[Bugfix] Fix the attn_metadata is None

[dragondream-chen]

What this PR does / why we need it?

Fix the bug " TypeError: 'NoneType' object is not iterable' " in vllm_ascend/compilation/acl_graph.py
The reason of that is the attn_metadata is none in the dummy_run of MTP.

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.12.0
• vLLM main: vllm-project/vllm@ad32e3e

[gemini-code-assist]

Code Review

This pull request addresses a TypeError that occurs when attn_metadata is None during a dummy run in MTP with ACL graph mode. The change correctly ensures that attn_metadata is constructed when aclgraph_runtime_mode is CUDAGraphMode.FULL, regardless of the skip_attn flag. This makes the logic more robust and resolves the potential for a crash. The fix is correct and well-targeted.

[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.

[yiz-liu]

Please update the name of skip_attn, as its current name no longer accurately reflects its semantics. It would be more appropriate to rename it to in_graph_capturing or a similarly descriptive term.

[dragondream-chen]

Please update the name of skip_attn, as its current name no longer accurately reflects its semantics. It would be more appropriate to rename it to in_graph_capturing or a similarly descriptive term.

DONE. The old name is indeed not very easy to understand.

[yiz-liu]

I’m afraid there is a misunderstanding here: skip_attn does not correspond to in_graph_capturing, in fact, it represents the opposite behavior. Please take a closer look at the underlying logic and revise it accordingly as soon as possible.
@dragondream-chen