Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
59bd5f6a
Unverified
Commit
59bd5f6a
authored
Dec 16, 2025
by
Wentao Ye
Committed by
GitHub
Dec 16, 2025
Browse files
[Feat] Enable eplb with default all2all backend (#30559)
Signed-off-by:
yewentao256
<
zhyanwentao@126.com
>
parent
00a8d762
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
4 additions
and
4 deletions
+4
-4
vllm/model_executor/layers/fused_moe/shared_fused_moe.py
vllm/model_executor/layers/fused_moe/shared_fused_moe.py
+4
-4
No files found.
vllm/model_executor/layers/fused_moe/shared_fused_moe.py
View file @
59bd5f6a
...
@@ -29,14 +29,14 @@ class SharedFusedMoE(FusedMoE):
...
@@ -29,14 +29,14 @@ class SharedFusedMoE(FusedMoE):
self
.
_shared_experts
=
shared_experts
self
.
_shared_experts
=
shared_experts
# Disable shared expert overlap if:
# Disable shared expert overlap if:
# - we are using eplb, because of correctness issues
# - we are using eplb
with non-default backend
, because of correctness issues
# - we are using flashinfer with DP, since there nothin
g
to gain
# - we are using flashinfer with DP, since there nothin
t
to gain
# - we are using marlin kernels
# - we are using marlin kernels
backend
=
self
.
moe_parallel_config
.
all2all_backend
self
.
use_overlapped
=
(
self
.
use_overlapped
=
(
use_overlapped
use_overlapped
and
not
(
and
not
(
# TODO(wentao): find the root cause and remove this condition
(
self
.
enable_eplb
and
backend
!=
"allgather_reducescatter"
)
self
.
enable_eplb
or
(
self
.
moe_config
.
use_flashinfer_cutlass_kernels
and
self
.
dp_size
>
1
)
or
(
self
.
moe_config
.
use_flashinfer_cutlass_kernels
and
self
.
dp_size
>
1
)
)
)
and
self
.
_shared_experts
is
not
None
and
self
.
_shared_experts
is
not
None
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment