[Bugfix] Revert max_prompt_len validation for decoder-only models.

[davidheineman]

A bugfix (#16156) improved input validation for multi-modal models.

However for decoder-only models it changed:

if len(prompt_ids) > max_prompt_len:

to

if len(prompt_ids) >= max_prompt_len:

This change breaks inputs for decoder-only models that provide inputs equal to max_prompt_len, which is valid behavior. This breaks LLM eval frameworks like OLMES and the LM Eval harness. See:

FIX #16445

This will revert this change to the original behavior (len(prompt_ids) > max_prompt_len). This fixes the v1 engine as well.

[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.

🚀

[DarkLight1337]

Nice catch!

[DarkLight1337]

Nice catch!