[Bugfix] Add MTP for opanpangu_pro_moe model, fix an initialization bug in StaticSinkAttention

[yt0428]

Purpose

This PR further add MTP for opanpangu_pro_moe model #28775 , and it also fix an initialization bug in StaticSinkAttention

For MTP support, the major modification is that we need to do a shallow copy for spec_decode_common_attn_metadata in gpu_model_runner.py. The block_table_tensor of common_metadata may be modified during the building of StaticSinkAttention, this will, in turn, affect spec_decode_common_attn_metadata, as it is a direct reference to common_metadata. A simple shallow copy can avoid this.

As for the initialization bug in StaticSinkAttention, we move the init of CustomOp to the beginning of the initialization to avoid the refresh of member variables.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request introduces support for the opanpangu_pro_moe model and addresses an initialization bug in StaticSinkAttention. The changes include adding the new model type to speculative configuration and model architecture convertors, as well as refining the weight loading process for the openpangu_mtp model. A critical fix involves ensuring a shallow copy of attention metadata to prevent unintended side effects during speculative decoding initialization.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

[yt0428]

@DarkLight1337 @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg
Hello, could you please give some reviews about this small PR? Many Thanks!!!

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @yt0428.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @yt0428.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[LucasWilkinson]

can we move this copy into StaticSinkAttentionBuilder?

[yt0428]

Unfortunately the builder is not responsible for the building of spec_decode_common_attn_metadata, it is handled by gpu_model_runner outside.

[github-actions]

This pull request has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this pull request should remain open. Thank you!

[github-actions]

This pull request has been automatically closed due to inactivity. Please feel free to reopen if you intend to continue working on it. Thank you!