add temp registry

Signed-off-by: Bangqi Zhu <[email protected]>

.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 /mnt/adapters
+
+COPY docker/adapter/adapter_config.json /mnt/adapters/
+COPY docker/adapter/adapter_model.safetensors /mnt/adapters/

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

@@ -22,11 +22,13 @@ import (
 var emptyAdapters = make([]kaitov1alpha1.AdapterSpec, 0)
 var DefaultStrength = "1.0"
 
+var fullImageName = os.Getenv("REGISTRY") + "/e2e-adapter: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",
+			Image: fullImageName,
 		},
 		Strength: &DefaultStrength,
 	},
@@ -35,7 +37,7 @@ var validAdapters = []kaitov1alpha1.AdapterSpec{
 var expectedInitContainers = []corev1.Container{
 	{
 		Name:  "falcon-7b-adapter",
-		Image: "aimodelsregistrytest.azurecr.io/adapter-falcon-7b-dolly-oai-busybox:0.0.2",
+		Image: fullImageName,
 	},
 }