From bd172e0569331cf1b7c3da4f31e98e58ce09d03b Mon Sep 17 00:00:00 2001 From: Sean Smith Date: Fri, 5 Apr 2024 12:09:31 +0900 Subject: [PATCH] Activate conda environment The conda needs to be activated otherwise the job will encounter an issue: ``` command torchrun not found ``` --- 3.test_cases/13.SM-dataparallel-deepspeed/exec_torchrun.sh | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/3.test_cases/13.SM-dataparallel-deepspeed/exec_torchrun.sh b/3.test_cases/13.SM-dataparallel-deepspeed/exec_torchrun.sh index 7454bae4..d04e7930 100755 --- a/3.test_cases/13.SM-dataparallel-deepspeed/exec_torchrun.sh +++ b/3.test_cases/13.SM-dataparallel-deepspeed/exec_torchrun.sh @@ -3,6 +3,10 @@ # Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved. # SPDX-License-Identifier: MIT-0 +# Activate conda environment +source ./conda/bin/activate +conda activate smdataparallel + export OMP_NUM_THREADS=1 export GPUS_PER_NODE=8 # p4d/p4de instances have 8 GPUs per node MASTER_NODE=$(scontrol show hostname | head -n 1)