Added the verification of num_max_dispatch_tokens_per_rank to the decode operator adaptation layer.

[oagniqgnat]

Motivation

If the SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK environment variable is configured on the framework side, it will be passed as the input parameter num_max_dispatch_tokens_per_rank of the decode operator to the operator for each rank to determine the size of the tensor and HCCL buffer to be allocated. However, the current logic cannot determine whether the SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK configured for each rank is large enough when the batch size is dynamic. Therefore, an assert check is added at the operator entry to ensure that the max_token configured for each rank is large enough.

Modification

Added the verification of num_max_dispatch_tokens_per_rank to the decode operator adaptation layer.

Benchmarking and Profiling

Not involved

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

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 (/tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci) or contact authorized users to do so.
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @oagniqgnat, 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 enhances the robustness of the decode operator adaptation layer by implementing a critical validation check. It ensures that the system's maximum dispatch token capacity per rank is always sufficient for the tokens being processed, thereby preventing potential issues related to token handling. Concurrently, the global batch size calculation has been streamlined, leveraging this new guarantee for a more direct and efficient computation.

Highlights

• Validation Added: A new host-side assertion (EP_HOST_ASSERT) has been introduced to verify that num_max_dispatch_tokens_per_rank is always greater than or equal to the current number of tokens (new_x.size(0)), ensuring proper token dispatch limits.
• Global Batch Size Calculation Refinement: The calculation for global_bs has been simplified by removing the std::max operation, now directly using num_max_dispatch_tokens_per_rank multiplied by num_ranks. This change reflects the new guarantee provided by the added assertion.

🧠 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 adds a verification for num_max_dispatch_tokens_per_rank in the low_latency_dispatch function and simplifies the calculation of global_bs accordingly. While adding verification is a good practice, the current implementation introduces a critical regression that can cause a crash when handling empty inputs (i.e., zero tokens). I have provided a detailed comment and a code suggestion to address this issue by making the logic robust for empty inputs while retaining the intended verification.