Fix minor docs issues and fix metric requests

[SriRangaTarun]

Fixes some minor issues in the docs, issue #15394, and issue #21014.

[gemini-code-assist]

Code Review

This pull request includes minor documentation fixes and updates to how request metrics are handled. The change from time.time() to time.monotonic() for recording arrival times is a good improvement for accuracy in time-interval measurements.

However, I've identified a critical issue in vllm/v1/engine/output_processor.py. The new logic for populating request metrics does not account for the case where statistics logging is disabled, which will cause a crash. I've provided a suggestion to fix this by adding a conditional check. Please address this to ensure the stability of the system.

[gemini-code-assist]

The new code to populate request_output.metrics assumes that req_state.stats is always available. However, req_state.stats is initialized to None if log_stats is False (see RequestState.__init__).

This will lead to an AttributeError when trying to access req_state.stats.arrival_time, causing a crash when log_stats is disabled.

To prevent this, you should add a check to ensure req_state.stats is not None before attempting to access its attributes.

Suggested change

	request_output.metrics = RequestMetrics(
	arrival_time=req_state.stats.arrival_time,
	last_token_time=req_state.stats.last_token_ts,
	first_scheduled_time=req_state.stats.scheduled_ts,
	first_token_time=req_state.stats.first_token_ts,
	time_in_queue=req_state.stats.scheduled_ts - req_state.stats.arrival_time,
	finished_time=time.monotonic()
	)
	if req_state.stats:
	request_output.metrics = RequestMetrics(
	arrival_time=req_state.stats.arrival_time,
	last_token_time=req_state.stats.last_token_ts,
	first_scheduled_time=req_state.stats.scheduled_ts,
	first_token_time=req_state.stats.first_token_ts,
	time_in_queue=req_state.stats.scheduled_ts - req_state.stats.arrival_time,
	finished_time=time.monotonic()
	)

[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.

🚀