Thiago Crepaldi

Oh no, there is a typo in the ONNX test! It should actually be `24`, not `14`! ```python import torch col = torch.tensor([[[1.0, 6.0, 11.0, 16.0, 21.0, 26, 31, 36,...

That is odd, on [ONNX main branch](https://github.com/onnx/onnx/blob/d6d03e0d9feea5bcfee7fa218802dceae37197c8/onnx/backend/test/case/node/col2im.py#L207), this is correct. 24 is there and was fixed by @liqun at https://github.com/onnx/onnx/pull/4769 It is safe to ignore this failure

> What's the real code actually doing? It seems reasonable for us to graph break here. My scenario is for full graph :) On my actual model, the user used...

@gramalingam @shubhambhokare1 should we complete or close this PR?

This is a repro for the schema error bug reported by https://github.com/onnx/onnx/issues/6098

maybe this? https://github.com/marketplace/actions/setup-cpp-c-c

> Thanks for the suggestion, I agree it would be nice to have such a feature, but I don't think this is easily possible to add it. First of all,...

> The code you show here has nothing at all to do with `accelerator.prepare()` though, or the `Accelerator` class at all. cc @SunMarc is there something we can do about...

> > The .to(device) calls are removed when Accelerate is used, so it shouldn't be a problem. > > I don't understand this. If we deactivate accelerate, the data won't...

Thanks @muellerzr `low_cpu_mem_usage=False` kind of works for the specific scenario I got. I understand the issues with fixing th generic problem at 2), so I will try to focus on...