perf: calculate grad on-the-fly for SiLUT #4678

caic99 · 2025-03-28T09:24:11Z

Current implementation of SiLUT involves one extra storage for saving the 1st-order gradient. This PR reduces the memory footprint by calculating the 1st-order gradient on-the-fly in silut_double_backward. It introduces an overhead of ~0.5% of calculation time.

I've tested this PR on OMat with 9 DPA-3 layers and batch size=auto:512.

Metric	Before	After	Improvement
Peak Memory	25.0G	21.7G	-13%
Speed (per 100 steps)	30.29s	30.46s	-0.56%

The correctness of this modification is covered by source/tests/pt/test_custom_activation.py.

Summary by CodeRabbit

Refactor
- Streamlined internal computation logic by refining variable naming for clarity.
- Updated public method signatures to return outputs in a structured tuple, ensuring more intuitive and consistent integration.

Copilot

Pull Request Overview

This PR refactors the SiLUT gradient computation to reduce memory consumption by calculating the first-order gradient on-the-fly.

Refactored silut_forward to use a temporary variable renamed from "tanh_part" to "tanh".
Changed silut_backward to return a single gradient tensor instead of a tuple.
Updated silut_double_backward and its usage in SiLUTGradFunction to return a tuple with improved on-the-fly gradient calculation.

Comments suppressed due to low confidence (1)

deepmd/pt/utils/utils.py:122

[nitpick] The variable name 'grad_mul_grad_grad_output' is unclear. Consider renaming it to better reflect its purpose, for example 'grad_second', to improve code readability.

grad_input, grad_mul_grad_grad_output = silut_double_backward_script(

coderabbitai · 2025-03-28T09:26:41Z

📝 Walkthrough

Walkthrough

The changes modify internal gradient computation functions in the deepmd/pt/utils/utils.py file. Variable names have been updated for clarity—specifically renaming tanh-related variables—and computation logic has been simplified. The silut_double_backward function now returns a tuple instead of a single tensor, and the associated backward method in the SiLUTGradFunction class has been adjusted to save only the necessary tensors and return matching gradient values.

Changes

File(s)	Change Summary
`deepmd/pt/.../utils.py`	- silut_forward: Renamed `tanh_part` to `tanh`. - silut_backward: Renamed `tanh_term` to `tanh` and updated `grad_tanh` calculation to use `tanh * tanh`; simplified the return statement to return only `grad * grad_output`. - silut_double_backward: Changed return type from `torch.Tensor` to `tuple[torch.Tensor, torch.Tensor]`, removed `grad_grad`, added `grad_grad_tanh` computation, and updated the return tuple. - SiLUTGradFunction.backward: Adjusted saved tensors (now only `x` and `grad_output`) and updated its return values to match the modified double backward calculations.

Suggested reviewers

njzjz

📜 Recent review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 9e1d69d and a54533d.

📒 Files selected for processing (1)

deepmd/pt/utils/utils.py (3 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

deepmd/pt/utils/utils.py

⏰ Context from checks skipped due to timeout of 90000ms (30)

GitHub Check: CodeQL
GitHub Check: Test Python (6, 3.12)
GitHub Check: Test Python (5, 3.12)
GitHub Check: Test Python (5, 3.9)
GitHub Check: Test Python (4, 3.12)
GitHub Check: Test Python (4, 3.9)
GitHub Check: Test Python (3, 3.12)
GitHub Check: Test Python (3, 3.9)
GitHub Check: Test Python (2, 3.12)
GitHub Check: Test Python (2, 3.9)
GitHub Check: Test Python (1, 3.12)
GitHub Check: Test Python (1, 3.9)
GitHub Check: Analyze (python)
GitHub Check: Analyze (javascript-typescript)
GitHub Check: Analyze (c-cpp)
GitHub Check: Build wheels for cp310-manylinux_aarch64
GitHub Check: Build wheels for cp311-win_amd64
GitHub Check: Build wheels for cp311-macosx_arm64
GitHub Check: Build C++ (clang, clang)
GitHub Check: Build wheels for cp311-macosx_x86_64
GitHub Check: Build C++ (rocm, rocm)
GitHub Check: Build C++ (cuda120, cuda)
GitHub Check: Build C library (2.14, >=2.5.0rc0,<2.15, libdeepmd_c_cu11.tar.gz)
GitHub Check: Build wheels for cp311-manylinux_x86_64
GitHub Check: Build C library (2.18, libdeepmd_c.tar.gz)
GitHub Check: Build C++ (cuda, cuda)
GitHub Check: Build wheels for cp311-manylinux_x86_64
GitHub Check: Test C++ (false)
GitHub Check: Build C++ (cpu, cpu)
GitHub Check: Test C++ (true)

