Sergii Dymchenko comments

Results 140 comments of


                                            Sergii Dymchenko

Is it possible run horovod with unequal steps on each worker?

I've added a very early work in progress PR for discussion here: https://github.com/horovod/horovod/pull/1058 I've asked several questions there, could you please have a look?

Is it possible run horovod with unequal steps on each worker?

@spatters Now I remember that the original design was to return the rank of the last worker, but I forgot to properly implement it - in my case the first...

Add cmake, zstd, tomli to requires

No longer relevant.

[dynamo] support group=None when rewriting collectives

@pytorchbot revert -m "As previous PR in stack was reverted https://github.com/pytorch/pytorch/pull/120118" -c weird

[dynamo] support group=None when rewriting collectives

Manual revert https://github.com/pytorch/pytorch/pull/120677

`torch.load` without `weights_only` parameter is unsafe

@araffin you should specify `weights_only=False` if you need pickle. Otherwise when soon PyTorch changes the default of `weights_only` to `True`, your code will break.

`torch.load` without `weights_only` parameter is unsafe

@araffin I don't think there is an announcement, but we're definitely thinking of it. See this comment https://github.com/pytorch/pytorch/issues/111806#issuecomment-1785685208

`torch.load` without `weights_only` parameter is unsafe

@araffin The PR that added the option is https://github.com/pytorch/pytorch/pull/86812, first release with it is PyTorch 1.13.0

[dynamo] use lazy disable dynamo for manual seed

@pytorchbot revert -m "breaking internal builds" -c ghfirst

[opinfo] conv3d

@pytorchbot rebase