Skip to content

Fix FLOPs inconsistency and add using_sparse_model flag #21743

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Devesh-Maheshwari wants to merge 10 commits into
base: master
Choose a base branch
from
+230 −144
Open

Fix FLOPs inconsistency and add using_sparse_model flag #21743

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
3a45223
added arguments
Devesh-Maheshwari May 26, 2026
6ac9f57
Fix Hopper FLOPs inconsistency and add using_sparse_model flag
Devesh-Maheshwari May 26, 2026
ed10cee
Merge branch 'master' into fix-flop-inconsistency
deependujha May 27, 2026
7e2bda6
Update src/lightning/fabric/utilities/throughput.py
Devesh-Maheshwari May 27, 2026
8f09c93
made _CUDA_FLOPS consistent with a*10^b so it is consistent
Devesh-Maheshwari May 27, 2026
bb8c49d
resolved the reviewers comments
Devesh-Maheshwari May 27, 2026
9f98086
Added changed logs
Devesh-Maheshwari Jun 1, 2026
2485332
Merge branch 'master' into fix-flop-inconsistency
deependujha Jun 1, 2026
94dc34c
update
deependujha Jun 1, 2026
19b8e9c
update
deependujha Jun 1, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions src/lightning/fabric/CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,11 +8,11 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).

### Added

-
- Added `using_sparse_model` and `sparse_cuda_acceleration_factor` parameters to `Throughput` so MFU defaults to the dense peak and opts into the sparse peak explicitly ([#21743](https://github.com/Lightning-AI/pytorch-lightning/pull/21743))

### Changed

-
- Normalized `_CUDA_FLOPS` values in `throughput.py` to consistent scientific notation (`a.bcde+N` with `1 <= a < 10`) ([#21743](https://github.com/Lightning-AI/pytorch-lightning/pull/21743))
Comment thread
deependujha marked this conversation as resolved.
Outdated

### Removed

Expand Down
Loading
Loading