From 4027642bdf5783e4cda7470772c7dc0a51517ad7 Mon Sep 17 00:00:00 2001
From: David Kyle <david.kyle@elastic.co>
Date: Fri, 22 Nov 2024 11:43:20 +0000
Subject: [PATCH] Explicitly set default chunking settings

---
 .../xpack/inference/DefaultEndPointsIT.java   | 20 +++++++++++++++++++
 .../ElasticsearchInternalService.java         |  5 +++--
 2 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/DefaultEndPointsIT.java b/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/DefaultEndPointsIT.java
index 69767ce0b24f0..ba3e48e11928d 100644
--- a/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/DefaultEndPointsIT.java
+++ b/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/DefaultEndPointsIT.java
@@ -51,6 +51,14 @@ public void tearDown() throws Exception {
         super.tearDown();
     }
 
+    public void testGet() throws IOException {
+        var elserModel = getModel(ElasticsearchInternalService.DEFAULT_ELSER_ID);
+        assertDefaultElserConfig(elserModel);
+
+        var e5Model = getModel(ElasticsearchInternalService.DEFAULT_E5_ID);
+        assertDefaultE5Config(e5Model);
+    }
+
     @SuppressWarnings("unchecked")
     public void testInferDeploysDefaultElser() throws IOException {
         var model = getModel(ElasticsearchInternalService.DEFAULT_ELSER_ID);
@@ -79,6 +87,7 @@ private static void assertDefaultElserConfig(Map<String, Object> modelConfig) {
             adaptiveAllocations,
             Matchers.is(Map.of("enabled", true, "min_number_of_allocations", 0, "max_number_of_allocations", 32))
         );
+        assertDefaultChunkingSettings(modelConfig);
     }
 
     @SuppressWarnings("unchecked")
@@ -113,6 +122,17 @@ private static void assertDefaultE5Config(Map<String, Object> modelConfig) {
             adaptiveAllocations,
             Matchers.is(Map.of("enabled", true, "min_number_of_allocations", 0, "max_number_of_allocations", 32))
         );
+        assertDefaultChunkingSettings(modelConfig);
+    }
+
+    @SuppressWarnings("unchecked")
+    private static void assertDefaultChunkingSettings(Map<String, Object> modelConfig) {
+        var chunkingSettings = (Map<String, Object>) modelConfig.get("chunking_settings");
+        assertThat(
+            modelConfig.toString(),
+            chunkingSettings,
+            Matchers.is(Map.of("strategy", "sentence", "max_chunk_size", 250, "sentence_overlap", 1))
+        );
     }
 
     public void testMultipleInferencesTriggeringDownloadAndDeploy() throws InterruptedException {
diff --git a/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticsearchInternalService.java b/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticsearchInternalService.java
index 718aeae979fe9..6d124906d65bd 100644
--- a/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticsearchInternalService.java
+++ b/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticsearchInternalService.java
@@ -862,6 +862,7 @@ public void updateModelsWithDynamicFields(List<Model> models, ActionListener<Lis
         );
     }
 
+    @Override
     public void defaultConfigs(ActionListener<List<Model>> defaultsListener) {
         preferredModelVariantFn.accept(defaultsListener.delegateFailureAndWrap((delegate, preferredModelVariant) -> {
             if (PreferredModelVariant.LINUX_X86_OPTIMIZED.equals(preferredModelVariant)) {
@@ -892,7 +893,7 @@ private List<Model> defaultConfigs(boolean useLinuxOptimizedModel) {
                 new AdaptiveAllocationsSettings(Boolean.TRUE, 0, 32)
             ),
             ElserMlNodeTaskSettings.DEFAULT,
-            null // default chunking settings
+            ChunkingSettingsBuilder.DEFAULT_SETTINGS
         );
         var defaultE5 = new MultilingualE5SmallModel(
             DEFAULT_E5_ID,
@@ -904,7 +905,7 @@ private List<Model> defaultConfigs(boolean useLinuxOptimizedModel) {
                 useLinuxOptimizedModel ? MULTILINGUAL_E5_SMALL_MODEL_ID_LINUX_X86 : MULTILINGUAL_E5_SMALL_MODEL_ID,
                 new AdaptiveAllocationsSettings(Boolean.TRUE, 0, 32)
             ),
-            null // default chunking settings
+            ChunkingSettingsBuilder.DEFAULT_SETTINGS
         );
         return List.of(defaultElser, defaultE5);
     }