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, 05:30 UTC
[CI/Build] add EP dependencies to docker (#21976)
Co-authored-by: Simon Mo <simon.mo@hey.com>
Zhewen Li committed
Aug 22, 2025 at 13:34 UTC
0483fabc746c79f6969b600665568255260d0b94
2 files changed
+15
.buildkite/test-pipeline.yaml
+7
@@ -843,3 +843,10 @@ steps:
843
commands:
844
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
845
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
846
+
847
+- label: Qwen MoE EP Test # optional
848
+ gpu: h200
849
+ optional: true
850
+ num_gpus: 2
851
+ commands:
852
+ - CUDA_VISIBLE_DEVICES=1,2 VLLM_ALL2ALL_BACKEND=deepep_high_throughput VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 /vllm-workspace/examples/offline_inference/data_parallel.py --model Qwen/Qwen1.5-MoE-A2.7B --tp-size=1 --dp-size=2 --max-model-len 2048
docker/Dockerfile
+8
@@ -458,6 +458,14 @@ RUN --mount=type=cache,target=/root/.cache/uv bash - <<'BASH'
458
fi
459
BASH
460
461
+# Install EP kernels(pplx-kernels and DeepEP), NixL
462
+COPY tools/ep_kernels/install_python_libraries.sh install_python_libraries.sh
463
+COPY tools/install_nixl.sh install_nixl.sh
464
+ENV CUDA_HOME=/usr/local/cuda
465
+RUN export TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-9.0a+PTX}" \
466
+ && bash install_python_libraries.sh \
467
+ && bash install_nixl.sh --force
468
+
469
#################### vLLM installation IMAGE ####################
470
471
#################### TEST IMAGE ####################