Loader Script
@mariosasko Shifting the conversation here for the JXL loader script prototype.
What's the official URLs I should use to download the dataset .jsonl and then .jxl files? Can I just use the same as the "raw" view or "download" button in the web frontend?
The script is working fine locally, just need to finalize it. A few more questions:
- What name do I need to give the script so it's picked up?
- What functions are run on the server, if any?
- Does the server provide an automatic .zip or archive of subfolders to download?
- Is the meta-data in the README overriden by that in the script?
- Is it possible to disable the dataset preview with the script?
Thanks!
Hi ! :)
What name do I need to give the script so it's picked up?
the same name as the dataset: td01_natural-ground-textures.py
What functions are run on the server, if any?
For the preview you mean ? We only use the datasets lib on our side to run the script and get the preview. So it runs the code in the dataset builder class.
Does the server provide an automatic .zip or archive of subfolders to download?
You can just git clone the repository to download it
Is the meta-data in the README overriden by that in the script?
yes - but please keep your readme updated ;)
Is it possible to disable the dataset preview with the script?
you can disable the preview with a YAML tag in the README.md: viewer: false
see the docs at https://huggingface.co/docs/hub/datasets-viewer#disable-the-viewer
Many thanks, almost there!
Is there a way I can just git push to the repository? I couldn't find a way to configure SSH keys like GitHub, just in case I missed something...
You must be logged in to push to the repo (this doc explains the steps: https://huggingface.co/docs/hub/repositories-getting-started#cloning-repositories).