A warning appears in DDP using ConvNeXt
Toshimasa-Matsumoto opened this issue · comments
Toshimasa-Matsumoto commented
When I start training using DDP with ConvNeXt, the following warning appears.
What caused the warning?
Also, is there a way to avoid the warning?
UserWarning: Grad strides do not match bucket view strides. This may indicate grad was not created according to the gradient layout contract, or that the param's strides changed since DDP was constructed. This is not an error, but may impair performance.
grad.sizes() = [1024, 1, 7, 7], strides() = [49, 1, 7, 1]
bucket_view.sizes() = [1024, 1, 7, 7], strides() = [49, 49, 7, 1] (Triggered internally at ../torch/csrc/distributed/c10d/reducer.cpp:323.)
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
Thank you.
Anh Tien Nguyen commented
I also got this warning.
I found this https://github.com/pytorch/pytorch/issues/47163. So the rule is you need add .contiguous()
after the permute
operation