Skip to content
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

Jump to bottom

Restrict GPU access from worker to deviceIds #3221

Closed
wants to merge 2 commits into from
Closed

Restrict GPU access from worker to deviceIds #3221

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2282882
Restrict GPU access from worker to deviceIds
namannandan Jul 2, 2024
eba5d7b
Set device Ids to gpu id when parallel level is not configured
namannandan Jul 2, 2024
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
20 changes: 7 additions & 13 deletions frontend/server/src/main/java/org/pytorch/serve/wlm/WorkerLifeCycle.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,16 +130,13 @@ private void startWorkerPython(int port, String deviceIds)
modelPath.getAbsolutePath(),
model.getModelArchive().getManifest().getModel().getHandler())));

if (model.getParallelLevel() > 0) {
if (model.getParallelType() != ParallelType.CUSTOM) {
attachRunner(argl, envp, port, deviceIds);
} else {
if (deviceIds != null) {
envp.add("CUDA_VISIBLE_DEVICES=" + deviceIds);
}
argl.add(EnvironmentUtils.getPythonRunTime(model));
}
} else if (model.getParallelLevel() == 0) {
if (deviceIds != null) {
envp.add("CUDA_VISIBLE_DEVICES=" + deviceIds);
}

if (model.getParallelLevel() > 0 && model.getParallelType() != ParallelType.CUSTOM) {
attachRunner(argl, envp, port, deviceIds);
} else {
argl.add(EnvironmentUtils.getPythonRunTime(model));
}

Expand Down Expand Up @@ -291,9 +288,6 @@ private void startWorkerCPP(int port, String runtimeType, String deviceIds)
private void attachRunner(
ArrayList<String> argl, List<String> envp, int port, String deviceIds) {
envp.add("LOGLEVEL=INFO");
if (deviceIds != null) {
envp.add("CUDA_VISIBLE_DEVICES=" + deviceIds);
}
ModelConfig.TorchRun torchRun = model.getModelArchive().getModelConfig().getTorchRun();
envp.add(String.format("OMP_NUM_THREADS=%d", torchRun.getOmpNumberThreads()));
argl.add("torchrun");
Expand Down
10 changes: 7 additions & 3 deletions frontend/server/src/main/java/org/pytorch/serve/wlm/WorkerThread.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -544,16 +544,20 @@ public void retry() {

protected String getDeviceIds() {
List<Integer> deviceIds;
if (gpuId == -1 || model.getParallelLevel() == 0) {
if (gpuId == -1) {
return null;
} else if (model.isHasCfgDeviceIds()) {
return model.getDeviceIds().subList(gpuId, gpuId + model.getParallelLevel()).stream()
.map(String::valueOf)
.collect(Collectors.joining(","));
} else {
deviceIds = new ArrayList<>(model.getParallelLevel());
for (int i = gpuId; i < gpuId + model.getParallelLevel(); i++) {
deviceIds.add(i);
if (model.getParallelLevel() > 0) {
for (int i = gpuId; i < gpuId + model.getParallelLevel(); i++) {
deviceIds.add(i);
}
} else {
deviceIds.add(gpuId);
}
return deviceIds.stream().map(String::valueOf).collect(Collectors.joining(","));
}
Expand Down
Loading