[8.19] [Obs AI Assistant] Enable ELSER and e5 on EIS (elastic#220993) (elastic#221148)

kibanamachine · viduni94 · web-flow · commit defd0043a54d · 2025-05-21T19:32:00.000+02:00
# Backport This will backport the following commits from `main` to `8.19`: - [[Obs AI Assistant] Enable ELSER and e5 on EIS (elastic#220993)](elastic#220993)  ### Questions ? Please refer to the [Backport tool documentation](https://github.com/sorenlouv/backport)  Co-authored-by: Viduni Wickramarachchi <viduni.wickramarachchi@elastic.co>
diff --git a/x-pack/platform/packages/shared/kbn-ai-assistant/src/utils/get_model_options_for_inference_endpoints.test.ts b/x-pack/platform/packages/shared/kbn-ai-assistant/src/utils/get_model_options_for_inference_endpoints.test.ts
@@ -17,6 +17,8 @@ import type { InferenceAPIConfigResponse } from '@kbn/ml-trained-models-utils';
 import {
   ELSER_ON_ML_NODE_INFERENCE_ID,
   E5_SMALL_INFERENCE_ID,
+  ELSER_IN_EIS_INFERENCE_ID,
+  E5_LARGE_IN_EIS_INFERENCE_ID,
 } from '@kbn/observability-ai-assistant-plugin/public';
 
 describe('getModelOptionsForInferenceEndpoints', () => {
@@ -44,6 +46,55 @@ describe('getModelOptionsForInferenceEndpoints', () => {
     ]);
   });
 
+  it('shows only ELSER in EIS when both ELSER models are available', () => {
+    const endpoints = [
+      { inference_id: ELSER_ON_ML_NODE_INFERENCE_ID },
+      { inference_id: ELSER_IN_EIS_INFERENCE_ID },
+      { inference_id: E5_SMALL_INFERENCE_ID },
+    ] as InferenceAPIConfigResponse[];
+
+    const options = getModelOptionsForInferenceEndpoints({
+      endpoints,
+    });
+
+    expect(options.map((o) => o.key)).toEqual([ELSER_IN_EIS_INFERENCE_ID, E5_SMALL_INFERENCE_ID]);
+  });
+
+  it('shows only E5-large in EIS when both E5 (small and large) models are available', () => {
+    const endpoints = [
+      { inference_id: ELSER_ON_ML_NODE_INFERENCE_ID },
+      { inference_id: E5_SMALL_INFERENCE_ID },
+      { inference_id: E5_LARGE_IN_EIS_INFERENCE_ID },
+    ] as InferenceAPIConfigResponse[];
+
+    const options = getModelOptionsForInferenceEndpoints({
+      endpoints,
+    });
+
+    expect(options.map((o) => o.key)).toEqual([
+      ELSER_ON_ML_NODE_INFERENCE_ID,
+      E5_LARGE_IN_EIS_INFERENCE_ID,
+    ]);
+  });
+
+  it('shows only EIS models when both ELSER and E5-large pre-configured endpoints are available in EIS', () => {
+    const endpoints = [
+      { inference_id: ELSER_ON_ML_NODE_INFERENCE_ID },
+      { inference_id: ELSER_IN_EIS_INFERENCE_ID },
+      { inference_id: E5_SMALL_INFERENCE_ID },
+      { inference_id: E5_LARGE_IN_EIS_INFERENCE_ID },
+    ] as InferenceAPIConfigResponse[];
+
+    const options = getModelOptionsForInferenceEndpoints({
+      endpoints,
+    });
+
+    expect(options.map((o) => o.key)).toEqual([
+      ELSER_IN_EIS_INFERENCE_ID,
+      E5_LARGE_IN_EIS_INFERENCE_ID,
+    ]);
+  });
+
   it('does not show any EIS models if EIS pre-configured endpoints are not available', () => {
     const endpoints = [
       { inference_id: ELSER_ON_ML_NODE_INFERENCE_ID },
diff --git a/x-pack/platform/packages/shared/kbn-ai-assistant/src/utils/get_model_options_for_inference_endpoints.ts b/x-pack/platform/packages/shared/kbn-ai-assistant/src/utils/get_model_options_for_inference_endpoints.ts
@@ -10,6 +10,8 @@ import { InferenceAPIConfigResponse } from '@kbn/ml-trained-models-utils';
 import {
   ELSER_ON_ML_NODE_INFERENCE_ID,
   E5_SMALL_INFERENCE_ID,
+  ELSER_IN_EIS_INFERENCE_ID,
+  E5_LARGE_IN_EIS_INFERENCE_ID,
 } from '@kbn/observability-ai-assistant-plugin/public';
 
 export interface ModelOptionsData {
@@ -48,6 +50,21 @@ export const e5SmallDescription = i18n.translate(
   }
 );
 
+const e5LargeTitle = i18n.translate(
+  'xpack.aiAssistant.welcomeMessage.knowledgeBase.model.e5largeTitle',
+  {
+    defaultMessage: 'E5-large (multilingual)',
+  }
+);
+
+const e5LargeDescription = i18n.translate(
+  'xpack.aiAssistant.welcomeMessage.knowledgeBase.model.e5largeDescription',
+  {
+    defaultMessage:
+      'E5 is an NLP model by Elastic designed to enhance multilingual semantic search by focusing on query context rather than keywords. E5-large is an optimized version for Intel® silicon.',
+  }
+);
+
 const PRECONFIGURED_INFERENCE_ENDPOINT_METADATA: Record<
   string,
   { title: string; description: string }
@@ -56,19 +73,40 @@ const PRECONFIGURED_INFERENCE_ENDPOINT_METADATA: Record<
     title: elserTitle,
     description: elserDescription,
   },
+  [ELSER_IN_EIS_INFERENCE_ID]: {
+    title: elserTitle,
+    description: elserDescription,
+  },
   [E5_SMALL_INFERENCE_ID]: {
     title: e5SmallTitle,
     description: e5SmallDescription,
   },
+  [E5_LARGE_IN_EIS_INFERENCE_ID]: {
+    title: e5LargeTitle,
+    description: e5LargeDescription,
+  },
 };
 
