Set default similarity for Cohere model to cosine (elastic#125370)

Cohere embeddings are expected to be normalized to unit vectors, but due to floating point precision issues,
our check ({@link DenseVectorFieldMapper#isNotUnitVector(float)}) often fails.
This change fixes this bug by setting the default similarity for newly created Cohere inference endpoint to cosine.

Closes elastic#122878

Author: jimczi

docs/changelog/125370.yaml

@@ -0,0 +1,6 @@
+pr: 125370
+summary: Set default similarity for Cohere model to cosine
+area: Machine Learning
+type: bug
+issues:
+ - 122878

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

@@ -14,6 +14,7 @@
 import org.elasticsearch.common.util.LazyInitializable;
 import org.elasticsearch.core.Nullable;
 import org.elasticsearch.core.TimeValue;
+import org.elasticsearch.index.mapper.vectors.DenseVectorFieldMapper;
 import org.elasticsearch.inference.ChunkedInference;
 import org.elasticsearch.inference.ChunkingSettings;
 import org.elasticsearch.inference.InferenceServiceConfiguration;
@@ -335,15 +336,15 @@ public Model updateModelWithEmbeddingDetails(Model model, int embeddingSize) {
     }
 
     /**
-     * Return the default similarity measure for the embedding type.
-     * Cohere embeddings are normalized to unit vectors therefor Dot
-     * Product similarity can be used and is the default for all Cohere
-     * models.
+     * Returns the default similarity measure for the embedding type.
+     * Cohere embeddings are expected to be normalized to unit vectors, but due to floating point precision issues,
+     * our check ({@link DenseVectorFieldMapper#isNotUnitVector(float)}) often fails.
+     * Therefore, we use cosine similarity to ensure compatibility.
      *
-     * @return The default similarity.
+     * @return The default similarity measure.
      */
     static SimilarityMeasure defaultSimilarity() {
-        return SimilarityMeasure.DOT_PRODUCT;
+        return SimilarityMeasure.COSINE;
     }
 
     @Override

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/cohere/CohereServiceTests.java

@@ -1577,7 +1577,7 @@ public void testChunkedInfer_BatchesCalls_Bytes() throws IOException {
     }
 
     public void testDefaultSimilarity() {
-        assertEquals(SimilarityMeasure.DOT_PRODUCT, CohereService.defaultSimilarity());
+        assertEquals(SimilarityMeasure.COSINE, CohereService.defaultSimilarity());
     }
 
     public void testInfer_StreamRequest() throws Exception {