Skip to content

[NPU] Adapt cann 8.5: use sfa and lightning indexer op from cann#17489

Closed
randgun wants to merge 1 commit intosgl-project:mainfrom
randgun:main
Closed

[NPU] Adapt cann 8.5: use sfa and lightning indexer op from cann#17489
randgun wants to merge 1 commit intosgl-project:mainfrom
randgun:main

Conversation

@randgun
Copy link
Copy Markdown
Contributor

@randgun randgun commented Jan 21, 2026 •
edited
Loading

Motivation

There are some bugs of sparse flash attention and lightning indexer in custom_ops package. Fixed ops are intergrated in CANN 8.5 so we update these ops and call them using torch_npu.

lightning_indexer: https://gitcode.com/cann/ops-transformer/blob/master/attention/lightning_indexer/README.md
sfa: https://gitcode.com/cann/ops-transformer/blob/master/attention/sparse_flash_attention/README.md

Modifications

Accuracy Tests

image

Benchmarking and Profiling

Checklist

Review Process

  1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
  2. Get approvals from CODEOWNERS and other reviewers.
  3. Trigger CI tests with comments or contact authorized users to do so.
    • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
  4. After green CI and required approvals, ask Merge Oncalls to merge.

@github-actions github-actions Bot added the npu label Jan 21, 2026
@gemini-code-assist
Copy link
Copy Markdown
Contributor

gemini-code-assist Bot commented Jan 21, 2026

Summary of Changes

Hello @randgun, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request focuses on adapting the NPU backend to ensure compatibility with CANN 8.5. It involves updating the invocation of core NPU operators for sparse flash attention and lightning indexing to align with changes in the torch_npu library's API. This ensures the system can leverage the latest NPU toolkit features and maintain correct functionality.

Highlights

  • NPU Operator Migration: The pull request updates the usage of NPU-specific operators, migrating calls from torch.ops.custom to torch_npu for both npu_sparse_flash_attention and npu_lightning_indexer.
  • API Adaptation for NPU Operators: The npu_sparse_flash_attention operator now returns multiple values, requiring unpacking (e.g., attn_out, _, _ = ...), and new parameters (attention_mode=2, return_softmax_lse=False) have been added to its calls. Similarly, npu_lightning_indexer now returns a tuple, necessitating indexing ([0]) to retrieve the primary output.
  • Removed custom_ops Dependency: The explicit import and associated warning for the custom_ops module have been removed, indicating that the functionalities previously provided by custom_ops are now integrated directly into torch_npu.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature Command Description
Code Review /gemini review Performs a code review for the current pull request in its current state.
Pull Request Summary /gemini summary Provides a summary of the current pull request in its current state.
Comment @gemini-code-assist Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help /gemini help Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

  1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist[bot]
gemini-code-assist Bot reviewed Jan 21, 2026
View reviewed changes
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request updates the NPU backend operations to align with the CANN 8.5 API. Specifically, it migrates from torch.ops.custom.npu_sparse_flash_attention and torch.ops.custom.npu_lightning_indexer to their torch_npu counterparts. The changes also correctly adapt to the new return signatures of these operations, which now return multiple values, by unpacking and using only the relevant first element. Additionally, new parameters attention_mode=2 and return_softmax_lse=False are added to the sparse flash attention calls. The removal of the custom_ops import is a good cleanup, simplifying dependencies. The changes are consistent and appear to be a necessary adaptation to the updated CANN library.

@randgun randgun changed the title Adapt cann 8.5: use sfa and lightning indexer op from cann [NPU] Adapt cann 8.5: use sfa and lightning indexer op from cann Jan 21, 2026
@ping1jing2 ping1jing2 mentioned this pull request Jan 23, 2026
Merged
@randgun randgun closed this Jan 23, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@merrymercy merrymercy Awaiting requested review from merrymercy

@Fridge003 Fridge003 Awaiting requested review from Fridge003 Fridge003 is a code owner

@ispobock ispobock Awaiting requested review from ispobock

@Qiaolin-Yu Qiaolin-Yu Awaiting requested review from Qiaolin-Yu

@hebiao064 hebiao064 Awaiting requested review from hebiao064

@ping1jing2 ping1jing2 Awaiting requested review from ping1jing2 ping1jing2 is a code owner

@iforgetmyname iforgetmyname Awaiting requested review from iforgetmyname iforgetmyname is a code owner

1 more reviewer

@gemini-code-assist gemini-code-assist[bot] gemini-code-assist[bot] left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

npu

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@randgun