Warning: grad and param do not obey the gradient layout contract.

[mohaEs]

Has anybody gotten a similar warning when using it?

Warning: grad and param do not obey the gradient layout contract. This is not an error, but may impair performance.
grad.sizes() = [512, 256, 1, 1], strides() = [256, 1, 1, 1]
param.sizes() = [512, 256, 1, 1], strides() = [256, 1, 256, 256] (function operator())

[hcmea]

@mohaEs, I just started training BYOL with my custom dataset and got the same warning too. Did your model performed better on the downstream task? Does this warning effect the performance?

[zmzhang2000]

I got the same warning too. And the loss becomes NaN during training. Still not figured it out.

[zmzhang2000]

I got the same warning too. And the loss becomes NaN during training. Still not figured it out.

I set the input to be contiguous and then the warning disappeared.

input = input.contiguous()