A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 6, 21:45 UTC
[Bugfix] Fix DP/EP Shared Expert With Monolithic Kernels (#36061)
Signed-off-by: Robert Shaw <robshaw@redhat.com> Co-authored-by: Robert Shaw <robshaw@redhat.com>
Robert Shaw committed
Mar 11, 2026 at 12:07 UTC
b7e5a588d89003223bebc9b163413529f3db4cae
2 files changed
+2
-2
vllm/model_executor/layers/fused_moe/oracle/fp8.py
+1
-1
@@ -567,7 +567,7 @@ def make_fp8_moe_kernel(
567
experts,
568
shared_experts=(
569
shared_experts
570
- if moe_config.moe_parallel_config.use_all2all_kernels
570
+ if moe_config.moe_parallel_config.use_deepep_ll_kernels
571
else None
572
),
573
moe_parallel_config=moe_config.moe_parallel_config,
vllm/model_executor/layers/fused_moe/oracle/nvfp4.py
+1
-1
@@ -433,7 +433,7 @@ def make_nvfp4_moe_kernel(
433
experts,
434
shared_experts=(
435
shared_experts
436
- if moe_config.moe_parallel_config.use_all2all_kernels
436
+ if moe_config.moe_parallel_config.use_deepep_ll_kernels
437
else None
438
),
439
moe_parallel_config=moe_config.moe_parallel_config,