fix(grpc): update vLLM imports for inputs reorganization

[CatherineSue]

Description

Problem

vLLM reorganized its inputs module in vllm-project/vllm#35182 (commit ba2f0acc2, 2026-03-25), renaming and relocating several symbols. This change first appeared in v0.18.2rc0 and is now officially released in v0.19.0:

• vllm.multimodal.inputs.mm_inputs → vllm.inputs.engine.mm_input
• vllm.multimodal.inputs.MultiModalInputs → vllm.inputs.engine.MultiModalInput
• vllm.inputs.token_inputs → vllm.inputs.engine.tokens_input

The gRPC servicer was importing the old names, which breaks on vLLM >= v0.19.0.

Solution

Update the gRPC servicer imports to use the new module paths and renamed symbols from vllm.inputs.engine.

Changes

• Update grpc_servicer/smg_grpc_servicer/vllm/servicer.py:
  • Import MultiModalInput (was MultiModalInputs) from vllm.inputs.engine
  • Import mm_input (was mm_inputs) from vllm.inputs.engine
  • Import tokens_input (was token_inputs) from vllm.inputs.engine
  • Update type annotations and function calls to use the new names

Test Plan

• Verify imports resolve correctly against vLLM v0.19.0
• Run existing gRPC servicer tests to ensure no regression

Checklist

• cargo +nightly fmt passes
• cargo clippy --all-targets --all-features -- -D warnings passes
• (Optional) Documentation updated
• (Optional) Please join us on Slack #sig-smg to discuss, review, and merge PRs

Summary by CodeRabbit

• Refactor
  • Updated internal tokenization and multimodal input handling to align with the newer engine-level interfaces, improving compatibility and maintainability.
• Chores
  • Bumped the minimum vLLM dependency to a more recent release to match the updated interfaces.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: 7ca3a81f-0c26-4cf5-a93f-d6583046a0a0

📥 Commits

Reviewing files that changed from the base of the PR and between a4a79f7 and 194028c.

📒 Files selected for processing (1)

• grpc_servicer/pyproject.toml

---

📝 Walkthrough

Walkthrough

Replaced vLLM input constructors in the gRPC servicer: switched tokenized prompt import to vllm.inputs.engine.tokens_input and swapped multimodal input construction/type from mm_inputs/VllmMultiModalInputs to mm_input/VllmMultiModalInput in _build_preprocessed_mm_inputs. (≤50 words)

Changes

Cohort / File(s)	Summary
vLLM input constructor updates grpc_servicer/smg_grpc_servicer/vllm/servicer.py	Replaced vllm.inputs.token_inputs import with vllm.inputs.engine.tokens_input; changed _build_preprocessed_mm_inputs to construct multimodal inputs via vllm.inputs.engine.mm_input and updated return annotation from VllmMultiModalInputs to VllmMultiModalInput; preserved argument structure (prompt_token_ids, mm_kwargs, mm_hashes, mm_placeholders, prompt).
Dependency bump grpc_servicer/pyproject.toml	Raised optional vllm extra from vllm>=0.17.0 to vllm>=0.19.0.

Sequence Diagram(s)

(omitted)

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• refactor(grpc): use EngineClient interface instead of AsyncLLM in vLLM servicer #949 — Modifies the same servicer input construction to use engine-layer vLLM APIs in servicer.py.
• refactor(grpc): optimize vLLM servicer and remove duplicate server.py #662 — Alters _build_preprocessed_mm_inputs and related multimodal input handling.

Suggested labels

dependencies

Suggested reviewers

• njhill
• slin1237

Poem

🐰 Hopped through code with nimble paws,
Swapped old imports for fresher laws,
Tokens trimmed and mm stitched tight,
Engine-ready now — what a sight!
A tiny hop, then off to write.

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Title check	✅ Passed	The title accurately describes the main change: updating vLLM imports in the gRPC servicer due to vLLM's inputs module reorganization. It is specific, clear, and directly related to the primary objective of the PR.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/vllm-mm-inputs-rename

---

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

[gemini-code-assist]

Code Review

This pull request updates the vLLM multimodal input handling by migrating from the deprecated mm_inputs and MultiModalInputs to the newer mm_input and MultiModalInput classes from vllm.inputs.engine. These changes include updating imports, return type hints, and function calls within the _build_preprocessed_mm_inputs method. I have no feedback to provide.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@grpc_servicer/smg_grpc_servicer/vllm/servicer.py`:
- Around line 18-19: The project imports VllmMultiModalInput, mm_input, and
tokens_input from vllm (in grpc_servicer/smg_grpc_servicer/vllm/servicer.py),
which require vllm>=0.19.0; update the dependency constraint in
grpc_servicer/pyproject.toml by changing the vllm requirement to vllm =
["vllm>=0.19.0"] so the installed vllm release matches the new API
reorganization.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: c8443587-019b-4c52-8cea-1408800ea702

📥 Commits

Reviewing files that changed from the base of the PR and between 57a30b4 and a4a79f7.

📒 Files selected for processing (1)

• grpc_servicer/smg_grpc_servicer/vllm/servicer.py