[Eagle] Fix attn_mask index out of range in high concurrency situations

[wxsIcey]

What this PR does / why we need it?

• Fixes [Bug]: Multiple calls (maybe >100) to eagle3-qwen3-8b often incurs "attn_mask index out of range" #3024

Does this PR introduce any user-facing change?

N/A

How was this patch tested?

 python -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --served-model-name Eagle3 --port 8000  --model Qwen/Qwen3-8B   --seed 42     -tp 1  --speculative_config '{"model": "Tengyunw/qwen3_8b_eagle3", "draft_tensor_parallel_size": 1, "num_speculative_tokens": 5, "method": "eagle3"}'

Co-authored-by: liuruijin17 ricklrj@outlook.com

• vLLM version: v0.10.2
• vLLM main: vllm-project/vllm@52d0cb8

[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.

[gemini-code-assist]

Code Review

This pull request addresses an index out of range error related to attn_mask in high-concurrency scenarios. The fix involves pre-allocating the attention mask cache to its maximum possible size by using max_model_len, which effectively prevents race conditions during dynamic resizing. This also simplifies the code by removing a dependency on an environment variable and its corresponding os import. While the change is effective, I have raised a concern about the potential for increased memory consumption, which could be significant for models with a large max_model_len.

[wangxiyuan]

@liuruijin17 can you take a look this PR? Thanks.

[wxsIcey]

Please see: https://github.com/vllm-project/vllm-ascend/blob/main/vllm_ascend/worker/model_runner_v1.py#L333
In model runner, when constructing AttentionMask, vllm_config.model_config.max_model_len is also used