Skip to content

Add multi-host gpu tests to PyTorch/XLA #1011

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
vanbasten23 wants to merge 7 commits into
base: master
Choose a base branch
from
Draft

Add multi-host gpu tests to PyTorch/XLA #1011

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
b9c588b
need to add initContainer next
vanbasten23 Nov 8, 2023
ab974d2
added init container
vanbasten23 Nov 8, 2023
c41955a
print version
vanbasten23 Nov 8, 2023
4283cfd
added the containerPort
vanbasten23 Nov 8, 2023
243d7fc
fix another error
vanbasten23 Nov 8, 2023
10ad060
did some experiment
vanbasten23 Nov 21, 2023
a940186
Try kubectl patch
vanbasten23 Nov 22, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion tests/pytorch/nightly/common.libsonnet
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,7 +120,7 @@ local volumes = import 'templates/volumes.libsonnet';
},
GpuMixin:: {
local config = self,
imageTag+: '_cuda_11.8',
imageTag: 'nightly_3.10_cuda_12.1',

podTemplate+:: {
spec+: {
Expand Down
77 changes: 75 additions & 2 deletions tests/pytorch/nightly/resnet50-mp.libsonnet
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,12 +145,84 @@ local tpus = import 'templates/tpus.libsonnet';
memory: '40Gi',

// Disable XLA metrics report on GPU
command+: [
'--nometrics_debug',
command: [
'bash',
'-c',
|||
export PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
export LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64

nvidia-smi
nvcc -V

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121
pip install --user https://storage.googleapis.com/pytorch-xla-releases/wheels/cuda/12.1/torch_xla-2.1.0-cp310-cp310-manylinux_2_28_x86_64.whl

git clone --depth=1 https://github.com/pytorch/pytorch.git
cd pytorch
git clone https://github.com/pytorch/xla.git

while true
do
ip=$(getent hosts ptxla-hello-world-0.headless-svc-$(JOB_NAME) | awk {'print $1'})
if [ $? -eq 0 ] && [ \"${ip}\" != \"\" ]
then
break
else
sleep 10
fi
done
echo $ip

PJRT_DEVICE=CUDA torchrun --nnodes=4 --node_rank=$JOB_COMPLETION_INDEX --nproc_per_node=4 --rdzv_endpoint=$ip:12355 xla/test/test_train_mp_imagenet.py --fake_data --pjrt_distributed --batch_size=128 --num_epochs=1"
|||,
],
flags+: {
modelDir: null,
},

jobTemplate+:: {
spec+: {
completionMode: 'Indexed',
completions: 4,
parallelism: 4,
},
},

podTemplate+:: {
spec+: {
initContainerMap+:: {
'tpu-version': {
command: [
"echo JOB_NAME=$(JOB_NAME)",
"echo POD_NAME=$(POD_NAME)",
"kubectl patch job $(JOB_NAME) -p \'{\"spec\":{\"subdomain\": \"headless-svc-$(JOB_NAME)\"}}\'",
"kubectl expose headless-svc-$(JOB_NAME) --type='None' --selector='job-name: $(JOB_NAME)'",
],
"image": "google/cloud-sdk",
},
},
containerMap+:: {
train+: {
ports: [
{
containerPort: 1234,
},
],
},
},
// subdomain: 'headless-svc-$(JOB_NAME)', doesn't work.
// subdomain: "headless-svc-metadata.labels['job-name']", doesn't work.
tolerations: [
{
key: "nvidia.com/gpu",
operator: "Exists",
effect: "NoSchedule",
},
],

},
},
},
local v100x4 = self.v100x4,
v100x4:: gpu {
Expand Down Expand Up @@ -194,6 +266,7 @@ local tpus = import 'templates/tpus.libsonnet';
},

configs: [
resnet50 + functional + v100x4 + timeouts.Hours(2),
// PJRT
resnet50 + fake_data + v2_8 + timeouts.Hours(3) + pjrt,
resnet50 + fake_data + v3_8 + timeouts.Hours(2) + pjrt,
Expand Down