Skip to content

Commit 40d3e55

Browse files
huydhnpytorchmergebot
authored andcommitted
Temporary fix to skip NVIDIA driver installation from RHEL repo (pytorch#85569)
This is a temporary fix until torchrec and FBGEMM are updated to use PyTorch NVIDIA installation script instead of using the latest driver from RHEL repo. It might take a day or so to finish updating the 2 repos, so I want to have this in place to avoid any issue with NVIDIA driver till then. The driver from RHEL repo `515.65.01` is even newer than what we are using in PyTorch CI `515.57`. So everything should just work with both of them Pull Request resolved: pytorch#85569 Approved by: https://github.com/clee2000
1 parent 4befe45 commit 40d3e55

File tree

1 file changed

+6
-1
lines changed

1 file changed

+6
-1
lines changed

.github/scripts/install_nvidia_utils_linux.sh

+6-1
Original file line numberDiff line numberDiff line change
@@ -30,7 +30,12 @@ install_nvidia_driver_amzn2() {
3030
INSTALLED_DRIVER_VERSION=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader)
3131

3232
if [ "$INSTALLED_DRIVER_VERSION" != "$DRIVER_VERSION" ]; then
33-
echo "NVIDIA driver ($INSTALLED_DRIVER_VERSION) has been installed, but we expect to have $DRIVER_VERSION instead. Continuing with NVIDIA driver installation"
33+
# TODO
34+
# Remove this after torchrec and FBGEMM have both been updated to use
35+
# PyTorch NVIDIA installation script instead of using the latest driver
36+
# from RHEL repo
37+
HAS_NVIDIA_DRIVER=1
38+
echo "NVIDIA driver ($INSTALLED_DRIVER_VERSION) has been installed, but we expect to have $DRIVER_VERSION instead. Skipping NVIDIA driver installation for now until torchrec and FBGEMM are updated to use PyTorch NVIDIA installation script instead of RHEL repo"
3439
else
3540
HAS_NVIDIA_DRIVER=1
3641
echo "NVIDIA driver ($INSTALLED_DRIVER_VERSION) has already been installed. Skipping NVIDIA driver installation"

0 commit comments

Comments
 (0)