Adds 'late_chunking' parameter to the JinaAI inference endpoint

specification/inference/_types/CommonTypes.ts

@@ -1711,6 +1711,15 @@ export class JinaAITaskSettings {
    * * `search`: Use it for storing embeddings of search queries run against a vector database to find relevant documents.
    */
   task?: JinaAITextEmbeddingTask
+  /**
+   * For a `text_embedding` task, controls when text is split into chunks.
+   * When set to `true`, a request from Elasticsearch contains only chunks related to a single document. Instead of batching chunks across documents, Elasticsearch sends them in separate requests. This ensures that chunk embeddings retain context from the entire document, improving semantic quality.
+   *
+   * If a document exceeds the model's context limits, late chunking is automatically disabled for that document only and standard chunking is used instead.
+   *
+   * If not specified, defaults to `false`.
+   */
+  late_chunking?: boolean
   /**
    * For a `rerank` task, the number of most relevant documents to return.
    * It defaults to the number of the documents.