[ML] Custom Service add embedding type support

server/src/main/java/org/elasticsearch/TransportVersions.java

@@ -334,6 +334,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion ML_INFERENCE_IBM_WATSONX_COMPLETION_ADDED = def(9_115_0_00);
     public static final TransportVersion ESQL_SPLIT_ON_BIG_VALUES = def(9_116_0_00);
     public static final TransportVersion ESQL_LOCAL_RELATION_WITH_NEW_BLOCKS = def(9_117_0_00);
+    public static final TransportVersion ML_INFERENCE_CUSTOM_SERVICE_EMBEDDING_TYPE = def(9_118_0_00);
 
     /*
      * STOP! READ THIS FIRST! No, really,

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/ServiceUtils.java

@@ -1085,5 +1085,11 @@ public static void validateInputTypeAgainstAllowlist(
         }
     }
 
+    public static void checkByteBounds(short value) {
+        if (value < Byte.MIN_VALUE || value > Byte.MAX_VALUE) {
+            throw new IllegalArgumentException("Value [" + value + "] is out of range for a byte");
+        }
+    }
+
     private ServiceUtils() {}
 }

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/CustomModel.java

@@ -46,7 +46,7 @@ public CustomModel(
             inferenceId,
             taskType,
             service,
-            CustomServiceSettings.fromMap(serviceSettings, context, taskType, inferenceId),
+            CustomServiceSettings.fromMap(serviceSettings, context, taskType),
             CustomTaskSettings.fromMap(taskSettings),
             CustomSecretSettings.fromMap(secrets)
         );
@@ -66,7 +66,7 @@ public CustomModel(
             inferenceId,
             taskType,
             service,
-            CustomServiceSettings.fromMap(serviceSettings, context, taskType, inferenceId),
+            CustomServiceSettings.fromMap(serviceSettings, context, taskType),
             CustomTaskSettings.fromMap(taskSettings),
             CustomSecretSettings.fromMap(secrets),
             chunkingSettings

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/CustomService.java

@@ -333,12 +333,7 @@ private static CustomServiceSettings getCustomServiceSettings(CustomModel custom
         var similarityToUse = similarityFromModel == null ? SimilarityMeasure.DOT_PRODUCT : similarityFromModel;
 
         return new CustomServiceSettings(
-            new CustomServiceSettings.TextEmbeddingSettings(
-                similarityToUse,
-                embeddingSize,
-                serviceSettings.getMaxInputTokens(),
-                serviceSettings.elementType()
-            ),
+            new CustomServiceSettings.TextEmbeddingSettings(similarityToUse, embeddingSize, serviceSettings.getMaxInputTokens()),
             serviceSettings.getUrl(),
             serviceSettings.getHeaders(),
             serviceSettings.getQueryParameters(),

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/CustomServiceEmbeddingType.java

@@ -0,0 +1,50 @@
+/*
+ * 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; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+package org.elasticsearch.xpack.inference.services.custom;
+
+import org.elasticsearch.index.mapper.vectors.DenseVectorFieldMapper;
+
+import java.util.Locale;
+
+public enum CustomServiceEmbeddingType {
+    /**
+     * Use this when you want to get back the default float embeddings.
+     */
+    FLOAT(DenseVectorFieldMapper.ElementType.FLOAT),
+    /**
+     * Use this when you want to get back signed int8 embeddings.
+     */
+    BYTE(DenseVectorFieldMapper.ElementType.BYTE),
+    /**
+     * Use this when you want to get back binary embeddings.
+     */
+    BIT(DenseVectorFieldMapper.ElementType.BIT),
+    /**
+     * This is a synonym for BIT
+     */
+    BINARY(DenseVectorFieldMapper.ElementType.BIT);
+
+    private final DenseVectorFieldMapper.ElementType elementType;
+
+    CustomServiceEmbeddingType(DenseVectorFieldMapper.ElementType elementType) {
+        this.elementType = elementType;
+    }
+
+    @Override
+    public String toString() {
+        return name().toLowerCase(Locale.ROOT);
+    }
+
+    public DenseVectorFieldMapper.ElementType toElementType() {
+        return elementType;
+    }
+
+    public static CustomServiceEmbeddingType fromString(String name) {
+        return valueOf(name.trim().toUpperCase(Locale.ROOT));
+    }
+}

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/CustomServiceSettings.java

@@ -66,12 +66,7 @@ public class CustomServiceSettings extends FilteredXContentObject implements Ser
     private static final String RESPONSE_SCOPE = String.join(".", ModelConfigurations.SERVICE_SETTINGS, RESPONSE);
     private static final int DEFAULT_EMBEDDING_BATCH_SIZE = 10;
 
