replace AIP with IAM principal tag (#1530)

Author: maekawataiki

packages/cdk/lambda/speechToSpeechTask.ts

@@ -416,29 +416,8 @@ export const handler = async (event: { channelId: string; model: Model }) => {
   try {
     console.log('event', event);
 
-    // Speech-to-Speech Inference Profile Arn handling
-    // NOTE: InvokeModelWithBidirectionalStreamCommand currently does not support Inference Profile Arn.
-    // When AWS adds support, uncomment the code block below and change 'const' to 'let' for modelIdOrArn.
-
-    const modelIdOrArn = event.model.modelId; // Fallback to modelId for now
-
-    /*
-    // TODO: Uncomment this block when InvokeModelWithBidirectionalStreamCommand supports Inference Profile Arn
-    // Also change 'const modelIdOrArn' above to 'let modelIdOrArn'
-    try {
-      const speechToSpeechModels = JSON.parse(process.env.SPEECH_TO_SPEECH_MODEL_IDS || '[]');
-      const modelConfig = speechToSpeechModels.find((config: any) => config.modelId === event.model.modelId);
-      if (modelConfig?.inferenceProfileArn) {
-        modelIdOrArn = modelConfig.inferenceProfileArn;
-        console.log('DEBUG: Using Inference Profile ARN for speech-to-speech:', modelIdOrArn);
-      } else {
-        console.log('DEBUG: No inference profile ARN found, using modelId:', modelIdOrArn);
-      }
-    } catch (error) {
-      console.error('DEBUG: Error parsing SPEECH_TO_SPEECH_MODEL_IDS:', error);
-      console.log('DEBUG: Falling back to modelId:', modelIdOrArn);
-    }
-    */
+    // Use modelId directly for speech-to-speech
+    const modelIdOrArn = event.model.modelId;
 
     initialize();
 

packages/cdk/lambda/utils/bedrockApi.ts

@@ -24,7 +24,6 @@ import {
   BEDROCK_TEXT_GEN_MODELS,
   BEDROCK_IMAGE_GEN_MODELS,
   BEDROCK_VIDEO_GEN_MODELS,
-  getInferenceProfileArn,
 } from './models';
 import { streamingChunk } from './streamingChunk';
 import { initBedrockRuntimeClient } from './bedrockClient';
@@ -182,9 +181,8 @@ const bedrockApi: Omit<ApiInterface, 'invokeFlow'> = {
     const client = await initBedrockRuntimeClient({ region });
 
     // Image generation using Stable Diffusion or Titan Image Generator is not supported for the Converse API, so InvokeModelCommand is used.
-    const modelIdOrArn = getInferenceProfileArn(model.modelId) || model.modelId;
     const command = new InvokeModelCommand({
-      modelId: modelIdOrArn,
+      modelId: model.modelId,
       body: createBodyImage(model, params),
       contentType: 'application/json',
     });
@@ -205,9 +203,8 @@ const bedrockApi: Omit<ApiInterface, 'invokeFlow'> = {
       throw new Error('Video tmp buket is not defined');
     }
 
-    const modelIdOrArn = getInferenceProfileArn(model.modelId) || model.modelId;
     const command = new StartAsyncInvokeCommand({
-      modelId: modelIdOrArn,
+      modelId: model.modelId,
       modelInput: createBodyVideo(model, params),
       outputDataConfig: {
         s3OutputDataConfig: {

packages/cdk/lambda/utils/models.ts

@@ -41,9 +41,6 @@ const modelIds: ModelConfiguration[] = (
   .map((model) => ({
     modelId: model.modelId.trim(),
     region: model.region.trim(),
-    ...(model.inferenceProfileArn && {
-      inferenceProfileArn: model.inferenceProfileArn,
-    }),
   }))
   .filter((model) => model.modelId);
 // If there is a lightweight model among the available models, prioritize the lightweight model.
@@ -55,9 +52,6 @@ export const defaultModel: Model = {
   type: 'bedrock',
   modelId: defaultModelConfiguration.modelId,
   region: defaultModelConfiguration.region,
-  ...(defaultModelConfiguration.inferenceProfileArn && {
-    inferenceProfileArn: defaultModelConfiguration.inferenceProfileArn,
-  }),
 };
 
 const imageGenerationModels: ModelConfiguration[] = (
@@ -69,19 +63,13 @@ const imageGenerationModels: ModelConfiguration[] = (
     (model: ModelConfiguration): ModelConfiguration => ({
       modelId: model.modelId.trim(),
       region: model.region.trim(),
-      ...(model.inferenceProfileArn && {
-        inferenceProfileArn: model.inferenceProfileArn,
-      }),
     })
   )
   .filter((model) => model.modelId);
 export const defaultImageGenerationModel: Model = {
   type: 'bedrock',
   modelId: imageGenerationModels?.[0]?.modelId ?? '',
   region: imageGenerationModels?.[0]?.region ?? '',
-  ...(imageGenerationModels?.[0]?.inferenceProfileArn && {
-    inferenceProfileArn: imageGenerationModels[0].inferenceProfileArn,
-  }),
 };
 
 const videoGenerationModels: ModelConfiguration[] = (
@@ -93,19 +81,13 @@ const videoGenerationModels: ModelConfiguration[] = (
     (model: ModelConfiguration): ModelConfiguration => ({
       modelId: model.modelId.trim(),
       region: model.region.trim(),
-      ...(model.inferenceProfileArn && {
-        inferenceProfileArn: model.inferenceProfileArn,
-      }),
     })
   )
   .filter((model) => model.modelId);
 export const defaultVideoGenerationModel: Model = {
   type: 'bedrock',
   modelId: videoGenerationModels?.[0]?.modelId ?? '',
   region: videoGenerationModels?.[0]?.region ?? '',
-  ...(videoGenerationModels?.[0]?.inferenceProfileArn && {
-    inferenceProfileArn: videoGenerationModels[0].inferenceProfileArn,
-  }),
 };
 
 // Model Params
@@ -420,27 +402,6 @@ const mergeConverseInferenceParams = (
     },
   }) as ConverseInferenceParams;
 
-// Get inference profile ARN from modelId
-export const getInferenceProfileArn = (modelId: string): string | undefined => {
-  const textModelConfig = modelIds.find((config) => config.modelId === modelId);
-  if (textModelConfig?.inferenceProfileArn) {
-    return textModelConfig.inferenceProfileArn;
-  }
-  const imageModelConfig = imageGenerationModels.find(
-    (config) => config.modelId === modelId
-  );
-  if (imageModelConfig?.inferenceProfileArn) {
-    return imageModelConfig.inferenceProfileArn;
-  }
-  const videoModelConfig = videoGenerationModels.find(
-    (config) => config.modelId === modelId
-  );
-  if (videoModelConfig?.inferenceProfileArn) {
-    return videoModelConfig.inferenceProfileArn;
-  }
-  return undefined;
-};
-
 // API call, extract string from output, etc.
 
 const createConverseCommandInput = (
@@ -539,9 +500,8 @@ const createConverseCommandInput = (
 
   const guardrailConfig = createGuardrailConfig();
 
-  const modelIdOrArn = getInferenceProfileArn(model.modelId) || model.modelId;
   const converseCommandInput: ConverseCommandInput = {
-    modelId: modelIdOrArn,
+    modelId: model.modelId,
     messages: conversationWithCache,
     system: systemContextWithCache,
     inferenceConfig: params.inferenceConfig,