Merge remote-tracking branch 'upstream/main'

Author: m-rafeeq

tests/common/support/environment.go

@@ -59,6 +59,7 @@ const (
 	storageBucketMnistDir        = "AWS_STORAGE_BUCKET_MNIST_DIR"
 	storageBucketFashionMnistDir = "AWS_STORAGE_BUCKET_FASHION_MNIST_DIR"
 	storageBucketOsftDir         = "AWS_STORAGE_BUCKET_OSFT_DIR"
+	storageBucketSftDir          = "AWS_STORAGE_BUCKET_SFT_DIR"
 
 	// Name of existing namespace to be used for test
 	testNamespaceNameEnvVar = "TEST_NAMESPACE_NAME"
@@ -191,6 +192,11 @@ func GetStorageBucketOsftDir() (string, bool) {
 	return storage_bucket_osft_dir, exists
 }
 
+func GetStorageBucketSftDir() (string, bool) {
+	storage_bucket_sft_dir, exists := os.LookupEnv(storageBucketSftDir)
+	return storage_bucket_sft_dir, exists
+}
+
 func GetPipIndexURL() string {
 	return lookupEnvOrDefault(pipIndexURL, "https://pypi.python.org/simple")
 }

tests/trainer/kubeflow_sdk_test.go

@@ -34,3 +34,9 @@ func TestOsftTrainingHubMultiNodeMultiGPU(t *testing.T) {
 	Tags(t, KftoCuda, MultiNodeMultiGpu(2, support.NVIDIA, 1)) // TODO: may need to be updated once https://issues.redhat.com/browse/RHOAIENG-30719 and https://issues.redhat.com/browse/RHOAIENG-24552 are resolved
 	sdktests.RunOsftTrainingHubMultiGpuDistributedTraining(t)
 }
+
+// TestSftTrainingHubMultiNodeMultiGPU tests SFT training using TrainingHubTrainer
+func TestSftTrainingHubMultiNodeMultiGPU(t *testing.T) {
+	Tags(t, KftoCuda, MultiNodeMultiGpu(2, support.NVIDIA, 1))
+	sdktests.RunSftTrainingHubMultiGpuDistributedTraining(t)
+}