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
+12 -16
View File
@@ -69,6 +69,7 @@ def resolve_layer_fused_shared_expert(
Raises:
ValueError: If requested shared-expert fusion is quantization-incompatible.
"""
# NOTE: is_fusion_moe_shared_experts_enabled is decorated with @if_aiter_supported
# that returns None if AITER is not available.
@@ -164,20 +165,19 @@ def _count_expert_num_tokens(
def count_expert_num_tokens(
topk_ids: torch.Tensor, num_local_experts: int, expert_map: torch.Tensor | None
) -> torch.Tensor:
"""
Count the number to tokens assigned to each expert.
"""Count the number to tokens assigned to each expert.
Parameters:
- topk_ids (torch.Tensor): Tensor mapping each token to its
list of experts.
- num_local_experts (int): Number of experts in this rank.
- expert_map (Optional[torch.Tensor]): A tensor mapping expert indices
from the global expert space to the local expert space of the expert
parallel shard.
Args:
topk_ids (torch.Tensor): Tensor mapping each token to its list of experts.
num_local_experts (int): Number of experts in this rank.
expert_map (Optional[torch.Tensor]): A tensor mapping expert indices
from the global expert space to the local expert space of the expert
parallel shard.
Returns:
A tensor of size num_local_experts, where tensor[i] holds the number
of tokens assigned to the ith expert.
"""
assert topk_ids.dtype.is_signed, "The kernel uses -1 to represent invalid topk_ids"
expert_num_tokens = torch.empty(
@@ -202,8 +202,7 @@ def count_expert_num_tokens(
def _resize_cache(x: torch.Tensor, v: tuple[int, ...]) -> torch.Tensor:
"""
Shrink the given tensor and apply the given view to it. This is
"""Shrink the given tensor and apply the given view to it. This is
used to resize the intermediate fused_moe caches.
"""
assert prod(v) <= x.numel(), (
@@ -226,8 +225,7 @@ def _fp8_quantize(
per_act_token: bool,
block_shape: list[int] | None = None,
) -> tuple[torch.Tensor, torch.Tensor]:
"""
Perform fp8 quantization on the inputs. If a block_shape
"""Perform fp8 quantization on the inputs. If a block_shape
is provided, the output will be blocked.
"""
if block_shape is None:
@@ -252,11 +250,9 @@ def _int8_quantize(
per_act_token: bool,
block_shape: list[int] | None = None,
) -> tuple[torch.Tensor, torch.Tensor]:
"""
Perform int8 quantization on the inputs. If a block_shape
"""Perform int8 quantization on the inputs. If a block_shape
is provided, the output will be blocked.
"""
# If weights are per-channel (per_channel_quant=True), then
# activations apply per-token quantization. Otherwise, assume
# activation tensor-wise fp8/int8 quantization, dynamic or static