[TPU] Calculate block size only when not set.

[QiliangCui]

The change calculates the block size from max-model-len and tries to use as the largest block size below 256. While in some configuration, smaller block size may give better performance.

This change is to re-enable the argument block-size and fallback to to PallasAttentionBackend.get_page_size when block-size is not explicit set.

[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.

🚀

[yaochengji]

Should we add warning when the block_size set by user is not optimal? cc @bythew3i

[QiliangCui]

If we know what value is optimal for the user, I think we should add.

The reason I want to put back the block-size parameter is that I found the value from PallasAttentionBackend.get_page_size is not a good value for me when my max-model-len is large.

In my benchmark testing, my max-model-len is 4096 which results a 256 from get_page_size. But block_size=16 gives me better result.