Jeff Daily

icon indicating an email [email protected]

icon company AMD icon location Richland, WA

Results 210 comments of Jeff Daily

[ROCm] Add extra cuda_to_hip_mappings.py

@pytorchbot merge

[ROCm] Add extra cuda_to_hip_mappings.py

@pytorchbot merge

[ROCm] Add extra cuda_to_hip_mappings.py

pytorchbot merge didn't close this automatically. Closing manually.

Convert `ForeachFuncInfo` to `dataclass`

ciflow/rocm is added to this PR. It should provide a ROCm signal now. Thank you @huydhn for the revert and adding of the label.

[ROCm] TunableOp for scaled_mm

> Thanks @jeffdaily ! Do you have some benchmark performance number with this PR on FP8 scaled mm? I do not, I was deferring to any benchmarks you were running.

[ROCm] TunableOp for scaled_mm

> Hi @jeffdaily , looks like the speedup on FP8 is minor. Is it because hipBLASLt is not well tuned yet? That would be my assumption, but let's check with...

[ROCm] TunableOp for scaled_mm

@pytorchbot merge

[ROCm] Implement forward AD for miopen_batch_norm

@andrewor14 How long will it take for #119496 to land? If this PR is ready now and the other is not, I'm in favor of landing this one. Would doing...

[ROCm] Implement forward AD for miopen_batch_norm

@andrewor14 or @albanD, since it sounds decided, can we get an approval so we can merge?

[ROCm] Implement forward AD for miopen_batch_norm

@pytorchbot merge