[FusedDeepMoe] Delete padding

[kaniel-outis]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @kaniel-outis, 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 refactors the fused_deep_moe function by eliminating an internal padding and unpadding mechanism. The change streamlines the function's execution by ensuring that it operates directly on the provided input tensors without any internal modifications for padding, suggesting a shift in how padding is managed or a simplification of the data flow within the system.

Highlights

• Padding Logic Removal: The explicit padding and unpadding logic for input tensors (x, expert_ids, and expert_scales_optional) within the fused_deep_moe function has been entirely removed, simplifying the function's internal operations.
• Direct Input Utilization: The aclnnFusedDeepMoe command now directly uses the original, unpadded input tensors (x, expert_ids, and expert_scales_optional), eliminating the need for intermediate padded versions.
• Batch Size Calculation Update: The calculation for the batch size (bs) has been updated to directly reference expert_ids.size(0), reflecting the removal of the internal padding mechanism that previously used a padded index tensor.

Ignored Files

• Ignored by pattern: .github/workflows/** (1)
  • .github/workflows/pr-test-npu.yml

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 removes padding logic from the fused_deep_moe function. The changes are mostly correct, removing the padding and unpadding blocks and updating the call to aclnnFusedDeepMoe. However, I found a critical issue where a variable (new_topk_idx) that was previously initialized in the removed padding logic is still being used, which will cause the code to use stale data. Please see my comment for details on how to fix this.

[gemini-code-assist]

You've correctly updated this line to use expert_ids instead of this->new_topk_idx. A similar change is required on line 619 for the global_bs calculation. Since the padding logic that set new_topk_idx was removed, it now holds a stale value within this function, which will lead to incorrect behavior.

Please update line 619 to use expert_ids as well:

int64_t global_bs = std::max(expert_ids.size(0), num_max_dispatch_tokens_per_rank) * num_ranks;

[kaniel-outis]

fixed, int64_t global_bs = std::max(expert_ids.size(0), num_max_dispatch_tokens_per_rank) * num_ranks;