gpetters94

Okay, I've addressed the changes and rebased to main. Let me know if I'm good to merge.

> Actually, don't we only accept the `padding` as constant int list? Why do we need the `getDynamicZeroPaddedTensor` API? The calculation for the correct padding, `gradInputPaddingValues`, requires a bunch of...

> > > Actually, don't we only accept the `padding` as constant int list? Why do we need the `getDynamicZeroPaddedTensor` API? > > > > > > The calculation for...

This is stale, closing.

> Can you identify which test is failing and when it was added and if it ever worked? From my testing it seems to only be the two in `train_models.py`,...

It was added in [this commit](https://github.com/llvm/torch-mlir/commit/10c8e3c5936c43eff31bfe929eeffa2996494ea9). I haven't been able to make it work even reverting to that commit but I'll ask @pashu123 if he verified it when he wrote...

The Pytorch op doesn't require the extra batch dimension, right? It should just be a matter of adding one or two extra nested for loops to the logic, i.e. ```...

Hmm. In that case could we just handle the case where the group count is constant? That'd mean going from `? -> [G, ?]`, and it's how we're already handling...

So if I'm understanding correctly, the named op should look something like [this](https://github.com/llvm/llvm-project/commit/19f2e0fc981529eb5d5746ae6aa5f5e0b13122aa), and the main issue is that actually expanding the NCHW/FCHW tensors to fit into the new op...

I've split the PRs, this one is for masked_fill now.