Merge remote-tracking branch 'upstream/main'

Author: dchourasia

go.mod

@@ -7,7 +7,7 @@ require (
 	github.com/matoous/go-nanoid/v2 v2.1.0
 	github.com/onsi/gomega v1.32.0
 	github.com/project-codeflare/appwrapper v0.8.0
-	github.com/project-codeflare/codeflare-common v0.0.0-20250128135036-f501cd31fe8b
+	github.com/project-codeflare/codeflare-common v0.0.0-20250317102908-1c124db97844
 	github.com/prometheus/client_golang v1.20.4
 	github.com/prometheus/common v0.57.0
 	github.com/ray-project/kuberay/ray-operator v1.1.1

go.sum

@@ -366,8 +366,8 @@ github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZb
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/project-codeflare/appwrapper v0.8.0 h1:vWHNtXUtHutN2EzYb6rryLdESnb8iDXsCokXOuNYXvg=
 github.com/project-codeflare/appwrapper v0.8.0/go.mod h1:FMQ2lI3fz6LakUVXgN1FTdpsc3BBkNIZZgtMmM9J5UM=
-github.com/project-codeflare/codeflare-common v0.0.0-20250128135036-f501cd31fe8b h1:MOmv/aLx/kcHd7PBErx8XNSTW180s8Slf/uVM0uV4rw=
-github.com/project-codeflare/codeflare-common v0.0.0-20250128135036-f501cd31fe8b/go.mod h1:DPSv5khRiRDFUD43SF8da+MrVQTWmxNhuKJmwSLOyO0=
+github.com/project-codeflare/codeflare-common v0.0.0-20250317102908-1c124db97844 h1:hEjZ2pV4Fp81wytijJZ7uHWovKIqirVBA/t1F5hIrbA=
+github.com/project-codeflare/codeflare-common v0.0.0-20250317102908-1c124db97844/go.mod h1:DPSv5khRiRDFUD43SF8da+MrVQTWmxNhuKJmwSLOyO0=
 github.com/prometheus/client_golang v0.9.1/go.mod h1:7SWBe2y4D6OKWSNQJUaRYU/AaXPKyh/dDVn+NZz0KFw=
 github.com/prometheus/client_golang v1.0.0/go.mod h1:db9x61etRT2tGnBNRi70OPL5FsnadC4Ky3P0J6CfImo=
 github.com/prometheus/client_golang v1.7.1/go.mod h1:PY5Wy2awLA44sXw4AOSfFBetzPP4j5+D6mVACh+pe2M=

tests/kfto/kfto_mnist_training_test.go

@@ -63,7 +63,7 @@ func runKFTOPyTorchMnistJob(t *testing.T, accelerator Accelerator, image string,
 	download_mnist_dataset := readFile(test, "resources/download_mnist_datasets.py")
 	requirementsFileName := readFile(test, requirementsFile)
 
-	if accelerator.isGpu() {
+	if accelerator.IsGpu() {
 		mnist = bytes.Replace(mnist, []byte("accelerator=\"has to be specified\""), []byte("accelerator=\"gpu\""), 1)
 	} else {
 		mnist = bytes.Replace(mnist, []byte("accelerator=\"has to be specified\""), []byte("accelerator=\"cpu\""), 1)
@@ -90,7 +90,7 @@ func runKFTOPyTorchMnistJob(t *testing.T, accelerator Accelerator, image string,
 
 func createKFTOPyTorchMnistJob(test Test, namespace string, config corev1.ConfigMap, accelerator Accelerator, workerReplicas int, numProcPerNode int, baseImage string) *kftov1.PyTorchJob {
 	var backend string
-	if accelerator.isGpu() {
+	if accelerator.IsGpu() {
 		backend = "nccl"
 	} else {
 		backend = "gloo"
@@ -306,7 +306,7 @@ func createKFTOPyTorchMnistJob(test Test, namespace string, config corev1.Config
 		},
 	}
 
-	if accelerator.isGpu() {
+	if accelerator.IsGpu() {
 		// Update resource lists for GPU (NVIDIA/ROCm) usecase
 		tuningJob.Spec.PyTorchReplicaSpecs[kftov1.PyTorchJobReplicaTypeMaster].Template.Spec.Containers[0].Resources.Requests[corev1.ResourceName(accelerator.ResourceLabel)] = resource.MustParse(fmt.Sprint(numProcPerNode))
 		tuningJob.Spec.PyTorchReplicaSpecs[kftov1.PyTorchJobReplicaTypeMaster].Template.Spec.Containers[0].Resources.Limits[corev1.ResourceName(accelerator.ResourceLabel)] = resource.MustParse(fmt.Sprint(numProcPerNode))

tests/kfto/support.go

@@ -30,22 +30,6 @@ import (
 	corev1 "k8s.io/api/core/v1"
 )
 
-type Accelerator struct {
-	ResourceLabel                 string
-	PrometheusGpuUtilizationLabel string
-}
-
-var (
-	NVIDIA = Accelerator{ResourceLabel: "nvidia.com/gpu", PrometheusGpuUtilizationLabel: "DCGM_FI_DEV_GPU_UTIL"}
-	AMD    = Accelerator{ResourceLabel: "amd.com/gpu"}
-	CPU    = Accelerator{}
-)
-
-// Method to check if the accelerator is a GPU
-func (a Accelerator) isGpu() bool {
-	return a != CPU
-}
-
 //go:embed resources/*
 var files embed.FS