fix(JinaV5OmniWrapper): use Query/Document prefix for all text tasks

The text-matching, classification and clustering adapters were trained
WITH the "Query: "/"Document: " prefix, not without it. Running them
without any prefix causes dramatic score drops (e.g. STS12: 0.85→0.38,
Banking77: 0.90→0.30, SprintDuplicateQuestions: 0.96→0.18).

New logic: audio/video non-retrieval tasks skip the prefix (unchanged);
text and image tasks always use the prefix regardless of adapter type.

Update unit tests to reflect the corrected expected behavior.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: florian-hoenicke

mteb/models/model_implementations/jina_models.py

@@ -842,15 +842,16 @@ def encode(
                 task_metadata.simplified_task_type
             )
         task = jina_task_name or "retrieval"
-        # Non-retrieval adapters were trained without the Query/Document prefix.
-        if task == "retrieval":
+        # All text and image adapters use Query/Document prefix.
+        # Audio/video non-retrieval adapters do not use the prefix.
+        if (has_video or has_audio) and task != "retrieval":
+            prompt = ""
+        else:
             prompt = (
                 "Query: "
                 if prompt_type and prompt_type == PromptType.query
                 else "Document: "
             )
-        else:
-            prompt = ""
 
         logger.info(
             f"Using prompt=`{prompt}` and task={task} for task={task_metadata.name} prompt_type={prompt_type}"

tests/test_models/test_jina_v5_omni_wrapper.py

@@ -72,10 +72,11 @@ def _omni_prompts() -> dict:
     [
         (PromptType.query, "Retrieval", "retrieval", "Query: "),
         (PromptType.document, "Retrieval", "retrieval", "Document: "),
-        (PromptType.document, "Clustering", "clustering", ""),
-        (PromptType.query, "STS", "text-matching", ""),
-        (PromptType.document, "Classification", "classification", ""),
-        (PromptType.query, "PairClassification", "text-matching", ""),
+        # Text tasks: all adapters now use prefix (trained with prefix)
+        (PromptType.document, "Clustering", "clustering", "Document: "),
+        (PromptType.query, "STS", "text-matching", "Query: "),
+        (PromptType.document, "Classification", "classification", "Document: "),
+        (PromptType.query, "PairClassification", "text-matching", "Query: "),
     ],
 )
 def test_prefix_by_task_type(prompt_type, task_type, expected_task, expected_prompt):
@@ -89,9 +90,10 @@ def test_prefix_by_task_type(prompt_type, task_type, expected_task, expected_pro
     "prompt_type,task_type,expected_task,expected_prompt",
     [
         (PromptType.query, "Any2AnyRetrieval", "retrieval", "Query: "),
-        (PromptType.document, "ImageClustering", "clustering", ""),
-        (PromptType.query, "VisualSTS(eng)", "text-matching", ""),
-        (PromptType.query, "AudioPairClassification", "text-matching", ""),
+        # Text-input paths always get prefix regardless of task type label
+        (PromptType.document, "ImageClustering", "clustering", "Document: "),
+        (PromptType.query, "VisualSTS(eng)", "text-matching", "Query: "),
+        (PromptType.query, "AudioPairClassification", "text-matching", "Query: "),
     ],
 )
 def test_simplified_fallback(prompt_type, task_type, expected_task, expected_prompt):
@@ -108,9 +110,10 @@ def test_simplified_fallback(prompt_type, task_type, expected_task, expected_pro
         (PromptType.query, "ImageClassification", "retrieval", "Query: "),
         (PromptType.query, "AudioClassification", "retrieval", "Query: "),
         (PromptType.query, "ZeroShotClassification", "retrieval", "Query: "),
-        (PromptType.document, "Compositionality", "clustering", ""),
+        # Text-input paths always get prefix
+        (PromptType.document, "Compositionality", "clustering", "Document: "),
         (PromptType.query, "AudioPairClassification", "retrieval", "Query: "),
-        (PromptType.document, "ImageClustering", "text-matching", ""),
+        (PromptType.document, "ImageClustering", "text-matching", "Document: "),
     ],
 )
 def test_omni_overrides(prompt_type, task_type, expected_task, expected_prompt):