[ML] Add prefix strings option to trained models

docs/changelog/101978.yaml

@@ -0,0 +1,5 @@
+pr: 101978
+summary: Add prefix strings option to trained models
+area: Machine Learning
+type: enhancement
+issues: []

docs/reference/ml/trained-models/apis/put-trained-models.asciidoc

@@ -443,7 +443,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -469,7 +469,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the 
+Refer to <<tokenization-properties>> to review the
 properties of the `tokenization` object.
 =====
 
@@ -488,7 +488,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -514,7 +514,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenizati
 Recommended to set `max_sentence_length` to `386` with `128` of `span` and set
 `truncate` to `none`.
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -546,7 +546,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-text-classific
 
 `num_top_classes`::::
 (Optional, integer)
-Specifies the number of top class predictions to return. Defaults to all classes 
+Specifies the number of top class predictions to return. Defaults to all classes
 (-1).
 
 `results_field`::::
@@ -557,7 +557,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -580,7 +580,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -599,7 +599,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-text-similarit
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 
@@ -634,7 +634,7 @@ include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-results-field]
 (Optional, object)
 include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-nlp-tokenization]
 +
-Refer to <<tokenization-properties>> to review the properties of the 
+Refer to <<tokenization-properties>> to review the properties of the
 `tokenization` object.
 =====
 ====
@@ -701,6 +701,32 @@ the platform identifiers used by Elasticsearch, so one of, `linux-x86_64`,
 For portable models (those that work independent of processor architecture or
 OS features), leave this field unset.
 
+//Begin prefix_strings
+`prefix_strings`::
+(Optional, object)
+Certain NLP models are trained in such a way that a prefix string should
+be applied to the input text before the input is evaluated. The prefix
+may be different depending on the intention. For asymmetric tasks such
+as infromation retrieval the prefix applied to a passage as it is indexed
+can be different to the prefix applied when searching those passages.
+
+`prefix_strings` has 2 options, a prefix string that is always applied
+in the search context and one that is always applied when ingesting the
+docs. Both are optional.
++
+.Properties of `prefix_strings`
+[%collapsible%open]
+====
+`search`:::
+(Optional, string)
+The prefix string to prepend to the input text for requests
+originating from a search query.
+`ingest`:::
+(Optional, string)
+The prefix string to prepend to the input text for requests
+at ingest where the Inference ingest processor is used. // TODO is there a shortcut for Inference ingest processor?
+====
+//End prefix_strings
 
 `tags`::
 (Optional, string)

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

@@ -161,7 +161,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion UNDESIRED_SHARD_ALLOCATIONS_COUNT_ADDED = def(8_530_00_0);
     public static final TransportVersion ML_INFERENCE_TASK_SETTINGS_OPTIONAL_ADDED = def(8_531_00_0);
     public static final TransportVersion DEPRECATED_COMPONENT_TEMPLATES_ADDED = def(8_532_00_0);
-
+    public static final TransportVersion ML_TRAINED_MODEL_PREFIX_STRINGS_ADDED = def(8_533_00_0);
     /*
      * STOP! READ THIS FIRST! No, really,
      *        ____ _____ ___  ____  _        ____  _____    _    ____    _____ _   _ ___ ____    _____ ___ ____  ____ _____ _

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

@@ -150,13 +150,14 @@ private static void checkUniqueness(int id, String uniqueId) {
 
     public static final MlConfigVersion V_10 = registerMlConfigVersion(10_00_00_99, "4B940FD9-BEDD-4589-8E08-02D9B480B22D");
     // V_11 is used in ELSER v2 package configs
-    public static final MlConfigVersion V_11 = registerMlConfigVersion(11_00_00_99, "79CB2950-57C7-11EE-AE5D-0800200C9A66");
+    public static final MlConfigVersion V_11 = registerMlConfigVersion(11_00_0_0_99, "79CB2950-57C7-11EE-AE5D-0800200C9A66");
+    public static final MlConfigVersion V_12 = registerMlConfigVersion(12_00_0_0_99, "Trained model config prefix strings added");
 
     /**
      * Reference to the most recent Ml config version.
      * This should be the Ml config version with the highest id.
      */
-    public static final MlConfigVersion CURRENT = V_11;
+    public static final MlConfigVersion CURRENT = V_12;
 
     /**
      * Reference to the first MlConfigVersion that is detached from the

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

@@ -23,6 +23,7 @@
 import org.elasticsearch.xcontent.ToXContentObject;
 import org.elasticsearch.xcontent.XContentBuilder;
 import org.elasticsearch.xcontent.XContentParser;
+import org.elasticsearch.xpack.core.ml.inference.TrainedModelPrefixStrings;
 import org.elasticsearch.xpack.core.ml.inference.trainedmodel.EmptyConfigUpdate;
 import org.elasticsearch.xpack.core.ml.inference.trainedmodel.InferenceConfigUpdate;
 import org.elasticsearch.xpack.core.ml.utils.ExceptionsHelper;
@@ -88,6 +89,7 @@ public static Builder parseRequest(String id, XContentParser parser) {
         // input and so cannot construct a document.
         private final List<String> textInput;
         private boolean highPriority;
+        private TrainedModelPrefixStrings.PrefixType prefixType = TrainedModelPrefixStrings.PrefixType.NONE;
 
