[main][bugfix] Fix fullgraph padding bug in mtp eagle refactor

[lilinsiman]

What this PR does / why we need it?

The condition for determining padding in the fullgraph overlay with MTP and PCP has been modified to accommodate corner cases where the shape capture size is manually specified.

Does this PR introduce any user-facing change?

no

How was this patch tested?

ut and tests

• vLLM version: v0.13.0
• vLLM main: vllm-project/vllm@2f4e654

[gemini-code-assist]

Code Review

This pull request addresses a bug in the padding logic for full graph mode, specifically in scenarios involving MTP (Multi-Token Prediction) and PCP (Prefill Context Parallelism). The modification correctly constrains the padding condition by ensuring that the number of input tokens does not exceed the maximum size of a captured graph. By using min(max_decode_tokens, self.cudagraph_batch_sizes[-1]), the change prevents the padding logic from being erroneously applied to batches that are too large for graph replay and would fall back to eager execution. This is a solid bug fix that enhances the robustness of the full graph execution path, particularly for cases with manually configured graph capture sizes.

[yiz-liu]

I suggest

max_decode_tokens = min(self.scheduler_config.max_num_seqs * self.uniform_decode_query_len, self.cudagraph_batch_sizes[-1])

It's more resonable.

[lilinsiman]

done

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.