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>, please review it.
- Generate unit testing code for this file.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai generate unit testing code for this file.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
- @coderabbitai read src/utils.ts and generate unit testing code.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
- @coderabbitai help me debug CodeRabbit configuration file.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai full review to do a full review from scratch and review all the files again.
@coderabbitai summary to regenerate the summary of the PR.
@coderabbitai generate docstrings to generate docstrings for this PR.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai plan to trigger planning for file edits and PR creation.
@coderabbitai configuration to show the current CodeRabbit configuration for the repository.
@coderabbitai help to get help.

Other keywords and placeholders

Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (`.coderabbit.yaml`)

You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
Please see the configuration documentation for more information.
If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

deepmd/pt/utils/utils.py

Co-authored-by: Duo <50307526+iProzd@users.noreply.github.com> Signed-off-by: Chun Cai <amoycaic@gmail.com>

codecov · 2025-03-29T09:43:15Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.80%. Comparing base (0918b22) to head (249219e).
Report is 79 commits behind head on devel.

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #4678      +/-   ##
==========================================
- Coverage   84.80%   84.80%   -0.01%     
==========================================
  Files         692      692              
  Lines       66396    66402       +6     
  Branches     3539     3538       -1     
==========================================
+ Hits        56306    56310       +4     
- Misses       8949     8951       +2     
  Partials     1141     1141

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

perf: calculate grad on-the-fly for SiLUT

9e1d69d

caic99 requested review from Copilot and iProzd March 28, 2025 09:24

github-actions bot added the Python label Mar 28, 2025

Copilot AI reviewed Mar 28, 2025

View reviewed changes

iProzd requested changes Mar 28, 2025

View reviewed changes

deepmd/pt/utils/utils.py Outdated Show resolved Hide resolved

Update deepmd/pt/utils/utils.py

a54533d

Co-authored-by: Duo <50307526+iProzd@users.noreply.github.com> Signed-off-by: Chun Cai <amoycaic@gmail.com>

caic99 requested a review from iProzd March 28, 2025 09:38

iProzd approved these changes Mar 29, 2025

View reviewed changes

Merge branch 'devel' into perf-silut

249219e

njzjz added this pull request to the merge queue Mar 29, 2025

Merged via the queue into deepmodeling:devel with commit 52f8ece Mar 29, 2025
60 checks passed

caic99 deleted the perf-silut branch March 30, 2025 05:27

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf: calculate grad on-the-fly for SiLUT #4678

perf: calculate grad on-the-fly for SiLUT #4678

Uh oh!

caic99 commented Mar 28, 2025 •

edited by coderabbitai bot

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

coderabbitai bot commented Mar 28, 2025 •

edited

Loading

Walkthrough

Changes

Suggested reviewers

Chat

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

CodeRabbit Configuration File (`.coderabbit.yaml`)

Documentation and Community

Uh oh!

Uh oh!

codecov bot commented Mar 29, 2025 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

perf: calculate grad on-the-fly for SiLUT #4678

perf: calculate grad on-the-fly for SiLUT #4678

Uh oh!

Conversation

caic99 commented Mar 28, 2025 • edited by coderabbitai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Uh oh!

coderabbitai bot commented Mar 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Suggested reviewers

Chat

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

CodeRabbit Configuration File (.coderabbit.yaml)

Documentation and Community

Uh oh!

Uh oh!

codecov bot commented Mar 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

caic99 commented Mar 28, 2025 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Mar 28, 2025 •

edited

Loading

CodeRabbit Configuration File (`.coderabbit.yaml`)

codecov bot commented Mar 29, 2025 •

edited

Loading