Change nccl version to 2.20.3

NCCL 2.19.4 has performance regression.

4.validation_and_observability/0.nccl-tests/0.nccl-tests.Dockerfile

@@ -2,10 +2,10 @@
 # SPDX-License-Identifier: MIT-0
 FROM nvidia/cuda:12.2.0-devel-ubuntu22.04
 
-ARG EFA_INSTALLER_VERSION=1.30.0
+ARG EFA_INSTALLER_VERSION=1.31.0
 ARG AWS_OFI_NCCL_VERSION=v1.8.1-aws
 ARG NCCL_TESTS_VERSION=2.13.9
-ARG NCCL_VERSION=2.19.4
+ARG NCCL_VERSION=2.20.3
 
 RUN apt-get update -y
 RUN apt-get remove -y --allow-change-held-packages \
@@ -102,4 +102,4 @@ RUN git clone -b v${NCCL_TESTS_VERSION} https://github.com/NVIDIA/nccl-tests.git
     NVCC_GENCODE="-gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -gencode=arch=compute_90,code=sm_90"
 
 RUN rm -rf /var/lib/apt/lists/*
-ENV LD_PRELOAD /opt/nccl/build/lib/libnccl.so
+ENV LD_PRELOAD /opt/nccl/build/lib/libnccl.so