[ML] Default inference endpoint for ELSER (#113873)

Adds a default configuration for the ELSER model. The config uses
adaptive allocations to automatically scale. Min number of allocations
is set to 1 for this PR, a follow up with change that to 0 and enable
scale from 0.

This end point is always visible in the GET API.

```
GET _inference

{
  "endpoints": [
    {
      "inference_id": ".elser-2",
      "task_type": "sparse_embedding",
      "service": "elser",
      "service_settings": {
        "num_threads": 1,
        "model_id": ".elser_model_2",
        "adaptive_allocations": {
          "enabled": true,
          "min_number_of_allocations": 1,
          "max_number_of_allocations": 8
        }
      },
      "task_settings": {}
    }
  ]
}
```

The default configuration can be used against without any prior setup.
If the model is not downloaded it is automatically downloaded. If it is
not deployed it is deployed

```
POST _inference/.elser-2
{
  "input": "Automagically deploy and infer" 
}
 ... 

{
  "sparse_embedding": [
    {
      "is_truncated": false,
      "embedding": {
        "##fer": 2.2107008,
        "deployment": 2.1624098,
        "deploy": 2.144009,
        "auto": 1.9384763,
```

### Follow up tasks - [ ] Add default config for the E5 text embedding
model - [ ] Select platform specific version - [ ] Scale from 0 - [ ]
Chunking settings - What happens when the end point is deleted, can it
be deleted? - Can the default config be modified - chunking settings for
example? Probably not

Author: davidkyle

docs/changelog/113873.yaml

@@ -0,0 +1,5 @@
+pr: 113873
+summary: Default inference endpoint for ELSER
+area: Machine Learning
+type: enhancement
+issues: []

docs/reference/rest-api/usage.asciidoc

@@ -206,7 +206,12 @@ GET /_xpack/usage
   "inference": {
     "available" : true,
     "enabled" : true,
-    "models" : []
+    "models" : [{
+        "service": "elasticsearch",
+        "task_type": "SPARSE_EMBEDDING",
+        "count": 1
+      }
+    ]
   },
   "logstash" : {
     "available" : true,

server/src/main/java/org/elasticsearch/inference/InferenceService.java

@@ -191,4 +191,13 @@ default Set<TaskType> supportedStreamingTasks() {
     default boolean canStream(TaskType taskType) {
         return supportedStreamingTasks().contains(taskType);
     }
+
+    /**
+     * A service can define default configurations that can be
+     * used out of the box without creating an endpoint first.
+     * @return Default configurations provided by this service
+     */
+    default List<UnparsedModel> defaultConfigs() {
+        return List.of();
+    }
 }

server/src/main/java/org/elasticsearch/inference/UnparsedModel.java

@@ -0,0 +1,24 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the "Elastic License
+ * 2.0", the "GNU Affero General Public License v3.0 only", and the "Server Side
+ * Public License v 1"; you may not use this file except in compliance with, at
+ * your election, the "Elastic License 2.0", the "GNU Affero General Public
+ * License v3.0 only", or the "Server Side Public License, v 1".
+ */
+
+package org.elasticsearch.inference;
+
+import java.util.Map;
+
+/**
+ * Semi parsed model where inference entity id, task type and service
+ * are known but the settings are not parsed.
+ */
+public record UnparsedModel(
+    String inferenceEntityId,
+    TaskType taskType,
+    String service,
+    Map<String, Object> settings,
+    Map<String, Object> secrets
+) {}

test/test-clusters/src/main/java/org/elasticsearch/test/cluster/FeatureFlag.java

@@ -19,7 +19,8 @@ public enum FeatureFlag {
     TIME_SERIES_MODE("es.index_mode_feature_flag_registered=true", Version.fromString("8.0.0"), null),
     FAILURE_STORE_ENABLED("es.failure_store_feature_flag_enabled=true", Version.fromString("8.12.0"), null),
     CHUNKING_SETTINGS_ENABLED("es.inference_chunking_settings_feature_flag_enabled=true", Version.fromString("8.16.0"), null),
-    INFERENCE_SCALE_TO_ZERO("es.inference_scale_to_zero_feature_flag_enabled=true", Version.fromString("8.16.0"), null);
+    INFERENCE_SCALE_TO_ZERO("es.inference_scale_to_zero_feature_flag_enabled=true", Version.fromString("8.16.0"), null),
+    INFERENCE_DEFAULT_ELSER("es.inference_default_elser_feature_flag_enabled=true", Version.fromString("8.16.0"), null);
 
