How to determine if a file is a safetensor file

[bernt-matthias]

Is there a good and fast way to determine if a file is a safetensors file. We would like to avoid reading the whole header.

Background we are currently trying to add safetensors as a datatype to the Galaxy project: galaxyproject/galaxy#20754

[julien-c]

is using the file extension an option?

[bernt-matthias]

Thanks for the feedback @julien-c. File extension is helpful for a 1st guess, but it would be better to be able to decide based on the contents.

[julien-c]

there's no magic number in safetensors so the way to go could be to check whether the start of the header looks like JSON:

note the full header is usually pretty small (proportional to the number of tensors)