fix: on GB200 use single-thread checkpoint save to avoid Cpu OOM (NVIDIA-NeMo#1703)

guyueh1 · parthmannan · commit 40511230cd92 · 2026-01-15T10:41:36.000-08:00
Signed-off-by: Guyue Huang &lt;guyueh@nvidia.com&gt;
Signed-off-by: Parth Mannan &lt;pmannan@nvidia.com&gt;
diff --git a/3rdparty/Megatron-LM-workspace/Megatron-LM b/3rdparty/Megatron-LM-workspace/Megatron-LM
@@ -1 +1 @@
-Subproject commit 25a62edf77b5130f888328ca8119d6a76117cf23
+Subproject commit b73ae5cdab9d409fcface2b2f3c375710abe6911
