **In the widget on the right, this parameter is set to its default value: "This example is {}.", so different results are expected.**
## Demo and tutorial
If you want to see this model in action, we have created a basic tutorial using [Rubrix](https://www.rubrix.ml/), a free and open-source tool to *explore, annotate, and monitor data for NLP*.
The tutorial shows you how to evaluate this classifier for news categorization in Spanish, and how it could be used to build a training set for training a supervised classifier (which might be useful if you want obtain more precise results or improve the model over time).
You can [find the tutorial here](https://rubrix.readthedocs.io/en/master/tutorials/zeroshot_data_annotation.html).
See the video below showing the predictions within the annotation process (see that the predictions are almost correct for every example).
<video width="100%" controls><source src="https://github.com/recognai/rubrix-materials/raw/main/tutorials/videos/zeroshot_selectra_news_data_annotation.mp4" type="video/mp4"></video>
## Metrics
| Model | Params | XNLI (acc) | \*MLSUM (acc) |
| --- | --- | --- | --- |
| [zs BETO](https://huggingface.co/Recognai/bert-base-spanish-wwm-cased-xnli) | 110M | 0.799 | 0.530 |
| zs SELECTRA medium | 41M | **0.807** | **0.589** |
| [zs SELECTRA small](https://huggingface.co/Recognai/zeroshot_selectra_small) | **22M** | 0.795 | 0.446 |
\*evaluated with zero-shot learning (ZSL)
- **XNLI**: The stated accuracy refers to the test portion of the [XNLI dataset](https://huggingface.co/datasets/xnli), after finetuning the model on the training portion.
- **MLSUM**: For this accuracy we take the test set of the [MLSUM dataset](https://huggingface.co/datasets/mlsum) and classify the summaries of 5 selected labels. For details, check out our [evaluation notebook](https://github.com/recognai/selectra/blob/main/zero-shot_classifier/evaluation.ipynb)
## Training
Check out our [training notebook](https://github.com/recognai/selectra/blob/main/zero-shot_classifier/training.ipynb) for all the details.
## Authors
- David Fidalgo ([GitHub](https://github.com/dcfidalgo))
- Daniel Vila ([GitHub](https://github.com/dvsrepo))
- Francisco Aranda ([GitHub](https://github.com/frascuchon))
- Javier Lopez ([GitHub](https://github.com/javispp))