From ff45ddff1bc5e5a88443f5164d13b86708072214 Mon Sep 17 00:00:00 2001
From: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Date: Tue, 5 Aug 2025 18:33:17 +0100
Subject: [PATCH] fix(runtimes): Set numProcPerNode: 1 in DeepSpeed Runtime

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
---
 manifests/base/runtimes/deepspeed_distributed.yaml | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/manifests/base/runtimes/deepspeed_distributed.yaml b/manifests/base/runtimes/deepspeed_distributed.yaml
index 869e97dbbb..2ebc62b226 100644
--- a/manifests/base/runtimes/deepspeed_distributed.yaml
+++ b/manifests/base/runtimes/deepspeed_distributed.yaml
@@ -8,9 +8,7 @@ spec:
   mlPolicy:
     numNodes: 1
     mpi:
-      # TODO (andreyvelich): Change num proc to 1 and remove container resources after we
-      # allow to override it via TrainJob APIs.
-      numProcPerNode: 4
+      numProcPerNode: 1
       mpiImplementation: OpenMPI
       sshAuthMountPath: /home/mpiuser/.ssh
       runLauncherAsNode: true