fix(torchtitan): correct MFU on MI350X/MI355X via peak FLOPS patch#772
Merged
Conversation
fix(torchtitan): correct MFU on MI350X/MI355X via peak FLOPS patch TorchTitan computes MFU as num_flops_per_token * tps / gpu_peak_flops, where gpu_peak_flops comes from a hardcoded device-name table in torchtitan.tools.utils.get_peak_flops. The vendored TorchTitan does not know about the AMD MI350 series (gfx950), so the lookup falls through to the A100 fallback (312 TFLOPS), making the MFU denominator ~8x too small and reporting impossible values (e.g. 563%). Throughput (TFLOP/s) itself was correct; only MFU was affected. MI300X was unaffected since it is present in the upstream table. Add a setup-phase patch that wraps get_peak_flops and returns the correct BF16 dense peaks from AMD's product pages: - MI355X: 2500 TFLOPS (matches upstream TorchTitan) - MI350X: 2300 TFLOPS (not yet covered upstream) The patch runs before MetricsProcessor caches gpu_peak_flops and delegates all other devices to the original implementation, so it is safe to keep after the vendored TorchTitan is updated.
WangLingxun
requested review from
Xiaoming-AMD,
limou102 and
wenxie-amd
as code owners
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
TorchTitan computes MFU as num_flops_per_token * tps / gpu_peak_flops, where gpu_peak_flops comes from a hardcoded device-name table in torchtitan.tools.utils.get_peak_flops. The vendored TorchTitan does not know about the AMD MI350 series (gfx950), so the lookup falls through to the A100 fallback (312 TFLOPS), making the MFU denominator ~8x too small and reporting impossible values (e.g. 563%). Throughput (TFLOP/s) itself was correct; only MFU was affected. MI300X was unaffected since it is present in the upstream table.
Add a setup-phase patch that wraps get_peak_flops and returns the correct BF16 dense peaks from AMD's product pages:
The patch runs before MetricsProcessor caches gpu_peak_flops and delegates all other devices to the original implementation, so it is safe to keep after the vendored TorchTitan is updated.