Siavash Nazari

Yes. Requirements is the typical Torch-MLIR steps in [development.md](https://github.com/llvm/torch-mlir/blob/main/development.md) with the only addition of installing QPyTorch using `pip install qtorch`. More detailed info on setting up QPyTorch is linked [here](https://github.com/Tiiiger/QPyTorch#installation).

Since the next steps in the investigation needed changes in QPyTorch side, I created [this fork](https://github.com/Svoch/QPyTorch/), I also had to update the script a little bit due to the changes...

Below is the IR I get lowering the model above to Torch Dialect. If I've understood correctly, even though rewriters for [AtenMmOp](https://github.com/llvm/torch-mlir/blob/5888c4f7dc5aad504ef8c77442ebbf3c0393623c/lib/Conversion/TorchToTosa/TorchToTosa.cpp#L1497) and [AtenLinearOp](https://github.com/llvm/torch-mlir/blob/5888c4f7dc5aad504ef8c77442ebbf3c0393623c/lib/Conversion/TorchToTosa/TorchToTosa.cpp#L1533) do exist in TorchToTosa lowering pass,...

@silvasean - Yes. with the Torch to TOSA conversion of the Transpose Op merged, this issue can be marked as fixed. Please note that the symptom, i.e. `error: failed to...

Here is the stack trace of the error I'm hitting: ``` RuntimeError: Arguments for call are not valid. The following variants are available: aten::__contains__.int_list(int[] l, int item) -> (bool): Expected...

Thank you @silvasean! This is very helpful, and it makes sense why the op is not being picked up. Let me modify the registration method somehow and I'll update this...

I was able to modify QPyTorch operation bindings such that `block_quantize_nearest` and `block_quantize_stochastic` are registered via `TORCH_LIBRARY` and not `PYBIND11` anymore. Seems like Torch-MLIR is able to pick up these...

More details on steps to reproduce this error: #### 1. QPyTorch Modifications The modifications for QPyTorch operator bindings is present on the `master` branch in [this QPyTorch fork](https://github.com/Svoch/QPyTorch). The steps...

Sounds great @silvasean! This is very exciting update on the custom Op support front, and looks like a very good timing. I'll follow the example for the ExampleIdentity custom Op...

Thank you @silvasean and @rdadolf for the details! Quick update: I kept hitting some KeyErrors in `registry.py` running `update_torch_ods.sh`. I think it _may_ be due to a torch versioning mismatch...