Save long tensors using wandb.save OR wandb.log?

Hi all,
In my experimentation, I want to save some tensor_10k_size for a dataset which has 90k samples. so that I can compare how the tensors look for different datasets. It’s part of my thesis.
Which is a better option to save tensors in my scripts using wandb.save(‘filename’)
or
wandb.log(tensor)

Thanks,
Prachi

Hi Prachi,

It completely depends on what you are trying to do with the tensor. If you are trying to visualize the tensor (a tensor is usually presented as a histogram - unless it is of shape (W, H, C) in which case it can be coerced to an image - it can be logged using wandb.log.

If you want to load up those tensors again to be used in your code -wandb.save would be the way to go.

Thanks,
Ramit

1 Like

Hi Ramit,
That was my hazy view, you made it clear.

Thanks,
Prachi