Add pydocstyle to the ruff rules (#52136)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Signed-off-by: Harry Mellor <hej.mellor@gmail.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5
parent
804e5377aa
commit
8c1557a79c
@@ -25,9 +25,7 @@ logger = init_logger(__name__)
|
||||
|
||||
|
||||
class CompilerInterface:
|
||||
"""
|
||||
The interface for a compiler that can be used by vLLM.
|
||||
"""
|
||||
"""The interface for a compiler that can be used by vLLM."""
|
||||
|
||||
# The name of the compiler, e.g. inductor.
|
||||
# This is a class-level attribute.
|
||||
@@ -36,8 +34,7 @@ class CompilerInterface:
|
||||
def initialize_cache(
|
||||
self, cache_dir: str, disable_cache: bool = False, prefix: str = ""
|
||||
) -> None:
|
||||
"""
|
||||
when the vLLM process uses `cache_dir` as the cache directory,
|
||||
"""When the vLLM process uses `cache_dir` as the cache directory,
|
||||
the compiler should initialize itself with the cache directory,
|
||||
e.g. by re-directing its own cache directory to a sub-directory.
|
||||
|
||||
@@ -52,8 +49,7 @@ class CompilerInterface:
|
||||
pass
|
||||
|
||||
def compute_hash(self, vllm_config: VllmConfig) -> str:
|
||||
"""
|
||||
Gather all the relevant information from the vLLM config,
|
||||
"""Gather all the relevant information from the vLLM config,
|
||||
to compute a hash so that we can cache the compiled model.
|
||||
|
||||
See [`VllmConfig.compute_hash`][vllm.config.VllmConfig.compute_hash]
|
||||
@@ -71,8 +67,7 @@ class CompilerInterface:
|
||||
compile_range: Range,
|
||||
key: str | None = None,
|
||||
) -> tuple[Callable[..., Any] | None, Any | None]:
|
||||
"""
|
||||
Compile the graph with the given example inputs and compiler config,
|
||||
"""Compile the graph with the given example inputs and compiler config,
|
||||
with a range. The `compile_range` specifies the range of the inputs,
|
||||
it could be concrete size (if compile_sizes is provided), e.g. [4, 4]
|
||||
or a range [5, 8].
|
||||
@@ -105,8 +100,7 @@ class CompilerInterface:
|
||||
graph_index: int,
|
||||
compile_range: Range,
|
||||
) -> Callable[..., Any]:
|
||||
"""
|
||||
Load the compiled function from the handle.
|
||||
"""Load the compiled function from the handle.
|
||||
Raises an error if the handle is invalid.
|
||||
|
||||
The handle is the second return value of the `compile` function.
|
||||
@@ -115,8 +109,7 @@ class CompilerInterface:
|
||||
|
||||
|
||||
class AlwaysHitShapeEnv:
|
||||
"""
|
||||
Why do we need this class:
|
||||
"""Why do we need this class:
|
||||
|
||||
For normal `torch.compile` usage, every compilation will have
|
||||
one Dynamo bytecode compilation and one Inductor compilation.
|
||||
@@ -160,7 +153,8 @@ def _get_vllm_functorch_config() -> dict[str, Any]:
|
||||
"""Return the functorch config overrides that vLLM applies at compile time.
|
||||
|
||||
Used by both set_functorch_config() and get_inductor_factors() to ensure
|
||||
the compile-time config and cache key are always consistent."""
|
||||
the compile-time config and cache key are always consistent.
|
||||
"""
|
||||
cfg: dict[str, Any] = {}
|
||||
if not envs.VLLM_USE_MEGA_AOT_ARTIFACT:
|
||||
cfg["bundled_autograd_cache"] = False
|
||||
@@ -249,8 +243,7 @@ def _patch_standalone_compile_atomic_save() -> None:
|
||||
|
||||
|
||||
class InductorStandaloneAdaptor(CompilerInterface):
|
||||
"""
|
||||
The adaptor for the Inductor compiler.
|
||||
"""The adaptor for the Inductor compiler.
|
||||
Requires PyTorch 2.8+.
|
||||
This is not on by default yet, but we plan to turn it on by default for
|
||||
PyTorch 2.8.
|
||||
@@ -447,9 +440,7 @@ class InductorStandaloneAdaptor(CompilerInterface):
|
||||
|
||||
|
||||
class InductorAdaptor(CompilerInterface):
|
||||
"""
|
||||
The adaptor for the Inductor compiler, version 2.5, 2.6, 2.7.
|
||||
"""
|
||||
"""The adaptor for the Inductor compiler, version 2.5, 2.6, 2.7."""
|
||||
|
||||
name = "inductor"
|
||||
|
||||
@@ -727,8 +718,7 @@ class InductorAdaptor(CompilerInterface):
|
||||
return compiled_graph
|
||||
|
||||
def metrics_context(self) -> contextlib.AbstractContextManager[Any]:
|
||||
"""
|
||||
This method returns the Dynamo metrics context (if it exists,
|
||||
"""This method returns the Dynamo metrics context (if it exists,
|
||||
otherwise a null context). It is used by various compile components.
|
||||
Present in torch>=2.6, it's used inside FxGraphCache in
|
||||
torch==2.6 (but not after). It might also be used in various other
|
||||
|
||||
Reference in New Issue
Block a user