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:
Harry Mellor
2026-09-16 16:44:29 +00:00
committed by GitHub
co-authored by Claude Opus 5
parent 804e5377aa
commit 8c1557a79c
2002 changed files with 15024 additions and 14269 deletions
@@ -69,7 +69,8 @@ class NCCLTrainerInitInfo(TrainerInitInfo):
propagates them to the worker at `trainer_init` so the two sides cannot
disagree. Note this defaults to packed, unlike the worker-side
`NCCLWeightTransferInitInfo`, whose default only applies when no trainer
ships a value. `backend` is the factory dispatch key."""
ships a value. `backend` is the factory dispatch key.
"""
backend: ClassVar[str] = "nccl"
@@ -113,8 +114,7 @@ class NCCLWeightTransferUpdateInfo(WeightTransferUpdateInfo):
class NCCLWeightTransferEngine(
WeightTransferEngine[NCCLWeightTransferInitInfo, NCCLWeightTransferUpdateInfo]
):
"""
Weight transfer engine using NCCL for communication between trainer and workers.
"""Weight transfer engine using NCCL for communication between trainer and workers.
This implementation uses NCCL broadcast operations to transfer dense
checkpoint-format weights from the trainer (rank 0) to all inference workers
@@ -142,14 +142,14 @@ class NCCLWeightTransferEngine(
self.packed_num_buffers = DEFAULT_PACKED_NUM_BUFFERS
def init_transfer_engine(self, init_info: NCCLWeightTransferInitInfo) -> None:
"""
Initialize NCCL process group with the trainer and record the
"""Initialize NCCL process group with the trainer and record the
trainer-supplied wire params so the worker decodes exactly as the
trainer encodes.
Args:
init_info: NCCL initialization info containing master address, port,
rank offset, world size, and the packed wire params
"""
self.packed = init_info.packed
self.packed_buffer_size_bytes = init_info.packed_buffer_size_bytes
@@ -175,8 +175,7 @@ class NCCLWeightTransferEngine(
finalize_layerwise_reload(self.model, self.model_config)
def receive_weights(self, update_info: NCCLWeightTransferUpdateInfo) -> None:
"""
Receive weights from trainer via NCCL broadcast.
"""Receive weights from trainer via NCCL broadcast.
Whether to use packed broadcasting (and the buffer geometry) is read
from `self.packed` / `self.packed_*`, set at the init handshake from the
@@ -186,6 +185,7 @@ class NCCLWeightTransferEngine(
Args:
update_info: NCCL update info containing parameter names, dtypes,
and shapes
"""
if self.model_update_group is None:
raise RuntimeError(
@@ -365,7 +365,8 @@ class NCCLTrainerWeightTransferEngine(TrainerWeightTransferEngine[NCCLTrainerIni
def _broadcast(self, source: WeightSource, meta: list[ParamMeta]) -> None:
"""Iterate the source (materializing each tensor — a collective on all
ranks) and, on the sender, broadcast from rank 0, packed or one-by-one.
Non-sender ranks only replay the iteration to stay in the collective."""
Non-sender ranks only replay the iteration to stay in the collective.
"""
if not self.is_sender:
for _ in source:
pass