[workflow] disable p2p via shared memory on non-nvlink machine (#1086)

1bd8a72f · Frank Lee · GitHub · 65ee6dcc · 1bd8a72f
Unverified Commit 1bd8a72f authored Jun 09, 2022 by Frank Lee Committed by GitHub Jun 09, 2022
Hide whitespace changes
Inline Side-by-side

Showing with 2 additions and 1 deletion

.github/workflows/build.yml .github/workflows/build.yml +2 -1

No files found.
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -15,7 +15,7 @@ jobs:
    runs-on: [self-hosted, gpu]
    container:
      image: hpcaitech/pytorch-cuda:1.10.1-11.3.0
-      options: --shm-size=2gb --gpus all --rm -v /data/scratch/cifar-10:/data/scratch/cifar-10
+      options: --gpus all --rm -v /data/scratch/cifar-10:/data/scratch/cifar-10
    timeout-minutes: 40
    steps:
      - uses: actions/checkout@v2
@@ -34,3 +34,4 @@ jobs:
          PYTHONPATH=$PWD pytest tests
        env:
          DATA: /data/scratch/cifar-10
+          NCCL_SHM_DISABLE: 1