Aflowers/bump vllm 0.11.0 debug

[alec-flowers]

Overview:

Do Not Merge

Details:

Where should the reviewer start?

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

• closes GitHub issue: #xxx

Summary by CodeRabbit

• New Features
  • Updated AI runtime components to newer minor versions, improving compatibility and stability.
• Refactor
  • Aligned internal imports with upstream module changes; no user-facing behavior impacted.
• Tests
  • Adjusted test initialization timing to improve stability during automated runs.
• Chores
  • Refreshed build configurations for containerized environments to match updated component versions.

[coderabbitai]

Walkthrough

Bumps VLLM and FlashInfer versions in container build and install scripts, updates a PromptLogprobs import path in an example utility, and adds a long sleep in pytest configuration.

Changes

Cohort / File(s)	Summary of Change
Container version bumps container/Dockerfile.vllm, container/deps/vllm/install_vllm.sh	Update VLLM ref v0.10.2 → v0.11.0 and FlashInfer ref v0.3.0 → v0.3.1; remove historical cherry-pick block and committer env from install script; add EOF newline.
Example import adjustment examples/multimodal/utils/protocol.py	Import PromptLogprobs from vllm.logprobs instead of vllm.sequence; no runtime logic changes.
Test startup delay tests/conftest.py	Add import time and a 4-hour time.sleep(60 * 60 * 4) in pytest_configure.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  participant Runner as pytest Runner
  participant Conftest as tests/conftest.py

  Runner->>Conftest: import and run pytest_configure()
  Note over Conftest: Introduced delay
  Conftest->>Conftest: time.sleep(4 hours)
  Conftest-->>Runner: return
  Runner->>Runner: collect and execute tests

Loading

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

A bunny taps the Docker drum—thump-thump, versions rise,
FlashInfer blinks, vLLM blithely reaches new skies.
A hop in imports, tidy trails aligned,
But tests now nap for hours—time redefined!
I’ll brew some clover tea and wait, ears high, wide-eyed.

Pre-merge checks

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description Check	⚠️ Warning	The PR description follows the template headings but contains only placeholder text and a “Do Not Merge” note without any actual summary or details. It fails to inform reviewers about the modifications or where they should focus.	Replace all placeholders under Overview, Details, and Where should the reviewer start with concrete descriptions of the changes. List the key files to review, remove the “Do Not Merge” warning, and reference the actual issue number if applicable.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title Check	✅ Passed	The title indicates bumping VLLM to version 0.11.0 but includes an unnecessary “Aflowers/” prefix and the term “debug,” and it omits the related FlashInfer version bump. It thus refers to a real aspect of the change but is noisy and does not clearly convey the full scope.

---

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 and usage tips.}

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

container/deps/vllm/install_vllm.sh (1)

201-201: Update stale version reference in precompiled wheel URL.

The VLLM_PRECOMPILED_WHEEL_LOCATION still references version 0.10.2 instead of 0.11.0. This will cause the script to download the wrong wheel when building from source on AMD64.

Apply this diff to update the version reference:

-        export VLLM_PRECOMPILED_WHEEL_LOCATION="https://vllm-wheels.s3.us-west-2.amazonaws.com/${VLLM_REF}/vllm-0.10.2-cp38-abi3-manylinux1_x86_64.whl"
+        export VLLM_PRECOMPILED_WHEEL_LOCATION="https://vllm-wheels.s3.us-west-2.amazonaws.com/${VLLM_REF}/vllm-0.11.0-cp38-abi3-manylinux1_x86_64.whl"

Alternatively, verify the exact wheel filename by running:

#!/bin/bash
# Description: Verify the correct wheel filename for VLLM 0.11.0
aws s3 ls s3://vllm-wheels/v0.11.0/ --region us-west-2 --no-sign-request | grep -E 'cp38-abi3-manylinux.*x86_64.whl'

🧹 Nitpick comments (1)

examples/multimodal/utils/protocol.py (1)

26-26: Remove unused noqa directive.

Static analysis reports that the # noqa: F401 directive on this line is no longer needed.

As per static analysis hints

Apply this diff:

-from vllm.multimodal.inputs import MultiModalUUIDDict  # noqa: F401
+from vllm.multimodal.inputs import MultiModalUUIDDict

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 60975b5 and 7d09e62.

📒 Files selected for processing (4)

• container/Dockerfile.vllm (1 hunks)
• container/deps/vllm/install_vllm.sh (4 hunks)
• examples/multimodal/utils/protocol.py (1 hunks)
• tests/conftest.py (2 hunks)

🧰 Additional context used

🪛 Ruff (0.13.3)

examples/multimodal/utils/protocol.py

26-26: Unused noqa directive (non-enabled: F401)

Remove unused noqa directive

(RUF100)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: Build and Test - dynamo

🔇 Additional comments (4)

examples/multimodal/utils/protocol.py (1)

25-25: LGTM! Import path aligns with VLLM 0.11.0.

The import source change for PromptLogprobs from vllm.sequence to vllm.logprobs correctly reflects the module restructuring in VLLM 0.11.0.

container/Dockerfile.vllm (1)

18-20: LGTM! Version bumps are consistent.

The VLLM (v0.10.2 → v0.11.0) and FlashInfer (v0.3.0 → v0.3.1) version updates align with the PR objectives and are consistent with changes in container/deps/vllm/install_vllm.sh.

container/deps/vllm/install_vllm.sh (2)

16-16: LGTM! Version bumps are consistent.

The VLLM (v0.10.2 → v0.11.0) and FlashInfer (v0.3.0 → v0.3.1) version updates align with the PR objectives and are consistent with changes in container/Dockerfile.vllm.

Also applies to: 32-32

---

134-135: LGTM! Cherry-pick block appropriately replaced with placeholder.

The removal of the historical git cherry-pick and replacement with a TODO comment is appropriate for future maintenance.

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

Remove debug sleep blocking test execution.

The 4-hour delay in pytest_configure will block all tests from running. This appears to be a debug artifact left in the code.

Apply this diff to remove the debug sleep:

-import time
-
 import pytest
 
 from tests.utils.constants import TEST_MODELS

 def pytest_configure(config):
-    time.sleep(60 * 60 * 4)
     # Defining model morker to avoid `'model' not found in `markers` configuration option`
     # error when pyproject.toml is not available in the container
     config.addinivalue_line("markers", "model: model id used by a test or parameter")

Also applies to: 28-29

🤖 Prompt for AI Agents

In tests/conftest.py around line 20 (and also lines 28-29), there is a debug
time.sleep call that blocks test execution for hours; remove the time.sleep(...)
calls so pytest_configure no longer sleeps, and if the only use of the time
module was for these sleeps, also remove the import time line to avoid an unused
import.