[Pooling] Honor max_embed_len for chunked embeddings

[taneem-ibrahim]

Purpose

Chunked embedding requests ignored an explicitly configured max_embed_len because tokenization unconditionally used an unlimited token count. This change honors the configured limit before chunk creation. It only affects explicitly configured limits; max_embed_len=None retains existing unlimited chunked processing. MRV2 execution and valid embedding outputs are unchanged.

Reproducer

Configuration:

max_model_len=4
enable_chunked_processing=true
max_embed_len=8
input_tokens=9

On Main:

{'max_total_tokens': None, 'accepted_tokens': 9}

On this branch:

HTTP 400 BadRequestError

An eight-token boundary request still returns HTTP 200.

Test Plan

46 focused pooling tests passed
5 MRV2 long-text E2E tests passed
MRV2 HTTP probe: 8 tokens → 200; 9 tokens → 400
File-scoped pre-commit passed

AI Assistance

OpenAI Codex (GPT-5) assisted with drafting the change

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Team

Run ID: a6edd130-0812-45db-a288-12894ad47962

📥 Commits

Reviewing files that changed from the base of the PR and between 144e79c and abe148e.

📒 Files selected for processing (1)

• vllm/entrypoints/pooling/base/protocol.py

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • Chunked embedding requests now enforce the configured maximum embedding length during tokenization, preventing unbounded token processing.

Walkthrough

Chunked pooling tokenization now limits max_total_tokens to pooler_config.max_embed_len instead of leaving the limit unset.

Changes

Pooling tokenization

Layer / File(s)	Summary
Apply chunk embedding limit vllm/entrypoints/pooling/base/protocol.py	The chunked-processing branch sets max_total_tokens to pooler_config.max_embed_len.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Merge Risk: ⚪ Minimal · up to abe14

Chunked embedding requests now reject inputs exceeding an explicitly configured embedding limit while preserving unlimited and non-chunked behavior. No current merge-blocking risk remains.

Suggested reviewers: kernelclint, jperezdealgaba

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 1 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description check	✅ Passed	The description clearly explains the fix to honor max_embed_len for chunked embedding requests. It also states the behavior for max_embed_len=None and summarizes validation results.
Title check	✅ Passed	The title clearly and concisely identifies the main change: honoring max_embed_len for chunked embeddings.

• Fix all pre-merge checks with AI

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}