xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-23 19:00:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix pre-commit

Browse Source 
Signed-off-by: ilmarkov <markovilya197@gmail.com>
This commit is contained in:
ilmarkov 2025-12-09 12:21:31 +00:00
parent 599648b6b7
commit b462872804
2 changed files with 7 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/distributed/eplb/policy/abstract.py
View File

@ -16,6 +16,7 @@ class AbstractEplbPolicy(ABC):
num_groups: int,
num_nodes: int,
num_ranks: int,
old_global_expert_indices: torch.Tensor | None = None,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
"""
Entry point for expert-parallelism load balancer.
@ -28,7 +29,9 @@ class AbstractEplbPolicy(ABC):
num_groups: number of expert groups
num_nodes: number of server nodes
num_ranks: number of ranks, must be a multiple of `num_nodes`
old_global_expert_indices: [layers, num_logical_experts], the old global
expert indices. Used to avoid unnecessary weight copying
for experts moving within one rank.
Returns:
physical_to_logical_map: [layers, num_replicas], the expert
index of each replica

4
vllm/distributed/eplb/policy/default.py
View File

@ -328,7 +328,9 @@ class DefaultEplbPolicy(AbstractEplbPolicy):
num_nodes: number of server nodes, where the intra-node network
(e.g, NVLink) is faster
num_ranks: number of ranks, must be a multiple of `num_nodes`
old_global_expert_indices: [layers, num_logical_experts], the old global
expert indices. Used to avoid unnecessary weight copying
for experts moving within one rank.
Returns:
phy2log: [layers, num_replicas], the expert
index of each replica