-    public static CustomServiceSettings fromMap(
-        Map<String, Object> map,
-        ConfigurationParseContext context,
-        TaskType taskType,
-        String inferenceId
-    ) {
+    public static CustomServiceSettings fromMap(Map<String, Object> map, ConfigurationParseContext context, TaskType taskType) {
         ValidationException validationException = new ValidationException();
 
         var textEmbeddingSettings = TextEmbeddingSettings.fromMap(map, taskType, validationException);
@@ -137,22 +132,12 @@ public static CustomServiceSettings fromMap(
         );
     }
 
-    public record TextEmbeddingSettings(
-        @Nullable SimilarityMeasure similarityMeasure,
-        @Nullable Integer dimensions,
-        @Nullable Integer maxInputTokens,
-        @Nullable DenseVectorFieldMapper.ElementType elementType
-    ) implements ToXContentFragment, Writeable {
+    public static class TextEmbeddingSettings implements ToXContentFragment, Writeable {
 
         // This specifies float for the element type but null for all other settings
-        public static final TextEmbeddingSettings DEFAULT_FLOAT = new TextEmbeddingSettings(
-            null,
-            null,
-            null,
-            DenseVectorFieldMapper.ElementType.FLOAT
-        );
+        public static final TextEmbeddingSettings DEFAULT_FLOAT = new TextEmbeddingSettings(null, null, null);
         // This refers to settings that are not related to the text embedding task type (all the settings should be null)
-        public static final TextEmbeddingSettings NON_TEXT_EMBEDDING_TASK_TYPE_SETTINGS = new TextEmbeddingSettings(null, null, null, null);
+        public static final TextEmbeddingSettings NON_TEXT_EMBEDDING_TASK_TYPE_SETTINGS = new TextEmbeddingSettings(null, null, null);
 
         public static TextEmbeddingSettings fromMap(Map<String, Object> map, TaskType taskType, ValidationException validationException) {
             if (taskType != TaskType.TEXT_EMBEDDING) {
@@ -162,24 +147,42 @@ public static TextEmbeddingSettings fromMap(Map<String, Object> map, TaskType ta
             SimilarityMeasure similarity = extractSimilarity(map, ModelConfigurations.SERVICE_SETTINGS, validationException);
             Integer dims = removeAsType(map, DIMENSIONS, Integer.class);
             Integer maxInputTokens = removeAsType(map, MAX_INPUT_TOKENS, Integer.class);
-            return new TextEmbeddingSettings(similarity, dims, maxInputTokens, DenseVectorFieldMapper.ElementType.FLOAT);
+            return new TextEmbeddingSettings(similarity, dims, maxInputTokens);
+        }
+
+        private final SimilarityMeasure similarityMeasure;
+        private final Integer dimensions;
+        private final Integer maxInputTokens;
+
+        public TextEmbeddingSettings(
+            @Nullable SimilarityMeasure similarityMeasure,
+            @Nullable Integer dimensions,
+            @Nullable Integer maxInputTokens
+        ) {
+            this.similarityMeasure = similarityMeasure;
+            this.dimensions = dimensions;
+            this.maxInputTokens = maxInputTokens;
         }
 
         public TextEmbeddingSettings(StreamInput in) throws IOException {
-            this(
-                in.readOptionalEnum(SimilarityMeasure.class),
-                in.readOptionalVInt(),
-                in.readOptionalVInt(),
-                in.readOptionalEnum(DenseVectorFieldMapper.ElementType.class)
-            );
+            this.similarityMeasure = in.readOptionalEnum(SimilarityMeasure.class);
+            this.dimensions = in.readOptionalVInt();
+            this.maxInputTokens = in.readOptionalVInt();
+
+            if (in.getTransportVersion().before(TransportVersions.ML_INFERENCE_CUSTOM_SERVICE_EMBEDDING_TYPE)) {
+                in.readOptionalEnum(DenseVectorFieldMapper.ElementType.class);
+            }
         }
 
         @Override
         public void writeTo(StreamOutput out) throws IOException {
             out.writeOptionalEnum(similarityMeasure);
             out.writeOptionalVInt(dimensions);
             out.writeOptionalVInt(maxInputTokens);
-            out.writeOptionalEnum(elementType);
+
+            if (out.getTransportVersion().before(TransportVersions.ML_INFERENCE_CUSTOM_SERVICE_EMBEDDING_TYPE)) {
+                out.writeOptionalEnum(null);
+            }
         }
 
         @Override
@@ -193,8 +196,23 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
             if (maxInputTokens != null) {
                 builder.field(MAX_INPUT_TOKENS, maxInputTokens);
             }
+
             return builder;
         }
+
+        @Override
+        public boolean equals(Object o) {
+            if (o == null || getClass() != o.getClass()) return false;
+            TextEmbeddingSettings that = (TextEmbeddingSettings) o;
+            return similarityMeasure == that.similarityMeasure
+                && Objects.equals(dimensions, that.dimensions)
+                && Objects.equals(maxInputTokens, that.maxInputTokens);
+        }
+
+        @Override
+        public int hashCode() {
+            return Objects.hash(similarityMeasure, dimensions, maxInputTokens);
+        }
     }
 
     private final TextEmbeddingSettings textEmbeddingSettings;
@@ -300,7 +318,12 @@ public Integer dimensions() {
 
