Merge branch 'awslabs:main' into main

CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## UNRELEASED
 
+## v1.1.0
+
 ### **Added**
 
 - added managed autoscaling config to `sagemaker-endpoint` module

DEPLOYMENT.md

@@ -5,7 +5,7 @@
 1. Clone the repository and checkout a release branch using the below command:
 
 ```
-git clone --origin upstream --branch release/1.0.0 https://github.com/awslabs/aiops-modules
+git clone --origin upstream --branch release/1.1.0 https://github.com/awslabs/aiops-modules
 ```
 The release version can be replaced with the version of interest.
 

VERSION

@@ -1 +1 @@
-1.0.0
+1.1.0

examples/manifests/sagemaker-hugging-face.yml

@@ -0,0 +1,24 @@
+name: hugging-face-mistral-endpoint
+path: modules/fmops/sagemaker-hugging-face-endpoint
+targetAccount: primary
+parameters:
+  - name: hugging-face-model-id
+    value: mistralai/Mistral-7B-Instruct-v0.1
+  - name: instance-type
+    value: ml.g5.2xlarge
+  - name: deep-learning-container-image
+    value: huggingface-pytorch-tgi-inference:2.0.1-tgi1.1.0-gpu-py39-cu118-ubuntu20.04
+  - name: hugging-face-token-secret-name
+    value: aiops/huggingface-token
+  - name: vpc_id
+    valueFrom:
+      moduleMetadata:
+        group: networking
+        name: networking
+        key: VpcId
+  - name: subnet_ids
+    valueFrom:
+      moduleMetadata:
+        group: networking
+        name: networking
+        key: PrivateSubnetIds

manifests/fmops-modules.yaml

@@ -18,26 +18,3 @@ parameters:
         group: networking
         name: networking
         key: PrivateSubnetIds
----
-name: hugging-face-mistral-endpoint
-path: modules/fmops/sagemaker-hugging-face-endpoint
-targetAccount: primary
-parameters:
-  - name: hugging-face-model-id
-    value: mistralai/Mixtral-8x7B-Instruct-v0.1
-  - name: instance-type
-    value: ml.g5.2xlarge
-  - name: deep-learning-container-image
-    value: huggingface-pytorch-tgi-inference:2.0.1-tgi1.1.0-gpu-py39-cu118-ubuntu20.04
-  - name: vpc_id
-    valueFrom:
-      moduleMetadata:
-        group: networking
-        name: networking
-        key: VpcId
-  - name: subnet_ids
-    valueFrom:
-      moduleMetadata:
-        group: networking
-        name: networking
-        key: PrivateSubnetIds

modules/fmops/sagemaker-hugging-face-endpoint/README.md

@@ -24,6 +24,7 @@ The module uses [AWS Generative AI CDK Constructs](https://github.com/awslabs/ge
 
 - `vpc-id` - VPC id
 - `subnet-ids` - VPC subnet ids
+- `hugging-face-token-secret-name` - ID of the AWS secret with the Hugging Face access token
 
 ### Module Metadata Outputs
 

modules/fmops/sagemaker-hugging-face-endpoint/index.ts

@@ -17,6 +17,8 @@ const deepLearningContainerImage: string = process.env.SEEDFARMER_PARAMETER_DEEP
 const vpcId: string | undefined = process.env.SEEDFARMER_PARAMETER_VPC_ID;
 const subnetIds: string[] = JSON.parse(process.env.SEEDFARMER_PARAMETER_SUBNET_IDS || ("[]" as string));
 
+const hfTokenSecretName: string | undefined = process.env.SEEDFARMER_PARAMETER_HUGGING_FACE_TOKEN_SECRET_NAME;
+
 const app = new cdk.App();
 
 const stack = new SagemakerHuggingFaceEndpointStack(app, `${projectName}-${deploymentName}-${moduleName}`, {
@@ -28,6 +30,7 @@ const stack = new SagemakerHuggingFaceEndpointStack(app, `${projectName}-${deplo
   deepLearningContainerImage,
   vpcId,
   subnetIds,
+  hfTokenSecretName,
   env: { account, region },
 });
 

modules/fmops/sagemaker-hugging-face-endpoint/lib/sagemaker-hugging-face-endpoint-stack.ts