 export const getModelOptionsForInferenceEndpoints = ({
   endpoints,
 }: {
   endpoints: InferenceAPIConfigResponse[];
 }): ModelOptionsData[] => {
+  const hasElserEIS = endpoints.some((ep) => ep.inference_id === ELSER_IN_EIS_INFERENCE_ID);
+  const hasE5EIS = endpoints.some((ep) => ep.inference_id === E5_LARGE_IN_EIS_INFERENCE_ID);
+
   return endpoints
     .filter((endpoint) => {
+      // if ELSER exists in EIS, skip the other ELSER model
+      if (endpoint.inference_id === ELSER_ON_ML_NODE_INFERENCE_ID && hasElserEIS) {
+        return false;
+      }
+
+      // if e5-large exists in EIS, skip the e5-small
+      if (endpoint.inference_id === E5_SMALL_INFERENCE_ID && hasE5EIS) {
+        return false;
+      }
+
       // Only include preconfigured endpoints and skip custom endpoints
       return Boolean(PRECONFIGURED_INFERENCE_ENDPOINT_METADATA[endpoint.inference_id]);
     })
diff --git a/x-pack/platform/plugins/shared/observability_ai_assistant/common/index.ts b/x-pack/platform/plugins/shared/observability_ai_assistant/common/index.ts
@@ -58,4 +58,5 @@ export {
   ELSER_IN_EIS_INFERENCE_ID,
   E5_SMALL_INFERENCE_ID,
   E5_LARGE_IN_EIS_INFERENCE_ID,
+  EIS_PRECONFIGURED_INFERENCE_IDS,
 } from './preconfigured_inference_ids';
diff --git a/x-pack/platform/plugins/shared/observability_ai_assistant/common/preconfigured_inference_ids.ts b/x-pack/platform/plugins/shared/observability_ai_assistant/common/preconfigured_inference_ids.ts
@@ -10,3 +10,8 @@ export const ELSER_IN_EIS_INFERENCE_ID = '.elser-v2-elastic';
 
 export const E5_SMALL_INFERENCE_ID = '.multilingual-e5-small-elasticsearch';
 export const E5_LARGE_IN_EIS_INFERENCE_ID = '.multilingual-e5-large-elastic'; // TODO: verify the inference ID once it's created in EIS
+
+export const EIS_PRECONFIGURED_INFERENCE_IDS = [
+  ELSER_IN_EIS_INFERENCE_ID,
+  E5_LARGE_IN_EIS_INFERENCE_ID,
+];
diff --git a/x-pack/platform/plugins/shared/observability_ai_assistant/public/index.ts b/x-pack/platform/plugins/shared/observability_ai_assistant/public/index.ts
@@ -73,6 +73,7 @@ export {
   ELSER_IN_EIS_INFERENCE_ID,
   E5_SMALL_INFERENCE_ID,
   E5_LARGE_IN_EIS_INFERENCE_ID,
+  EIS_PRECONFIGURED_INFERENCE_IDS,
 } from '../common';
 
 export type {
diff --git a/x-pack/platform/plugins/shared/observability_ai_assistant/server/service/inference_endpoint.ts b/x-pack/platform/plugins/shared/observability_ai_assistant/server/service/inference_endpoint.ts
@@ -16,7 +16,7 @@ import {
 import { InferenceAPIConfigResponse } from '@kbn/ml-trained-models-utils';
 import pRetry from 'p-retry';
 import { CoreSetup } from '@kbn/core/server';
-import { KnowledgeBaseState } from '../../common';
+import { EIS_PRECONFIGURED_INFERENCE_IDS, KnowledgeBaseState } from '../../common';
 import { ObservabilityAIAssistantConfig } from '../config';
 import {
   getConcreteWriteIndex,
@@ -128,12 +128,28 @@ export async function getKbModelStatus({
     inferenceId = currentInferenceId;
   }
 
+  // check if inference ID is an EIS inference ID
+  const isPreConfiguredInferenceIdInEIS = EIS_PRECONFIGURED_INFERENCE_IDS.includes(inferenceId);
+
   let endpoint: InferenceInferenceEndpointInfo;
   try {
     endpoint = await getInferenceEndpoint({ esClient, inferenceId });
     logger.debug(
       `Inference endpoint "${inferenceId}" found with model id "${endpoint?.service_settings?.model_id}"`
     );
+
+    // if the endpoint is in EIS, the model doesn't have to be downloaded and deployed
+    // Therefore, return the KB state as READY if the endpoint exists
+    if (isPreConfiguredInferenceIdInEIS && endpoint.service === 'elastic') {
+      return {
+        endpoint,
+        enabled,
+        kbState: KnowledgeBaseState.READY,
+        currentInferenceId,
+        concreteWriteIndex,
+        isReIndexing,
+      };
+    }
   } catch (error) {
     if (!isInferenceEndpointMissingOrUnavailable(error)) {
       throw error;
@@ -242,6 +258,8 @@ export async function waitForKbModel({
         logger.debug('Knowledge base model is not yet ready. Retrying...');
         throw new Error('Knowledge base model is not yet ready');
       }
+
+      logger.debug('Knowledge base model is ready.');
     },
     { retries: 30, factor: 2, maxTimeout: 30_000 }
   );
