From 90d0a54c4dae03f3e8c111babd359907e719355f Mon Sep 17 00:00:00 2001 From: Hongxia Yang <62075498+hongxiayang@users.noreply.github.com> Date: Thu, 1 May 2025 02:27:06 -0400 Subject: [PATCH] [ROCm] Effort to reduce the number of environment variables in command line (#17229) Signed-off-by: Hongxia Yang --- docker/Dockerfile.rocm | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/docker/Dockerfile.rocm b/docker/Dockerfile.rocm index 0a512f993a1e..e60cf5e69a4c 100644 --- a/docker/Dockerfile.rocm +++ b/docker/Dockerfile.rocm @@ -114,6 +114,15 @@ COPY --from=export_vllm /examples ${COMMON_WORKDIR}/vllm/examples ENV RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1 ENV TOKENIZERS_PARALLELISM=false +# ENV that can improve safe tensor loading, and end-to-end time +ENV SAFETENSORS_FAST_GPU=1 + +# User-friendly environment setting for multi-processing to avoid below RuntimeError. +# RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, +# you must use the 'spawn' start method +# See https://pytorch.org/docs/stable/notes/multiprocessing.html#cuda-in-multiprocessing +ENV VLLM_WORKER_MULTIPROC_METHOD=spawn + # Performance environment variable. ENV HIP_FORCE_DEV_KERNARG=1