@@ -2,6 +2,7 @@ import * as cdk from "aws-cdk-lib";
 import { Construct } from "constructs";
 import * as ec2 from "aws-cdk-lib/aws-ec2";
 import * as iam from "aws-cdk-lib/aws-iam";
+import * as secrets from "aws-cdk-lib/aws-secretsmanager";
 import * as sagemaker from "aws-cdk-lib/aws-sagemaker";
 import {
   DeepLearningContainerImage,
@@ -19,6 +20,7 @@ interface SagemakerHuggingFaceEndpointStackProps extends cdk.StackProps {
   deepLearningContainerImage: string;
   vpcId?: string | undefined;
   subnetIds: string[];
+  hfTokenSecretName?: string;
 }
 
 export class SagemakerHuggingFaceEndpointStack extends cdk.Stack {
@@ -83,12 +85,17 @@ export class SagemakerHuggingFaceEndpointStack extends cdk.Stack {
       props.deepLearningContainerImage,
     );
 
+    const hfTokenEnvironmentVars = this.getSecretTokenVars(props.hfTokenSecretName);
+
     this.huggingFaceEndpoint = new HuggingFaceSageMakerEndpoint(this, "HuggingFace Endpoint", {
       modelId: props.huggingFaceModelID,
       instanceType: SageMakerInstanceType.of(props.instanceType),
       container: DeepLearningContainerImage.fromDeepLearningContainerImage(containerImageRepoName, containerImageTag),
       role: this.role,
       vpcConfig: vpcConfig,
+      environment: {
+        ...hfTokenEnvironmentVars,
+      },
     });
 
     cdk_nag.NagSuppressions.addResourceSuppressions(
@@ -111,4 +118,16 @@ export class SagemakerHuggingFaceEndpointStack extends cdk.Stack {
 
     return [parts[0], parts[1]];
   }
+
+  getSecretTokenVars(tokenSecretName: string | undefined): { [key: string]: string } {
+    if (!tokenSecretName) {
+      return {};
+    }
+
+    const secret = secrets.Secret.fromSecretNameV2(this, "HFTokenSecret", tokenSecretName);
+    const token = secret.secretValueFromJson("TOKEN").toString();
+    return {
+      HUGGING_FACE_HUB_TOKEN: token,
+    };
+  }
 }

modules/fmops/sagemaker-hugging-face-endpoint/test/sagemaker-hugging-face-endpoint.test.ts

@@ -1,4 +1,5 @@
 import * as cdk from "aws-cdk-lib";
+import * as secrets from "aws-cdk-lib/aws-secretsmanager";
 import { Annotations, Match, Template } from "aws-cdk-lib/assertions";
 import { SagemakerHuggingFaceEndpointStack } from "../lib/sagemaker-hugging-face-endpoint-stack";
 
@@ -16,6 +17,15 @@ describe("Sagemaker JumpStart Fm Endpoint Stack", () => {
   const account = "123456789";
   const region = "us-east-1";
 
+  const setupStack = new cdk.Stack(app, "setupStack", { env: { account, region } });
+  const secret = new secrets.Secret(setupStack, "secret", {
+    secretName: `${projectName}-${deploymentName}-${moduleName}`,
+    generateSecretString: {
+      secretStringTemplate: JSON.stringify({}),
+      generateStringKey: "TOKEN",
+    },
+  });
+
   const stack = new SagemakerHuggingFaceEndpointStack(app, `${projectName}-${deploymentName}-${moduleName}`, {
     projectName,
     deploymentName,
@@ -25,6 +35,7 @@ describe("Sagemaker JumpStart Fm Endpoint Stack", () => {
     deepLearningContainerImage,
     vpcId,
     subnetIds,
+    hfTokenSecretName: secret.secretName,
     env: { account, region },
   });
 
@@ -36,6 +47,18 @@ describe("Sagemaker JumpStart Fm Endpoint Stack", () => {
     template.hasResource("AWS::EC2::SecurityGroup", {});
   });
 
+  test("Hugging Face Token", () => {
+    const template = Template.fromStack(stack);
+
+    template.hasResourceProperties("AWS::SageMaker::Model", {
+      PrimaryContainer: {
+        Environment: {
+          HUGGING_FACE_HUB_TOKEN: Match.anyValue(),
+        },
+      },
+    });
+  });
+
   test("No CDK Nag Errors", () => {
     const errors = Annotations.fromStack(stack).findError("*", Match.stringLikeRegexp("AwsSolutions-.*"));
     expect(errors).toHaveLength(0);