We recommend sharing information about each model in a model_card
.
The canonical documentation about model cards is at https://huggingface.co/docs
One of the easiest way to get started is by using our template card. Simply copy template.README.md
to model_cards/USERNAME/MODELNAME/README.md
and fill it out while studying demo.README.md for an example of a model model_card
and referring to rest of this short document if you are not sure about some specific fields.
Here is a sample of a typical yaml metadata section:
---
language:
- ru
- en
thumbnail: https://raw.githubusercontent.com/JetRunner/BERT-of-Theseus/master/bert-of-theseus.png
tags:
- translation
- fsmt
license: apache-2.0
datasets:
- wmt19
metrics:
- bleu
- sacrebleu
---
Important:
- This section has to be first in the document, before any markdown sections.
- The data in this section doesn't need to be repeated again in the subsequent markdown sections - since it'll be automatically expanded into a user-friendly format on the website.
ISO 639-1 code for your language (e.g. ru
, en
, de
, or multilingual
)
"url to a thumbnail used in social sharing"
- array
- of
- tags
One of the valid license identifiers. e.g.:
- apache-2.0
- mit
One or more dataset identifiers.
Example:
- wmt19
- wmt16
You will find the supported list at https://huggingface.co/datasets
One or more metric identifiers. e.g.:
- bleu
- rouge
- sacrebleu
You will find the supported list at https://huggingface.co/metrics
You can embed local or remote images using ![](...)
# You can include sample code which will be formatted
Provide examples of latent issues and potential remediations.
Describe the data you used to train the model. If you initialized it with pre-trained weights, add a link to the pre-trained model card or repository with description of the pre-training data.
Preprocessing, hardware used, hyperparameters...
@inproceedings{...,
year={2020}
}