add hybrid model PD to NIXL connector

[nealvaidya]

Motivation

Add hybrid model support to NIXL connector based on the mooncake implementation

Modifications

Accuracy Tests

Benchmarking and Profiling

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 @nealvaidya, 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 integrates hybrid model support, particularly for Mamba states, into the NIXL connector. The changes involve modifying core data transfer structures to accommodate state-related information, implementing a dedicated RDMA transfer function for Mamba states, and updating the transfer lifecycle management to track the receipt and completion of these new state data types. This enhancement allows for more flexible and efficient disaggregation of hybrid models.

Highlights

• Hybrid Model State Support: Extended TransferInfo and KVArgsRegisterInfo to include fields for handling hybrid model states, specifically dst_state_indices and dst_state_data_ptrs, enabling the NIXL connector to manage these new data types.
• Mamba State Transfer Mechanism: Introduced a new _send_mamba_state method to facilitate the efficient transfer of Mamba model states via RDMA, ensuring specialized handling for this hybrid model type.
• State Transfer Status Tracking: Enhanced TransferStatus with received_state_per_pp and expects_state to monitor the completion of state data transfers across different pipeline parallel ranks, improving the robustness of the transfer process.
• Dynamic Buffer Registration: Added logic to register_buffer_to_engine to dynamically register state/extra pool data buffers if they are present, allowing the system to adapt to varying model requirements.

🧠 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 introduces support for transferring Mamba state data via RDMA within the disaggregation system. Key changes include adding dst_state_indices and dst_state_data_ptrs to TransferInfo and KVArgsRegisterInfo respectively, updating ZMQ message parsing to handle these new fields, and modifying TransferStatus to track state transfer progress. The KVSender class now includes a _send_mamba_state method for RDMA transfer and add_transfer_request is updated to initiate state transfers for Mamba models. A review comment points out a potential IndexError in _send_mamba_state because dst_state_indices is accessed at index 0 without a length check, even though it can be an empty list, and suggests adding an assertion for its length similar to prefill_state_indices.

[gemini-code-assist]

The code asserts the length of prefill_state_indices but not dst_state_indices. However, dst_state_indices is accessed at index 0 on line 621. If dst_state_indices is an empty list, this will cause an IndexError. It's possible for dst_state_indices to be empty as it's initialized to [] in TransferInfo.from_zmq if not present in the message. An assertion should be added to ensure dst_state_indices has at least one element before it's accessed.

Suggested change

	assert len(prefill_state_indices) == 1, "Mamba should have single state index"
	assert len(prefill_state_indices) == 1, "Mamba should have single state index"
	assert len(dst_state_indices) == 1, "Mamba should have single state index"