Currently, you hand off the output file to cog and then none of your code runs and the file gets left in place.
However, there's a class of bug where models can return outputs for previous predictions if some overwrite fails.
This also stops us from using tmp files as output files, since they never get deleted. Cog should probably delete those files (or even give the user a temp directory/overlay to work in).
Currently, you hand off the output file to cog and then none of your code runs and the file gets left in place.
However, there's a class of bug where models can return outputs for previous predictions if some overwrite fails.
This also stops us from using tmp files as output files, since they never get deleted. Cog should probably delete those files (or even give the user a temp directory/overlay to work in).