Quinn Dawkins

Dynamic shape handling for aten.index.Tensor

3

I'm looking to add support for dynamic shapes with multiple indexing tensors for the aten.index.Tensor op. The semantics for the op are summarized as follows. ``` IndexTensor for multiple input...

[RFC] Shape handling for output lists of tensors

8

For split-like ops the output type is a list of (potentially different shape) tensors which is not able to be handled by the current shape refinement pass. The solution introduced...

[RFC] Shape handling for `aten::_native_multi_head_attention`

1

Hi, I have been trying to look into adding a decomposition for the `aten::_native_multi_head_attention` op. The issue I have run into is that for certain inputs (specifically `need_weights=False`) the op...

Includes tests for generated IR and E2E. Only handles folding for constant string cases.

[TORCH] add flag to prevent demotion of doubles to floats

2

Needed for an accurate variance calculation for the torch-v-diffusion model - If there is a preferred and/or more robust way to include this please let me know

Hi, the issue I'm having is the same as in #411. The previous issue was dispatching a similar kernel to the one below with a single thread ``` __kernel void...

Initialize conversion passes from ONNX to Torch-MLIR backend contract

16

Includes LLVM lit tests for ONNXAddOp and ONNXConstantOp. This PR is a result of the RFC discussed in #1639. The following overview of the PR is collected from there, but...

[RFC] Adding an ONNX to Torch conversion

44

Hello everyone, in a recent RFC in torch-mlir a conversion from ONNX dialect to the Torch dialect in torch-mlir was proposed (https://github.com/llvm/torch-mlir/issues/1255). After feedback we are looking at adding that...

[RFC] Support for an embedded build of Torch-MLIR

4

Hi all, With the recent RFC in ONNX-MLIR for a conversion path from ONNX-MLIR to Torch-MLIR (#1255 and then https://github.com/onnx/onnx-mlir/issues/1639) there is a use case for building Torch-MLIR as part...