From 4fe3ea7e14426ec99bf02dfda2709e5dbcff0580 Mon Sep 17 00:00:00 2001
From: "will.gleich" <22464726+willgleich@users.noreply.github.com>
Date: Wed, 21 Aug 2024 16:42:00 -0600
Subject: [PATCH] Bump aws_ofi_nccl to 1.11.0

---
 docker/build_matrix.yaml        | 6 +++---
 docker/generate_build_matrix.py | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/docker/build_matrix.yaml b/docker/build_matrix.yaml
index ee74d12309..66a8a192e6 100644
--- a/docker/build_matrix.yaml
+++ b/docker/build_matrix.yaml
@@ -27,7 +27,7 @@
   - mosaicml/pytorch:latest
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.18.1
-- AWS_OFI_NCCL_VERSION: v1.9.1-aws
+- AWS_OFI_NCCL_VERSION: v1.11.0-aws
   BASE_IMAGE: nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
   CUDA_VERSION: 12.1.1
   IMAGE_NAME: torch-2-3-1-cu121-aws
@@ -97,7 +97,7 @@
   - mosaicml/pytorch:2.2.2_cu121-python3.11-ubuntu20.04
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.17.2
-- AWS_OFI_NCCL_VERSION: v1.9.1-aws
+- AWS_OFI_NCCL_VERSION: v1.11.0-aws
   BASE_IMAGE: nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
   CUDA_VERSION: 12.1.1
   IMAGE_NAME: torch-2-2-2-cu121-aws
@@ -165,7 +165,7 @@
   - mosaicml/pytorch:2.1.2_cu121-python3.10-ubuntu20.04
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.16.2
-- AWS_OFI_NCCL_VERSION: v1.9.1-aws
+- AWS_OFI_NCCL_VERSION: v1.11.0-aws
   BASE_IMAGE: nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
   CUDA_VERSION: 12.1.1
   IMAGE_NAME: torch-2-1-2-cu121-aws
diff --git a/docker/generate_build_matrix.py b/docker/generate_build_matrix.py
index 74d9c7fed4..85850334be 100644
--- a/docker/generate_build_matrix.py
+++ b/docker/generate_build_matrix.py
@@ -224,7 +224,7 @@ def _main():
         if interconnect != 'EFA':
             entry['AWS_OFI_NCCL_VERSION'] = ''
         else:
-            entry['AWS_OFI_NCCL_VERSION'] = 'v1.9.1-aws'
+            entry['AWS_OFI_NCCL_VERSION'] = 'v1.11.0-aws'
 
         pytorch_entries.append(entry)