add prepend support

2 files changed, 8 insertions, 0 deletions

diff --git a/config-model/src/test/cfg/application/embed/services.xml b/config-model/src/test/cfg/application/embed/services.xml
index 089adbd7517..ae0d0952630 100644
--- a/config-model/src/test/cfg/application/embed/services.xml
+++ b/config-model/src/test/cfg/application/embed/services.xml
@@ -12,6 +12,10 @@
       <transformer-token-type-ids>my_token_type_ids</transformer-token-type-ids>
       <transformer-output>my_output</transformer-output>
       <normalize>true</normalize>
+      <prepend>
+        <query>Represent this sentence for searching relevant passages:</query>
+        <document>passage:</document>
+      </prepend>
       <onnx-execution-mode>parallel</onnx-execution-mode>
       <onnx-intraop-threads>10</onnx-intraop-threads>
       <onnx-interop-threads>8</onnx-interop-threads>
diff --git a/config-model/src/test/java/com/yahoo/vespa/model/container/xml/EmbedderTestCase.java b/config-model/src/test/java/com/yahoo/vespa/model/container/xml/EmbedderTestCase.java
index fb1e176f707..6f629d99c92 100644
--- a/config-model/src/test/java/com/yahoo/vespa/model/container/xml/EmbedderTestCase.java
+++ b/config-model/src/test/java/com/yahoo/vespa/model/container/xml/EmbedderTestCase.java
@@ -88,6 +88,8 @@ public class EmbedderTestCase {
         var tokenizerCfg = assertHuggingfaceTokenizerComponentPresent(cluster);
         assertEquals("https://my/url/tokenizer.json", modelReference(tokenizerCfg.model().get(0), "path").url().orElseThrow().value());
         assertEquals(-1, tokenizerCfg.maxLength());
+        assertEquals("Represent this sentence for searching relevant passages:", embedderCfg.prependQuery());
+        assertEquals("passage:", embedderCfg.prependDocument());
     }
 
     @Test
@@ -101,6 +103,8 @@ public class EmbedderTestCase {
         var tokenizerCfg = assertHuggingfaceTokenizerComponentPresent(cluster);
         assertEquals("https://data.vespa.oath.cloud/onnx_models/multilingual-e5-base/tokenizer.json", modelReference(tokenizerCfg.model().get(0), "path").url().orElseThrow().value());
         assertEquals(-1, tokenizerCfg.maxLength());
+        assertEquals("Represent this sentence for searching relevant passages:", embedderCfg.prependQuery());
+        assertEquals("passage:", embedderCfg.prependDocument());
     }
 
     @Test