Umang Yadav
Umang Yadav
> I don't know how to make a test that would hit this case Is it possible to make a unit-test for the pass itself ? Test that would pass...
@pfultz2 @hgaspar @krzysz00 can you guys review and comment if this is correct ?
> What do you mean by expose? Are we adding this as a new data type? Yes, e.g. here https://github.com/ROCm/AMDMIGraphX/blob/3c0c782617ecd3554a2d17a9145b7bc015592a24/src/include/migraphx/shape.hpp#L65 We already have ref implementation for E4M3FN type and some...
There are two possibilities for the parameters and return values. 1. Parameters and/or return value are in OCP FP8. In that case we would require special operator. 2. Model has...
> Test of test_layernorm_large with this change show very similar numbers: (on banff-cyxtera-s81-2) > > With this change: `# bin/test_verify test_layernorm_large [ RUN ] test_layernorm_large [ COMPLETE ] test_layernorm_large (12240.9ms)...
This can be closed now i think. Split-k GEMM is integrated and enabled in MIGraphX
@fxmarty We don't have that logic implemented in MIGraphX yet. I have an issue open that describes the logic https://github.com/ROCm/AMDMIGraphX/issues/2717 It's pretty simple to convert, but haven't had chance to...
Closing this one for now untill we have more data or need
Fix for this issue should work for all the hardwares includign MI300. e.g. #2506 attempted fix for this by adding simplication for nested converts but it didnt' work on Mi300.
@lakhinderwalia FYI