add temp registry

Signed-off-by: Bangqi Zhu <bangqizhu@microsoft.com>

.github/workflows/e2e-workflow.yml

@@ -127,6 +127,14 @@ jobs:
         env:
           REGISTRY: ${{ env.REGISTRY }}
           VERSION: ${{ env.VERSION }}
+
+      - name: build adapter image
+        if: ${{ !inputs.isRelease }}
+        shell: bash
+        run: |
+          make docker-build-adapter
+        env:
+          REGISTRY: ${{ env.REGISTRY }}
 
       - name: create cluster
         shell: bash
@@ -200,6 +208,7 @@ jobs:
         env:
           AZURE_CLUSTER_NAME: ${{ env.CLUSTER_NAME }}
           RUN_LLAMA_13B: ${{ env.RUN_LLAMA_13B }}
+          REGISTRY: ${{ env.REGISTRY }}
           AI_MODELS_REGISTRY: ${{ secrets.E2E_ACR_AMRT_USERNAME }}.azurecr.io
           AI_MODELS_REGISTRY_SECRET: ${{ secrets.E2E_AMRT_SECRET_NAME }}
 

Makefile

@@ -190,6 +190,15 @@ docker-build-kaito: docker-buildx
 		--pull \
 		--tag $(REGISTRY)/$(IMG_NAME):$(IMG_TAG) .
 
+.PHONY: docker-build-adapter
+docker-build-adapter: docker-buildx
+	docker buildx build \
+		--file ./docker/adapter/Dockerfile \
+		--output=$(OUTPUT_TYPE) \
+		--platform="linux/$(ARCH)" \
+		--pull \
+		--tag $(REGISTRY)/e2e-adapter:0.0.1 .
+
 ##@ Deployment
 
 ifndef ignore-not-found

docker/adapter/Dockerfile

@@ -0,0 +1,6 @@
+FROM busybox:latest
+
+RUN mkdir -p /data
+
+COPY docker/adapter/adapter_config.json /data/
+COPY docker/adapter/adapter_model.safetensors /data/

docker/adapter/adapter_config.json

@@ -0,0 +1,29 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": {
+    "base_model_class": "FalconForCausalLM",
+    "parent_library": "transformers.models.falcon.modeling_falcon"
+  },
+  "base_model_name_or_path": "/workspace/tfs/weights",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 8,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "query_key_value"
+  ],
+  "task_type": null,
+  "use_rslora": false
+}

test/e2e/inference_with_adapters.go

@@ -4,9 +4,6 @@
 package e2e
 
 import (
-	"fmt"
-	"os"
-	"strconv"
 	"time"
 
 	kaitov1alpha1 "github.com/azure/kaito/api/v1alpha1"
@@ -22,20 +19,23 @@ import (
 var emptyAdapters = make([]kaitov1alpha1.AdapterSpec, 0)
 var DefaultStrength = "1.0"
 
+var imageName = "e2e-adapter"
+var fullImageName = utils.GetEnv("REGISTRY") + "/" + imageName + ":0.0.1"
+
 var validAdapters = []kaitov1alpha1.AdapterSpec{
 	{
 		Source: &kaitov1alpha1.DataSource{
-			Name:  "falcon-7b-adapter",
-			Image: "aimodelsregistrytest.azurecr.io/adapter-falcon-7b-dolly-oai-busybox:0.0.2",
+			Name:  imageName,
+			Image: fullImageName,
 		},
 		Strength: &DefaultStrength,
 	},
 }
 
 var expectedInitContainers = []corev1.Container{
 	{
-		Name:  "falcon-7b-adapter",
-		Image: "aimodelsregistrytest.azurecr.io/adapter-falcon-7b-dolly-oai-busybox:0.0.2",
+		Name:  imageName,
+		Image: fullImageName,
 	},
 }
 
@@ -75,30 +75,9 @@ func validateAdapters(workspaceObj *kaitov1alpha1.Workspace, expectedInitContain
 
 var _ = Describe("Workspace Preset", func() {
 	BeforeEach(func() {
-		var err error
-		runLlama13B, err = strconv.ParseBool(os.Getenv("RUN_LLAMA_13B"))
-		if err != nil {
-			// Handle error or set a default value
-			fmt.Print("Error: RUN_LLAMA_13B ENV Variable not set")
-			runLlama13B = false
-		}
-
-		aiModelsRegistry = utils.GetEnv("AI_MODELS_REGISTRY")
-		aiModelsRegistrySecret = utils.GetEnv("AI_MODELS_REGISTRY_SECRET")
-		supportedModelsYamlPath = utils.GetEnv("SUPPORTED_MODELS_YAML_PATH")
-
-		// Load stable model versions
-		configs, err := utils.GetModelConfigInfo(supportedModelsYamlPath)
-		if err != nil {
-			fmt.Printf("Failed to load model configs: %v\n", err)
-			os.Exit(1)
-		}
-
-		modelInfo, err = utils.ExtractModelVersion(configs)
-		if err != nil {
-			fmt.Printf("Failed to extract stable model versions: %v\n", err)
-			os.Exit(1)
-		}
+		loadTestEnvVars()
+
+		loadModelVersions()
 	})
 
