Skip to content

Chunk expensive numerical operations - #1016

Merged
ValterH merged 2 commits into
mem-opt/base-numerical-chunksfrom
mem-opt/4-chunk-numerical-ops
Sep 29, 2026
Merged

ValterH merged 2 commits into
mem-opt/base-numerical-chunksfrom
mem-opt/4-chunk-numerical-ops

Conversation

@ValterH

@ValterH ValterH commented Sep 28, 2026 •

Copy link
Copy Markdown
Collaborator

Split from #996 to keep each review focused on one behavior or optimization.

Add sdm._memory.split_size to balance chunks within the chunk memory limit, with a minimum of one item. Use it to compute RobustScale quantiles over column chunks and ClipSigma transforms over row chunks.

Chunking applies only on CUDA without autograd; the gradient path remains unchunked.

Depends on #1013 and #1014. The dependency-only base isolates this commit for review; rebase and retarget to main after both dependencies merge. Split from #996; related to #994.

@copy-pr-bot

copy-pr-bot Bot commented Sep 28, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

JingangQu and others added 2 commits September 29, 2026 00:09
- Compute RobustScale quantiles over column chunks and ClipSigma
  transforms over row chunks within the chunk memory limit.
- Keep the unchunked ClipSigma transform where gradients are required.

Signed-off-by: Jingang Qu <jqu@nvidia.com>
Co-authored-by: Cedric Lorenz <clorenz@nvidia.com>
Co-authored-by: Valter Hudovernik <vhudovernik@nvidia.com>
@ValterH
ValterH force-pushed the mem-opt/4-chunk-numerical-ops branch from 067a774 to 2f59f57 Compare September 28, 2026 22:09
@ValterH
ValterH added this pull request to stack #1021 September 28, 2026 22:10
@cedric-lorenz
cedric-lorenz force-pushed the mem-opt/4-chunk-numerical-ops branch from 2223079 to 2f59f57 Compare September 29, 2026 11:07
@ValterH
ValterH merged commit 20e572b into mem-opt/base-numerical-chunks Sep 29, 2026
8 of 11 checks passed
@ValterH
ValterH deleted the mem-opt/4-chunk-numerical-ops branch September 29, 2026 12:09
@ValterH
ValterH restored the mem-opt/4-chunk-numerical-ops branch September 29, 2026 12:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants