This would be useful in LLM use cases where the dataset might be small or just a simple file made of text. It would be convenient to be able to upload such a file from a Jupyter notebook or the UI and use it as input to an entrypoint rather than loading it in through the worker container's mounted volumes.
For vision-based use cases this is not as big a deal since it is fairly rare for the user to want to edit the dataset manually, but in language-based use cases, editing the dataset manually and re-uploading can add a lot of benefit, but it is currently cumbersome to do that.
This would be useful in LLM use cases where the dataset might be small or just a simple file made of text. It would be convenient to be able to upload such a file from a Jupyter notebook or the UI and use it as input to an entrypoint rather than loading it in through the worker container's mounted volumes.
For vision-based use cases this is not as big a deal since it is fairly rare for the user to want to edit the dataset manually, but in language-based use cases, editing the dataset manually and re-uploading can add a lot of benefit, but it is currently cumbersome to do that.