Skip to content

Commit cbfee9c

Browse files
committed
wip: tunercheck action
Signed-off-by: Raghu Raja <[email protected]>
1 parent a97f82b commit cbfee9c

File tree

1 file changed

+56
-0
lines changed

1 file changed

+56
-0
lines changed

.github/workflows/tunercheck.yaml

Lines changed: 56 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,56 @@
1+
name: Build Checks
2+
on: [push, pull_request]
3+
permissions:
4+
contents: read
5+
pull-requests: read
6+
7+
jobs:
8+
tuner-decisions-check:
9+
runs-on: ubuntu-22.04
10+
steps:
11+
- name: Install Dependencies
12+
run: |
13+
sudo apt-key del 7fa2af80
14+
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
15+
sudo dpkg -i cuda-keyring_1.1-1_all.deb
16+
sudo apt-get update
17+
sudo apt-get install cuda-toolkit libhwloc-dev
18+
pip install uv
19+
20+
21+
- name: Fetch and Install EFA Installer
22+
run: |
23+
curl -O https://efa-installer.amazonaws.com/aws-efa-installer-latest.tar.gz
24+
tar -xf aws-efa-installer-*.tar.gz
25+
pushd aws-efa-installer/
26+
sudo ./efa_installer.sh -y --skip-kmod
27+
popd
28+
29+
- uses: actions/checkout@v4
30+
- name: Build Plugin
31+
run: |
32+
set -x
33+
34+
# actions/checkout@v4 would drop the plugin source in $PWD,
35+
# so go ahead and build it.
36+
./autogen.sh
37+
./configure --with-mpi=/opt/amazon/openmpi \
38+
--with-libfabric=/opt/amazon/efa \
39+
--with-cuda=/usr/local/cuda/ \
40+
--enable-platform-aws \
41+
--prefix=$PWD/install
42+
make -j 2
43+
make install
44+
45+
- name: Tuner Check
46+
run: |
47+
pushd contrib/python/
48+
49+
OFI_NCCL_FORCE_PRODUCT_NAME=p5en.48xlarge uv run show-tuner-decisions ../..//install/lib/libnccl-ofi-tuner.so \
50+
--min-ranks-per-node 1 --max-ranks-per-node 8 \
51+
--min-nnodes 2 --max-nnodes 2048
52+
53+
OFI_NCCL_FORCE_PRODUCT_NAME=p5.48xlarge uv run show-tuner-decisions ../..//install/lib/libnccl-ofi-tuner.so \
54+
--min-ranks-per-node 1 --max-ranks-per-node 8 \
55+
--min-nnodes 2 --max-nnodes 2048
56+
popd

0 commit comments

Comments
 (0)