[Bugfix] Fix use_cascade_attention handling for Alibi-based models on vllm/v1

[h-sugi]

When using Alibi-based models like MPT, the following assertion error causes.

AssertionError: Cascade attention does not support ALiBi.

This is because that the determination logic for use_cascade in gpu_model_runner.py incorrectly uses a hard-coded use_alibi=False. Therefore, cascade_attention is wrongly enabled.

This PR allows setting use_alibi based on the alibi configuration specified in the config.json for MPT models.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[WoosukKwon]

@h-sugi Thanks for the PR! Left some minor comments.

[mergify]

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

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

[WoosukKwon]

@h-sugi Thanks for updating the PR! I will merge if the tests go well.