     @Override
     public DenseVectorFieldMapper.ElementType elementType() {
-        return textEmbeddingSettings.elementType;
+        var embeddingType = responseJsonParser.getEmbeddingType();
+        if (embeddingType != null) {
+            return embeddingType.toElementType();
+        }
+
+        return null;
     }
 
     public Integer getMaxInputTokens() {

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/response/BaseCustomResponseParser.java

@@ -22,7 +22,9 @@
 import java.util.Objects;
 import java.util.function.BiFunction;
 
-public abstract class BaseCustomResponseParser<T extends InferenceServiceResults> implements CustomResponseParser {
+import static org.elasticsearch.xpack.inference.services.ServiceUtils.checkByteBounds;
+
+public abstract class BaseCustomResponseParser implements CustomResponseParser {
 
     @Override
     public InferenceServiceResults parse(HttpResult response) throws IOException {
@@ -36,7 +38,7 @@ public InferenceServiceResults parse(HttpResult response) throws IOException {
         }
     }
 
-    protected abstract T transform(Map<String, Object> extractedField);
+    protected abstract InferenceServiceResults transform(Map<String, Object> extractedField);
 
     static List<?> validateList(Object obj, String fieldName) {
         validateNonNull(obj, fieldName);
@@ -97,6 +99,21 @@ static Float toFloat(Object obj, String fieldName) {
         return toNumber(obj, fieldName).floatValue();
     }
 
+    static List<Byte> convertToListOfBits(Object obj, String fieldName) {
+        return convertToListOfBytes(obj, fieldName);
+    }
+
+    static List<Byte> convertToListOfBytes(Object obj, String fieldName) {
+        return castList(validateList(obj, fieldName), BaseCustomResponseParser::toByte, fieldName);
+    }
+
+    static Byte toByte(Object obj, String fieldName) {
+        var shortValue = toNumber(obj, fieldName).shortValue();
+        checkByteBounds(shortValue);
+
+        return (byte) shortValue;
+    }
+
     private static Number toNumber(Object obj, String fieldName) {
         if (obj instanceof Number == false) {
             throw new IllegalArgumentException(

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/response/CompletionResponseParser.java

@@ -23,7 +23,7 @@
 import static org.elasticsearch.xpack.inference.services.ServiceUtils.extractRequiredString;
 import static org.elasticsearch.xpack.inference.services.custom.CustomServiceSettings.JSON_PARSER;
 
-public class CompletionResponseParser extends BaseCustomResponseParser<ChatCompletionResults> {
+public class CompletionResponseParser extends BaseCustomResponseParser {
 
     public static final String NAME = "completion_response_parser";
     public static final String COMPLETION_PARSER_RESULT = "completion_result";

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/response/CustomResponseParser.java

@@ -11,9 +11,17 @@
 import org.elasticsearch.inference.InferenceServiceResults;
 import org.elasticsearch.xcontent.ToXContentFragment;
 import org.elasticsearch.xpack.inference.external.http.HttpResult;
+import org.elasticsearch.xpack.inference.services.custom.CustomServiceEmbeddingType;
 
 import java.io.IOException;
 
 public interface CustomResponseParser extends ToXContentFragment, NamedWriteable {
     InferenceServiceResults parse(HttpResult response) throws IOException;
+
+    /**
+     * Returns the configured embedding type for this response parser. This should be overridden for text embedding parsers.
+     */
+    default CustomServiceEmbeddingType getEmbeddingType() {
+        return null;
+    }
 }

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/response/RerankResponseParser.java

@@ -26,7 +26,7 @@
 import static org.elasticsearch.xpack.inference.services.ServiceUtils.extractRequiredString;
 import static org.elasticsearch.xpack.inference.services.custom.CustomServiceSettings.JSON_PARSER;
 
-public class RerankResponseParser extends BaseCustomResponseParser<RankedDocsResults> {
+public class RerankResponseParser extends BaseCustomResponseParser {
 
     public static final String NAME = "rerank_response_parser";
     public static final String RERANK_PARSER_SCORE = "relevance_score";

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/custom/response/SparseEmbeddingResponseParser.java

@@ -26,7 +26,7 @@
 import static org.elasticsearch.xpack.inference.services.ServiceUtils.extractRequiredString;
 import static org.elasticsearch.xpack.inference.services.custom.CustomServiceSettings.JSON_PARSER;
 
-public class SparseEmbeddingResponseParser extends BaseCustomResponseParser<SparseEmbeddingResults> {
+public class SparseEmbeddingResponseParser extends BaseCustomResponseParser {
 
     public static final String NAME = "sparse_embedding_response_parser";
     public static final String SPARSE_EMBEDDING_TOKEN_PATH = "token_path";