Toby Roseman

I have not gone through the code. Before we can proceed here, we need code to reproduce the `grid_sample` issue (that doesn't involve loading `.pt` files). What error are you...

@Suraj209211 - I don't understand your last few messages. In order to make progress, we need step to reproduce this issue which does not involved loading `.pt` files.

@Suraj209211 - the code in your notebook doesn't look right. You're not passing the PyTorch and Core ML model the same input. So of course the output will be different....

In order to help you, we need to be able to reproduce the issue. Can you give us a minimal example to reproduce this issue? One with doesn't require loading...

Using `in_feats=torch.rand([1, 64000])` sounds good. However loading a `.pt` files, even one from Hugging Face, is insecure. Can you do some further investigation and give us PyTorch code that we...

I can reproduce the issue using your attached mlmodel. However I'm unable to create a MIL program that reproduces these results. It looks like you're using the `iOS15` opset, but...

> It seems like a 'pad' input is required for 'same'. That's not what I'm seeing. If you change my code: `pad_type='same_lower'` -> `pad_type='same'`, it works without specifying `pad`. >...

I've reviewed some internal ML Program configuration files. There are separate iOS15 and iOS16 ops for both `conv` and `conv_transpose`. The only difference between these versions seems to be that...

I'm not sure I understand your question. You should be able to use [the Builder API](https://apple.github.io/coremltools/docs-guides/source/model-intermediate-language.html?highlight=mb#create-a-mil-program) directly.

I would start by reviewing this document: https://machinelearning.apple.com/research/neural-engine-transformers