AMDMIGraphX

AMDMIGraphX copied to clipboard

Add flag to dump mlir modules to mxrs

3

pfultz2

Upgrade CI to 6.1.3

3

pfultz2

Bump ci rocm6.2

4

Updating CI to ROCm 6.2 and other appropriate dockers

TedThemistokleous

Introduce `--int4-weights` option in `migraphx-driver`. This would require changes in MIGraphX's naive quantizer to set range between `[0, 15]` During quantization it should also insert "pack" and "unpack" instructions.

1

lakhinderwalia

[INT4] Compress model by quantizing weights to int4

25

### Idea Use int4 as the compression technique to fit larger models onto Navi machines or possibly MI series machines. Weights would be compressed using encoding scheme that would pack...

umangyadav

Add ONNX parsing for SimplifiedLayerNormalization

6

turneram

This will help lower batch sizes, need to determine which models it benefits. Needed for exec_dashboard models

causten

Bitwise_and operator

5

* Implements the bitwise_and operator and ONNX parser. * Needed for support in TorchMIGraphX models

CharlieL7

Allow multiple outputs for the MLIR + Pointwise fusions

10

At present when doing `mlir + pointwise` fusion, It is not checking if "conv/gemm" has only use or not. Therefore for the contrived case, it can lead to pessimization where...

umangyadav

Update CI point for Onnxruntime builds - Updated with changes from 2024-08-02 - Auto-generated by [create-pull-request][1] - Update .onnxrt-commit to d0a6f57d74a62be07314780aa061d138200936d0 [1]: https://github.com/peter-evans/create-pull-request

github-actions[bot]