torchexplorer
torchexplorer copied to clipboard
Make grad / grad norm configurable
Right now, all i/o gradients are logged using the grad norm, and parameter gradients are logged using the raw gradient. It would be nice to make this configurable.
Probably best implementation would be to have the log
list take:
io_vals, io_grads, io_grad_norms, param_vals, param_grads, param_grad_norms
The attach
parameter should be [vals
, grads
, grad_norms
]