     public final String systemProperty;
     public final Version from;

x-pack/plugin/build.gradle

@@ -83,5 +83,6 @@ tasks.named("precommit").configure {
 tasks.named("yamlRestCompatTestTransform").configure({ task ->
   task.skipTest("security/10_forbidden/Test bulk response with invalid credentials", "warning does not exist for compatibility")
   task.skipTest("wildcard/30_ignore_above_synthetic_source/wildcard field type ignore_above", "Temporary until backported")
+  task.skipTest("inference/inference_crud/Test get all", "Assertions on number of inference models break due to default configs")
 })
 

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/action/StartTrainedModelDeploymentAction.java

@@ -237,7 +237,9 @@ public int computeNumberOfAllocations() {
             if (numberOfAllocations != null) {
                 return numberOfAllocations;
             } else {
-                if (adaptiveAllocationsSettings == null || adaptiveAllocationsSettings.getMinNumberOfAllocations() == null) {
+                if (adaptiveAllocationsSettings == null
+                    || adaptiveAllocationsSettings.getMinNumberOfAllocations() == null
+                    || adaptiveAllocationsSettings.getMinNumberOfAllocations() == 0) {
                     return DEFAULT_NUM_ALLOCATIONS;
                 } else {
                     return adaptiveAllocationsSettings.getMinNumberOfAllocations();

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/inference/assignment/AdaptiveAllocationsSettings.java

@@ -147,8 +147,8 @@ public AdaptiveAllocationsSettings merge(AdaptiveAllocationsSettings updates) {
     public ActionRequestValidationException validate() {
         ActionRequestValidationException validationException = new ActionRequestValidationException();
         boolean hasMinNumberOfAllocations = (minNumberOfAllocations != null && minNumberOfAllocations != -1);
-        if (hasMinNumberOfAllocations && minNumberOfAllocations < 1) {
-            validationException.addValidationError("[" + MIN_NUMBER_OF_ALLOCATIONS + "] must be a positive integer or null");
+        if (hasMinNumberOfAllocations && minNumberOfAllocations < 0) {
+            validationException.addValidationError("[" + MIN_NUMBER_OF_ALLOCATIONS + "] must be a non-negative integer or null");
         }
         boolean hasMaxNumberOfAllocations = (maxNumberOfAllocations != null && maxNumberOfAllocations != -1);
         if (hasMaxNumberOfAllocations && maxNumberOfAllocations < 1) {

x-pack/plugin/core/src/test/java/org/elasticsearch/xpack/core/ml/inference/assignment/AdaptiveAllocationSettingsTests.java

@@ -17,7 +17,7 @@ public class AdaptiveAllocationSettingsTests extends AbstractWireSerializingTest
     public static AdaptiveAllocationsSettings testInstance() {
         return new AdaptiveAllocationsSettings(
             randomBoolean() ? null : randomBoolean(),
-            randomBoolean() ? null : randomIntBetween(1, 2),
+            randomBoolean() ? null : randomIntBetween(0, 2),
             randomBoolean() ? null : randomIntBetween(2, 4)
         );
     }

x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/CustomElandModelIT.java

@@ -85,7 +85,7 @@ public void testSparse() throws IOException {
 
         var inferenceId = "sparse-inf";
         putModel(inferenceId, inferenceConfig, TaskType.SPARSE_EMBEDDING);
-        var results = inferOnMockService(inferenceId, List.of("washing", "machine"));
+        var results = infer(inferenceId, List.of("washing", "machine"));
         deleteModel(inferenceId);
         assertNotNull(results.get("sparse_embedding"));
     }