refactor(serve): split chat_completions endpoint into a package

[lvhan028]

Motivation

Improve the maintainence of v1/chat/completions

Modification

• Splits chat_completions.py into protocol/validation/logprobs/logits_processors/ serving modules.

• Chat-specific models move to endpoints/chat_completions/protocol.py; shared models stay in the top-level protocol.py with backward-compat re-exports.

[Copilot]

Pull request overview

Refactors the OpenAI /v1/chat/completions endpoint implementation from a single large module into a dedicated chat_completions package, aligning its layout with the existing responses/ package and adding migration tests to enforce backward-compatible import paths.

Changes:

• Introduces lmdeploy.serve.openai.endpoints.chat_completions as a package with split-out protocol, validation, logprobs, logits_processors, and serving modules.
• Moves chat-specific Pydantic models into endpoints/chat_completions/protocol.py while keeping shared models in lmdeploy.serve.openai.protocol.
• Adds migration equivalence tests to assert package structure and re-export/class-identity invariants.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_chat_completions_package_migration.py	Adds structural migration tests for package layout + model re-export identity.
lmdeploy/serve/openai/protocol.py	Removes chat-specific models and re-exports them for backward compatibility.
lmdeploy/serve/openai/endpoints/chat_completions/validation.py	Extracts request validation logic for chat completions.
lmdeploy/serve/openai/endpoints/chat_completions/serving.py	Updates chat completions serving module to use the new split modules.
lmdeploy/serve/openai/endpoints/chat_completions/protocol.py	New canonical home for chat-specific Pydantic models.
lmdeploy/serve/openai/endpoints/chat_completions/logprobs.py	Extracts logprobs construction helpers.
lmdeploy/serve/openai/endpoints/chat_completions/logits_processors.py	Extracts logit-bias logits processor.
lmdeploy/serve/openai/endpoints/chat_completions/init.py	Adds lazy register export to avoid circular imports.
lmdeploy/serve/openai/endpoints/init.py	Adds lazy create_openai_router export to avoid circular imports.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated no new comments.

Suppressed comments (2)

lmdeploy/serve/openai/protocol.py:428

• __getattr__ is newly introduced here but lacks a return type annotation, unlike the similar lazy exports added in endpoints/__init__.py and chat_completions/__init__.py. Adding -> Any improves consistency and static analysis without changing runtime behavior.

def __getattr__(name: str):

tests/test_chat_completions_package_migration.py:18

• The NOTE claims import lmdeploy.serve.openai.endpoints.chat_completions as pkg breaks because lmdeploy/__init__.py rebinds serve. Since lmdeploy/serve is a real subpackage, importing a submodule will overwrite the parent attribute, so this explanation is likely misleading. Consider updating the comment to reflect the actual caveat (e.g., that from lmdeploy import serve resolves to the deprecated function) or remove the claim about dotted imports being broken.

# NOTE: We use importlib.import_module rather than the dotted-`as` form
# ``import lmdeploy.serve.openai.endpoints.chat_completions as pkg`` because
# lmdeploy/__init__.py rebinds the ``serve`` attribute to a deprecated
# function (lmdeploy.api.serve), which shadows the ``lmdeploy.serve``
# subpackage attribute and breaks dotted-`as` imports. The submodule itself