MFU calculation for Quantized Workloads #2893
slabhs-aws
started this conversation in
General
Replies: 1 comment
-
|
See previous discussions here #560. In particular, I don't think there's a single metric which can represent e2e perf perfectly when mixed precision is used. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
MFU calculation is skipped for Quantized workloads here. Do we have plans for supporting it? We are especially interested for MXFP8 quantized workloads (via TorchAO)
Beta Was this translation helpful? Give feedback.
All reactions