will/bump_aws_ofi_nccl (#3253)

* Bump aws-ofi-nccl * update generate_build_matrix.py
mosaicml · May 8, 2024 · 9f7db04 · 9f7db04
1 parent 6898866
commit 9f7db04
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/docker/build_matrix.yaml b/docker/build_matrix.yaml
@@ -26,7 +26,7 @@
   - mosaicml/pytorch:2.3.0_cu121-python3.11-ubuntu20.04
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.18.0
-- AWS_OFI_NCCL_VERSION: v1.7.4-aws
+- AWS_OFI_NCCL_VERSION: v1.9.1-aws
   BASE_IMAGE: nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
   CUDA_VERSION: 12.1.1
   IMAGE_NAME: torch-2-3-0-cu121-aws
@@ -94,7 +94,7 @@
   - mosaicml/pytorch:2.2.1_cu121-python3.11-ubuntu20.04
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.17.1
-- AWS_OFI_NCCL_VERSION: v1.7.4-aws
+- AWS_OFI_NCCL_VERSION: v1.9.1-aws
   BASE_IMAGE: nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
   CUDA_VERSION: 12.1.1
   IMAGE_NAME: torch-2-2-1-cu121-aws
@@ -162,7 +162,7 @@
   - mosaicml/pytorch:2.1.2_cu121-python3.10-ubuntu20.04
   TARGET: pytorch_stage
   TORCHVISION_VERSION: 0.16.2
-- AWS_OFI_NCCL_VERSION: v1.7.4-aws
+- AWS_OFI_NCCL_VERSION: v1.9.1-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
@@ -224,7 +224,7 @@ def _main():
         if interconnect != 'EFA':
             entry['AWS_OFI_NCCL_VERSION'] = ''
         else:
-            entry['AWS_OFI_NCCL_VERSION'] = 'v1.7.4-aws'
+            entry['AWS_OFI_NCCL_VERSION'] = 'v1.9.1-aws'
 
         pytorch_entries.append(entry)