feat: Add Target Modules Env Variable and ETCD Cleanup

.github/actions/build-image-action/action.yml

@@ -81,8 +81,19 @@ runs:
         if ! kind get clusters | grep -q kind; then
           echo "Creating directory for etcd storage"
           sudo mkdir -p /mnt/storage/etcd
-          echo "Creating Kind cluster using kind-1es.yaml"
-          if ! kind create cluster --config .github/workflows/kind-cluster/kind-1es.yaml; then
+          echo "Creating Kind cluster"
+
+          if [[ "${{ inputs.runs_on }}" == "hostname:model-server" ]]; then
+            # Ensure clean etcd directory for mounting
+            sudo rm -rf /datadrive/etcd/*
+            config_file=".github/workflows/kind-cluster/kind.yaml"
+          else
+            # Ensure clean etcd directory for mounting
+            sudo rm -rf /mnt/storage/etcd/*
+            config_file=".github/workflows/kind-cluster/kind-1es.yaml"
+          fi
+
+          if ! kind create cluster --config $config_file; then
             echo "Failed to create the Kind cluster."
             exit 1
           fi

pkg/resources/manifests.go

@@ -189,7 +189,8 @@ func GenerateStatefulSetManifest(ctx context.Context, workspaceObj *kaitov1alpha
 func GenerateTuningJobManifest(ctx context.Context, wObj *kaitov1alpha1.Workspace, imageName string,
 	imagePullSecretRefs []corev1.LocalObjectReference, replicas int, commands []string, containerPorts []corev1.ContainerPort,
 	livenessProbe, readinessProbe *corev1.Probe, resourceRequirements corev1.ResourceRequirements, tolerations []corev1.Toleration,
-	initContainers []corev1.Container, sidecarContainers []corev1.Container, volumes []corev1.Volume, volumeMounts []corev1.VolumeMount) *batchv1.Job {
+	initContainers []corev1.Container, sidecarContainers []corev1.Container, volumes []corev1.Volume, volumeMounts []corev1.VolumeMount, 
+	envVars []corev1.EnvVar) *batchv1.Job {
 	labels := map[string]string{
 		kaitov1alpha1.LabelWorkspaceName: wObj.Name,
 	}
@@ -210,6 +211,7 @@ func GenerateTuningJobManifest(ctx context.Context, wObj *kaitov1alpha1.Workspac
 			ReadinessProbe: readinessProbe,
 			Ports:          containerPorts,
 			VolumeMounts:   volumeMounts,
+			Env:            envVars,
 		},
 	}, sidecarContainers...)
 

pkg/tuning/preset-tuning.go

@@ -315,8 +315,17 @@ func CreatePresetTuning(ctx context.Context, workspaceObj *kaitov1alpha1.Workspa
 		imagePullSecrets = append(imagePullSecrets, tuningImagePullSecrets...)
 	}
 
+	var envVars []corev1.EnvVar
+	presetName := strings.ToLower(string(workspaceObj.Tuning.Preset.Name))
+	// Append environment variable for default target modules if using Phi3 model
+	if strings.HasPrefix(presetName, "phi-3") {
+		envVars = append(envVars, corev1.EnvVar{
+			Name:  "DEFAULT_TARGET_MODULES",
+			Value: "k_proj,q_proj,v_proj,o_proj,gate_proj,down_proj,up_proj",
+		})
+	}
 	jobObj := resources.GenerateTuningJobManifest(ctx, workspaceObj, tuningImage, imagePullSecrets, *workspaceObj.Resource.Count, commands,
-		containerPorts, nil, nil, resourceReq, tolerations, initContainers, sidecarContainers, volumes, volumeMounts)
+		containerPorts, nil, nil, resourceReq, tolerations, initContainers, sidecarContainers, volumes, volumeMounts, envVars)
 
 	err = resources.CreateResource(ctx, jobObj, kubeClient)
 	if client.IgnoreAlreadyExists(err) != nil {

presets/models/supported_models.yaml

@@ -105,4 +105,36 @@ models:
     runtime: tfs
     tag: 0.0.1
     # Tag history:
-    # 0.0.1 - Initial Release
+    # 0.0.1 - Initial Release
+
+  - name: phi-3-small-8k-instruct
+    type: text-generation
+    version: https://huggingface.co/microsoft/Phi-3-small-8k-instruct/commit/69caae1f2acea34b26f535fecb1f2abb9a304695
+    runtime: tfs
+    tag: 0.0.1
+    # Tag history:
+    # 0.0.1 - Initial Release
+
+  - name: phi-3-small-128k-instruct
+    type: text-generation
+    version: https://huggingface.co/microsoft/Phi-3-small-128k-instruct/commit/f80aaa30bfc64c2b8ab214b541d9050e97163bc4
+    runtime: tfs
+    tag: 0.0.1
+    # Tag history:
+    # 0.0.1 - Initial Release
+
+  - name: phi-3-medium-4k-instruct
+    type: text-generation
+    version: https://huggingface.co/microsoft/Phi-3-medium-4k-instruct/commit/d194e4e74ffad5a5e193e26af25bcfc80c7f1ffc
+    runtime: tfs
+    tag: 0.0.1
+    # Tag history:
+    # 0.0.1 - Initial Release
+
+  - name: phi-3-medium-128k-instruct
+    type: text-generation
+    version: https://huggingface.co/microsoft/Phi-3-medium-128k-instruct/commit/cae1d42b5577398fd1be9f0746052562ae552886
+    runtime: tfs
+    tag: 0.0.1
+    # Tag history:
+    # 0.0.1 - Initial Release

presets/tuning/text-generation/cli.py

@@ -11,6 +11,9 @@
 from transformers import (BitsAndBytesConfig, DataCollatorForLanguageModeling,
                           PreTrainedTokenizer)
 
+target_modules_env = os.environ.get('DEFAULT_TARGET_MODULES', None) 
+DEFAULT_TARGET_MODULES = [module.strip() for module in target_modules_env.split(",")] if target_modules_env else None
+
 # Consider Future Support for other trainers
 # class TrainerTypes(Enum):
     # TRAINER = "Trainer"
@@ -35,7 +38,7 @@ class ExtLoraConfig(LoraConfig):
     Lora Config
     """
     init_lora_weights: bool = field(default=True, metadata={"help": "Enable initialization of LoRA weights"})
-    target_modules: Optional[List[str]] = field(default=None, metadata={"help": ("List of module names to replace with LoRA.")})
+    target_modules: Optional[List[str]] = field(default_factory=lambda: DEFAULT_TARGET_MODULES if DEFAULT_TARGET_MODULES else None, metadata={"help": "List of module names to replace with LoRA."})
     layers_to_transform: Optional[List[int]] = field(default=None, metadata={"help": "Layer indices to apply LoRA"})
     layers_pattern: Optional[List[str]] = field(default=None, metadata={"help": "Pattern to match layers for LoRA"})
     loftq_config: Dict[str, any] = field(default_factory=dict, metadata={"help": "LoftQ configuration for quantization"})