[0.13.0][Bugfix] fix npu memory is not released in cp

[weiguihua2]

What this PR does / why we need it?

The npu memory is not released in the qwen context parallel chunk scenario. This pr resolve it.

The attention backend uses multiple streams. Variables created in the main stream are directly used in the communication stream. As a result, the NPU GPU memory of the variables is not released. This problem can be avoided by using the clone mode.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

CI passed. The fix was verified by observing that NPU memory is now correctly released in the described scenario.

pick-from #6480

[gemini-code-assist]

Summary of Changes

Hello @weiguihua2, 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 resolves a critical NPU memory leak occurring in the context parallel chunking scenario, specifically when qwen is used. The issue stemmed from variables created in the main stream being directly used in a communication stream, preventing their memory from being released. By implementing a cloning mechanism, the PR ensures that memory associated with these variables is correctly freed, improving resource management and stability.

Highlights

• NPU Memory Release: Introduced a .clone() operation on local_context_output before its use in _gather_global_context_output within the cp_chunkedprefill_comm_stream context. This ensures proper memory release for NPU variables when multiple streams are involved, addressing a memory leak.

🧠 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]

Code Review

This pull request addresses a memory leak on NPU devices within the context parallel attention mechanism. The issue occurs when a tensor created in the main computation stream is used in a separate communication stream, preventing its memory from being released. The fix involves cloning the tensor before passing it to the communication stream, which correctly decouples their lifecycles and resolves the memory leak. The change is straightforward and includes a helpful comment explaining the rationale.

To align with the repository's style guide, I suggest updating the pull request title and summary as follows:

Suggested PR Title:

[Attention][BugFix] Fix NPU memory leak in context parallel attention

Suggested PR Summary:

### What this PR does / why we need it?
This pull request resolves an NPU memory leak that occurs during chunked prefill in context parallel (`CP`) scenarios.

In the attention backend, variables created in the main computation stream were being used directly in the communication stream. This cross-stream usage prevented the NPU memory for these variables from being released correctly. This patch fixes the issue by cloning the tensor before it's used in the communication stream, which ensures proper memory management.

### Does this PR introduce _any_ user-facing change?
No.

### How was this patch tested?
CI passed. The fix was verified by observing that NPU memory is now correctly released in the described scenario.