Increase values

Author: davidkyle

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

@@ -14,7 +14,7 @@ public interface RerankingInferenceService {
     /**
      * The default window size for small reranking models (512 input tokens).
      */
-    int CONSERVATIVE_DEFAULT_WINDOW_SIZE = 250;
+    int CONSERVATIVE_DEFAULT_WINDOW_SIZE = 300;
 
     /**
      * The reranking model's max window or an approximation of

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

@@ -396,7 +396,7 @@ public int rerankerWindowSize(String modelId) {
         // Alibaba's mGTE models support long context windows of up to 8192 tokens.
         // Using 1 token = 0.75 words, this translates to approximately 6144 words.
         // https://huggingface.co/Alibaba-NLP/gte-multilingual-reranker-base
-        return 5000;
+        return 5500;
     }
 
     public static class Configuration {

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

@@ -367,7 +367,7 @@ public int rerankerWindowSize(String modelId) {
         // Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank
         // Using 1 token = 0.75 words as a rough estimate, we get 3072 words
         // allowing for some headroom, we set the window size below 3072
-        return 2500;
+        return 2800;
     }
 
     public static class Configuration {

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

@@ -1064,7 +1064,7 @@ static TaskType inferenceConfigToTaskType(InferenceConfig config) {
     @Override
     public int rerankerWindowSize(String modelId) {
         // The Elastic reranker has a window size of 512 tokens.
-        // Return 250 words as a default that comfortably fits in the window.
+        // Return 300 words as a default that comfortably fits in the window.
         // TODO custom rerank models may have larger windows, make this configurable
         return RerankingInferenceService.CONSERVATIVE_DEFAULT_WINDOW_SIZE;
     }

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

@@ -392,7 +392,7 @@ public int rerankerWindowSize(String modelId) {
         // TODO make the rerank window size configurable
 
         if (modelId != null && modelId.endsWith("-004")) {
-            return 500;
+            return 600;
         } else {
             return RerankingInferenceService.CONSERVATIVE_DEFAULT_WINDOW_SIZE;
         }

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

@@ -353,7 +353,7 @@ public int rerankerWindowSize(String modelId) {
         // Jina AI rerank models have an 8000 token input length https://jina.ai/models/jina-reranker-v2-base-multilingual
         // Using 1 token = 0.75 words as a rough estimate, we get 6000 words
         // allowing for some headroom, we set the window size below 6000 words
-        return 5000;
+        return 5500;
     }
 
     public static class Configuration {

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

@@ -94,14 +94,14 @@ public class VoyageAIService extends SenderService implements RerankingInference
 
     private static final Map<String, Integer> RERANKERS_INPUT_SIZE = Map.of(
         "rerank-lite-1",
-        2500 // The smallest model has a 4K context length https://docs.voyageai.com/docs/reranker
+        2800 // The smallest model has a 4K context length https://docs.voyageai.com/docs/reranker
     );
 
     /**
      * Apart from rerank-lite-1 all other models have a context length of at least 8k.
      * This value is based on 1 token == 0.75 words and allowing for some overhead
      */
-    private static final int DEFAULT_RERANKER_INPUT_SIZE_WORDS = 5000;
+    private static final int DEFAULT_RERANKER_INPUT_SIZE_WORDS = 5500;
 
     public static final EnumSet<InputType> VALID_INPUT_TYPE_VALUES = EnumSet.of(
         InputType.INGEST,

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/alibabacloudsearch/AlibabaCloudSearchServiceTests.java

@@ -724,6 +724,6 @@ public InferenceService createInferenceService() {
 
     @Override
     protected void assertRerankerWindowSize(RerankingInferenceService rerankingInferenceService) {
-        assertThat(rerankingInferenceService.rerankerWindowSize("any model"), is(5000));
+        assertThat(rerankingInferenceService.rerankerWindowSize("any model"), is(5500));
     }
 }

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/azureaistudio/AzureAiStudioServiceTests.java

@@ -1691,7 +1691,7 @@ public InferenceService createInferenceService() {
 
     @Override
     protected void assertRerankerWindowSize(RerankingInferenceService rerankingInferenceService) {
-        assertThat(rerankingInferenceService.rerankerWindowSize("Any model"), is(250));
+        assertThat(rerankingInferenceService.rerankerWindowSize("Any model"), is(300));
     }
 
     private Map<String, Object> getRequestConfigMap(

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

@@ -1644,6 +1644,6 @@ public InferenceService createInferenceService() {
 
     @Override
     protected void assertRerankerWindowSize(RerankingInferenceService rerankingInferenceService) {
-        assertThat(rerankingInferenceService.rerankerWindowSize("any model"), is(2500));
+        assertThat(rerankingInferenceService.rerankerWindowSize("any model"), is(2800));
     }
 }