feat: Add DeepSeek Model for E2E (#850)

presets/workspace/models/deepseek/model.go

@@ -68,7 +68,7 @@ func (*llama8b) GetInferenceParameters() *model.PresetParam {
 				ModelRunParams:    deepseekLlama8bRunParams,
 			},
 			VLLM: model.VLLMParam{
-				BaseCommand:    baseCommandPresetDeepseekInference,
+				BaseCommand:    inference.DefaultVLLMCommand,
 				ModelName:      "/workspace/vllm/weights",
 				ModelRunParams: deepseekLlama8bRunParamsVLLM,
 			},

test/e2e/preset_test.go

@@ -27,15 +27,17 @@ import (
 )
 
 const (
-	PresetLlama2AChat            = "llama-2-7b-chat"
-	PresetLlama2BChat            = "llama-2-13b-chat"
-	PresetFalcon7BModel          = "falcon-7b"
-	PresetFalcon40BModel         = "falcon-40b"
-	PresetMistral7BInstructModel = "mistral-7b-instruct"
-	PresetQwen2_5Coder7BModel    = "qwen2.5-coder-7b-instruct"
-	PresetPhi2Model              = "phi-2"
-	PresetPhi3Mini128kModel      = "phi-3-mini-128k-instruct"
-	WorkspaceHashAnnotation      = "workspace.kaito.io/hash"
+	PresetLlama2AChat                   = "llama-2-7b-chat"
+	PresetLlama2BChat                   = "llama-2-13b-chat"
+	PresetFalcon7BModel                 = "falcon-7b"
+	PresetFalcon40BModel                = "falcon-40b"
+	PresetMistral7BInstructModel        = "mistral-7b-instruct"
+	PresetQwen2_5Coder7BModel           = "qwen2.5-coder-7b-instruct"
+	PresetPhi2Model                     = "phi-2"
+	PresetPhi3Mini128kModel             = "phi-3-mini-128k-instruct"
+	PresetDeepSeekR1DistillLlama8BModel = "deepseek-r1-distill-llama-8b"
+	PresetDeepSeekR1DistillQwen14BModel = "deepseek-r1-distill-qwen-14b"
+	WorkspaceHashAnnotation             = "workspace.kaito.io/hash"
 	// WorkspaceRevisionAnnotation represents the revision number of the workload managed by the workspace
 	WorkspaceRevisionAnnotation = "workspace.kaito.io/revision"
 )

test/e2e/preset_vllm_test.go

@@ -27,6 +27,26 @@ var _ = Describe("Workspace Preset on vllm runtime", func() {
 		}
 	})
 
+	It("should create a deepseek-distilled-llama-8b workspace with preset public mode successfully", func() {
+		numOfNode := 1
+		workspaceObj := createDeepSeekLlama8BWorkspaceWithPresetPublicModeAndVLLM(numOfNode)
+
+		defer cleanupResources(workspaceObj)
+		time.Sleep(30 * time.Second)
+
+		validateCreateNode(workspaceObj, numOfNode)
+		validateResourceStatus(workspaceObj)
+
+		time.Sleep(30 * time.Second)
+
+		validateAssociatedService(workspaceObj)
+		validateInferenceConfig(workspaceObj)
+
+		validateInferenceResource(workspaceObj, int32(numOfNode), false)
+
+		validateWorkspaceReadiness(workspaceObj)
+	})
+
 	It("should create a falcon workspace with preset public mode successfully", func() {
 		numOfNode := 1
 		workspaceObj := createFalconWorkspaceWithPresetPublicModeAndVLLM(numOfNode)
@@ -153,6 +173,20 @@ var _ = Describe("Workspace Preset on vllm runtime", func() {
 	})
 })
 
+func createDeepSeekLlama8BWorkspaceWithPresetPublicModeAndVLLM(numOfNode int) *kaitov1alpha1.Workspace {
+	workspaceObj := &kaitov1alpha1.Workspace{}
+	By("Creating a workspace CR with DeepSeek Distilled Llama 8B preset public mode and vLLM", func() {
+		uniqueID := fmt.Sprint("preset-deepseek-", rand.Intn(1000))
+		workspaceObj = utils.GenerateInferenceWorkspaceManifestWithVLLM(uniqueID, namespaceName, "", numOfNode, "Standard_NC24ads_A100_v4",
+			&metav1.LabelSelector{
+				MatchLabels: map[string]string{"kaito-workspace": "public-preset-e2e-test-deepseek-vllm"},
+			}, nil, PresetDeepSeekR1DistillLlama8BModel, kaitov1alpha1.ModelImageAccessModePublic, nil, nil, nil)
+
+		createAndValidateWorkspace(workspaceObj)
+	})
+	return workspaceObj
+}
+
 func createFalconWorkspaceWithPresetPublicModeAndVLLM(numOfNode int) *kaitov1alpha1.Workspace {
 	workspaceObj := &kaitov1alpha1.Workspace{}
 	By("Creating a workspace CR with Falcon 7B preset public mode and vLLM", func() {