         /**
          * Build a request from a list of documents as maps.
@@ -190,6 +192,11 @@ public Request(StreamInput in) throws IOException {
             if (in.getTransportVersion().onOrAfter(TransportVersions.V_8_8_0)) {
                 highPriority = in.readBoolean();
             }
+            if (in.getTransportVersion().onOrAfter(TransportVersions.ML_TRAINED_MODEL_PREFIX_STRINGS_ADDED)) {
+                prefixType = in.readEnum(TrainedModelPrefixStrings.PrefixType.class);
+            } else {
+                prefixType = TrainedModelPrefixStrings.PrefixType.NONE;
+            }
         }
 
         public int numberOfDocuments() {
@@ -232,6 +239,14 @@ public void setHighPriority(boolean highPriority) {
             this.highPriority = highPriority;
         }
 
+        public void setPrefixType(TrainedModelPrefixStrings.PrefixType prefixType) {
+            this.prefixType = prefixType;
+        }
+
+        public TrainedModelPrefixStrings.PrefixType getPrefixType() {
+            return prefixType;
+        }
+
         @Override
         public ActionRequestValidationException validate() {
             return null;
@@ -253,6 +268,9 @@ public void writeTo(StreamOutput out) throws IOException {
             if (out.getTransportVersion().onOrAfter(TransportVersions.V_8_8_0)) {
                 out.writeBoolean(highPriority);
             }
+            if (out.getTransportVersion().onOrAfter(TransportVersions.ML_TRAINED_MODEL_PREFIX_STRINGS_ADDED)) {
+                out.writeEnum(prefixType);
+            }
         }
 
         @Override
@@ -266,7 +284,8 @@ public boolean equals(Object o) {
                 && Objects.equals(inferenceTimeout, that.inferenceTimeout)
                 && Objects.equals(objectsToInfer, that.objectsToInfer)
                 && Objects.equals(textInput, that.textInput)
-                && (highPriority == that.highPriority);
+                && (highPriority == that.highPriority)
+                && (prefixType == that.prefixType);
         }
 
         @Override
@@ -276,7 +295,7 @@ public Task createTask(long id, String type, String action, TaskId parentTaskId,
 
         @Override
         public int hashCode() {
-            return Objects.hash(id, objectsToInfer, update, previouslyLicensed, inferenceTimeout, textInput, highPriority);
+            return Objects.hash(id, objectsToInfer, update, previouslyLicensed, inferenceTimeout, textInput, highPriority, prefixType);
         }
 
         public static class Builder {

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

@@ -26,6 +26,7 @@
 import org.elasticsearch.xcontent.ToXContentObject;
 import org.elasticsearch.xcontent.XContentBuilder;
 import org.elasticsearch.xcontent.XContentParser;
+import org.elasticsearch.xpack.core.ml.inference.TrainedModelPrefixStrings;
 import org.elasticsearch.xpack.core.ml.inference.trainedmodel.EmptyConfigUpdate;
 import org.elasticsearch.xpack.core.ml.inference.trainedmodel.InferenceConfigUpdate;
 import org.elasticsearch.xpack.core.ml.utils.ExceptionsHelper;
@@ -102,6 +103,7 @@ public static Request.Builder parseRequest(String id, XContentParser parser) {
         // and do know which field the model expects to find its
         // input and so cannot construct a document.
         private final List<String> textInput;
+        private TrainedModelPrefixStrings.PrefixType prefixType = TrainedModelPrefixStrings.PrefixType.NONE;
 
         public static Request forDocs(String id, InferenceConfigUpdate update, List<Map<String, Object>> docs, TimeValue inferenceTimeout) {
             return new Request(
@@ -156,6 +158,11 @@ public Request(StreamInput in) throws IOException {
             } else {
                 textInput = null;
             }
+            if (in.getTransportVersion().onOrAfter(TransportVersions.ML_TRAINED_MODEL_PREFIX_STRINGS_ADDED)) {
+                prefixType = in.readEnum(TrainedModelPrefixStrings.PrefixType.class);
+            } else {
+                prefixType = TrainedModelPrefixStrings.PrefixType.NONE;
+            }
         }
 
         public String getId() {
@@ -200,6 +207,14 @@ public boolean isHighPriority() {
             return highPriority;
         }
 
+        public void setPrefixType(TrainedModelPrefixStrings.PrefixType prefixType) {
+            this.prefixType = prefixType;
+        }
+
+        public TrainedModelPrefixStrings.PrefixType getPrefixType() {
+            return prefixType;
+        }
+
         @Override
         public ActionRequestValidationException validate() {
             ActionRequestValidationException validationException = super.validate();
@@ -226,6 +241,9 @@ public void writeTo(StreamOutput out) throws IOException {
             if (out.getTransportVersion().onOrAfter(TransportVersions.V_8_7_0)) {
                 out.writeOptionalStringCollection(textInput);
             }
+            if (out.getTransportVersion().onOrAfter(TransportVersions.ML_TRAINED_MODEL_PREFIX_STRINGS_ADDED)) {
+                out.writeEnum(prefixType);
+            }
         }
 
         @Override
@@ -243,12 +261,13 @@ public boolean equals(Object o) {
                 && Objects.equals(update, that.update)
                 && Objects.equals(inferenceTimeout, that.inferenceTimeout)
                 && Objects.equals(highPriority, that.highPriority)
-                && Objects.equals(textInput, that.textInput);
+                && Objects.equals(textInput, that.textInput)
+                && (prefixType == that.prefixType);
         }
 
         @Override
         public int hashCode() {
-            return Objects.hash(id, update, docs, inferenceTimeout, highPriority, textInput);
+            return Objects.hash(id, update, docs, inferenceTimeout, highPriority, textInput, prefixType);
         }
 
         @Override