Merge pull request #4675 from vespa-engine/bjorncs/openai-embedder-prepend-docs

docs: document <prepend> config for OpenAI embedder

Author: bjorncs

en/rag/embedding.html

@@ -732,6 +732,11 @@ <h3 id="adding-a-fixed-string-to-a-query-text">Adding a fixed string to a query
   Find a complete example in the <a href="https://github.com/vespa-engine/sample-apps/tree/master/colbert">ColBERT</a>
   sample application.
 </p>
+<p>
+  The <code>&lt;prepend&gt;</code> element is also supported by the
+  <a href="../reference/rag/embedding.html#openai-embedder-reference-config">OpenAI embedder</a>,
+  which is useful for OpenAI-compatible instruction-tuned models that expect a task-specific prefix.
+</p>
 <p>
   An alternative approach is using query profiles to prepend query data.
   If you need to add a standard wrapper or a prefix instruction around the input text you want to embed

en/reference/rag/embedding.html

@@ -661,6 +661,24 @@ <h3 id="openai-embedder-reference-config">OpenAI embedder reference config</h3>
       <td>element</td>
       <td>disabled</td>
     </tr>
+    <tr>
+      <td>prepend</td>
+      <td>Optional</td>
+      <td>Strings prepended to the text input before sending the embedding request. Useful for
+        OpenAI-compatible instruction-tuned models that expect a task-specific prefix.
+        <ul>
+          <li>Element &lt;query&gt; - Optional query prepend instruction.</li>
+          <li>Element &lt;document&gt; - Optional document prepend instruction.</li>
+        </ul>
+    <pre>{% highlight xml %}
+    <prepend>
+      <query>query: </query>
+      <document>passage: </document>
+    </prepend>{% endhighlight %}</pre>
+      </td>
+      <td>Optional &lt;query&gt; &lt;document&gt; elements.</td>
+      <td></td>
+    </tr>
   </tbody>
 </table>