 	It("should create a falcon workspace with adapter", func() {

test/e2e/preset_test.go

@@ -35,6 +35,34 @@ const (
 	PresetPhi2Model              = "phi-2"
 )
 
+func loadTestEnvVars() {
+	var err error
+	runLlama13B, err = strconv.ParseBool(os.Getenv("RUN_LLAMA_13B"))
+	if err != nil {
+		fmt.Print("Error: RUN_LLAMA_13B ENV Variable not set")
+		runLlama13B = false
+	}
+
+	aiModelsRegistry = utils.GetEnv("AI_MODELS_REGISTRY")
+	aiModelsRegistrySecret = utils.GetEnv("AI_MODELS_REGISTRY_SECRET")
+	supportedModelsYamlPath = utils.GetEnv("SUPPORTED_MODELS_YAML_PATH")
+}
+
+func loadModelVersions() {
+	// Load stable model versions
+	configs, err := utils.GetModelConfigInfo(supportedModelsYamlPath)
+	if err != nil {
+		fmt.Printf("Failed to load model configs: %v\n", err)
+		os.Exit(1)
+	}
+
+	modelInfo, err = utils.ExtractModelVersion(configs)
+	if err != nil {
+		fmt.Printf("Failed to extract stable model versions: %v\n", err)
+		os.Exit(1)
+	}
+}
+
 func createCustomWorkspaceWithAdapter(numOfNode int) *kaitov1alpha1.Workspace {
 	workspaceObj := &kaitov1alpha1.Workspace{}
 	By("Creating a workspace with adapter", func() {
@@ -356,35 +384,14 @@ var modelInfo map[string]string
 
 var _ = Describe("Workspace Preset", func() {
 	BeforeEach(func() {
-		var err error
-		runLlama13B, err = strconv.ParseBool(os.Getenv("RUN_LLAMA_13B"))
-		if err != nil {
-			// Handle error or set a default value
-			fmt.Print("Error: RUN_LLAMA_13B ENV Variable not set")
-			runLlama13B = false
-		}
-
-		aiModelsRegistry = utils.GetEnv("AI_MODELS_REGISTRY")
-		aiModelsRegistrySecret = utils.GetEnv("AI_MODELS_REGISTRY_SECRET")
-		supportedModelsYamlPath = utils.GetEnv("SUPPORTED_MODELS_YAML_PATH")
+		loadTestEnvVars()
 
-		// Load stable model versions
-		configs, err := utils.GetModelConfigInfo(supportedModelsYamlPath)
-		if err != nil {
-			fmt.Printf("Failed to load model configs: %v\n", err)
-			os.Exit(1)
-		}
-
-		modelInfo, err = utils.ExtractModelVersion(configs)
-		if err != nil {
-			fmt.Printf("Failed to extract stable model versions: %v\n", err)
-			os.Exit(1)
-		}
+		loadModelVersions()
 	})
 
-	It("should create a falcon workspace with preset public mode successfully", func() {
+	It("should create a mistral workspace with preset public mode successfully", func() {
 		numOfNode := 1
-		workspaceObj := createFalconWorkspaceWithPresetPublicMode(numOfNode)
+		workspaceObj := createMistralWorkspaceWithPresetPublicMode(numOfNode)
 
 		defer cleanupResources(workspaceObj)
 		time.Sleep(30 * time.Second)
@@ -401,9 +408,9 @@ var _ = Describe("Workspace Preset", func() {
 		validateWorkspaceReadiness(workspaceObj)
 	})
 
-	It("should create a mistral workspace with preset public mode successfully", func() {
+	It("should create a Phi-2 workspace with preset public mode successfully", func() {
 		numOfNode := 1
-		workspaceObj := createMistralWorkspaceWithPresetPublicMode(numOfNode)
+		workspaceObj := createPhi2WorkspaceWithPresetPublicMode(numOfNode)
 
 		defer cleanupResources(workspaceObj)
 		time.Sleep(30 * time.Second)
@@ -420,9 +427,9 @@ var _ = Describe("Workspace Preset", func() {
 		validateWorkspaceReadiness(workspaceObj)
 	})
 
-	It("should create a Phi-2 workspace with preset public mode successfully", func() {
+	It("should create a falcon workspace with preset public mode successfully", func() {
 		numOfNode := 1
-		workspaceObj := createPhi2WorkspaceWithPresetPublicMode(numOfNode)
+		workspaceObj := createFalconWorkspaceWithPresetPublicMode(numOfNode)
 
 		defer cleanupResources(workspaceObj)
 		time.Sleep(30 * time.Second)