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 14, 04:15 UTC
[Bugfix] Fix modular kernel tests (#27707)
Signed-off-by: Bill Nell <bnell@redhat.com>
bnellnm committed
Oct 29, 2025 at 04:14 UTC
1891cf605ac015016cca38acc9e1950f70586f2e
2 files changed
+3
.buildkite/test-pipeline.yaml
+2
@@ -498,6 +498,8 @@ steps:
498
- tests/kernels/moe
499
- vllm/model_executor/layers/fused_moe/
500
- vllm/distributed/device_communicators/
501
+ - vllm/envs.py
502
+ - vllm/config
503
commands:
504
- pytest -v -s kernels/moe --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
505
parallelism: 2
tests/kernels/moe/modular_kernel_tools/common.py
+1
@@ -138,6 +138,7 @@ class Config:
138
}
139
140
backend = self.all2all_backend()
141
+ vllm_config.parallel_config.all2all_backend = backend
142
if backend is not None:
143
env_dict.update({"VLLM_ALL2ALL_BACKEND": backend})
144