Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-12 09:51:31 -07:00
committed by GitHub
parent 9bb38130cb
commit 8fcaaf6a16
944 changed files with 9490 additions and 10121 deletions
+7 -8
View File
@@ -29,7 +29,6 @@ physical experts.
import time
from collections.abc import Sequence
from dataclasses import dataclass
from typing import Optional, Union
import torch
from torch.distributed import ProcessGroup, all_reduce
@@ -186,9 +185,9 @@ class EplbState:
model: MixtureOfExperts,
device: torch.device,
parallel_config: ParallelConfig,
global_expert_load: Optional[torch.Tensor] = None,
old_global_expert_indices: Optional[torch.Tensor] = None,
rank_mapping: Optional[dict[int, int]] = None,
global_expert_load: torch.Tensor | None = None,
old_global_expert_indices: torch.Tensor | None = None,
rank_mapping: dict[int, int] | None = None,
) -> "EplbState":
"""
Build the initial EPLB state.
@@ -439,9 +438,9 @@ class EplbState:
model: MixtureOfExperts,
is_profile: bool = False,
execute_shuffle: bool = True,
global_expert_load: Optional[torch.Tensor] = None,
rank_mapping: Optional[dict[int, int]] = None,
) -> Optional[torch.Tensor]:
global_expert_load: torch.Tensor | None = None,
rank_mapping: dict[int, int] | None = None,
) -> torch.Tensor | None:
"""
Rearrange the experts according to the current load.
"""
@@ -611,7 +610,7 @@ class EplbState:
def _node_count_with_rank_mapping(
pg: Union[ProcessGroup, StatelessProcessGroup],
pg: ProcessGroup | StatelessProcessGroup,
rank_mapping: dict[int, int],
) -> int:
if isinstance(pg, ProcessGroup):
+1 -2
View File
@@ -8,7 +8,6 @@ This involves the exchange of expert weights between GPUs.
from collections.abc import Iterable, MutableSequence, Sequence
from functools import partial
from typing import Optional
import torch
from torch.distributed import (
@@ -253,7 +252,7 @@ def rearrange_expert_weights_inplace(
expert_weights: Sequence[Iterable[torch.Tensor]],
ep_group: ProcessGroup,
is_profile: bool = False,
rank_mapping: Optional[dict[int, int]] = None,
rank_mapping: dict[int, int] | None = None,
) -> None:
"""
Rearranges the expert weights in place according to the new expert indices.