fix tests to be compatible with SentenceTransformers v5

mteb/models/jina_models.py

@@ -10,10 +10,10 @@
 from sentence_transformers import __version__ as st_version
 
 from mteb.encoder_interface import PromptType
+from mteb.languages import PROGRAMMING_LANGS
 from mteb.model_meta import ModelMeta
 from mteb.models.sentence_transformer_wrapper import SentenceTransformerWrapper
 from mteb.requires_package import requires_package
-from mteb.languages import PROGRAMMING_LANGS
 
 logger = logging.getLogger(__name__)
 
@@ -234,8 +234,8 @@ def __init__(
         )
         requires_package(self, "peft", model, "pip install 'mteb[jina-v4]'")
         requires_package(self, "torchvision", model, "pip install 'mteb[jina-v4]'")
-        import peft  # noqa: F401
         import flash_attn  # noqa: F401
+        import peft  # noqa: F401
         import transformers  # noqa: F401
 
         super().__init__(model, revision, model_prompts, **kwargs)
@@ -284,8 +284,7 @@ def encode(
 def get_programming_task_override(
     task_name: str, current_task_name: str | None
 ) -> str | None:
-    """
-    Check if task involves programming content and override with 'code' task if so.
+    """Check if task involves programming content and override with 'code' task if so.
 
     Args:
         task_name: Original task name to check

tests/test_benchmark/mock_models.py

@@ -133,6 +133,8 @@ class MockSentenceTransformer(SentenceTransformer):
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
+        # by default, in SentenceTransformer, prompts are `{"query": "", "document": ""}`
+        self.prompts = {}
 
     def encode(
         self,

tests/test_benchmark/test_benchmark.py

@@ -9,7 +9,6 @@
 import numpy as np
 import pytest
 import torch
-from sentence_transformers import SentenceTransformer
 
 import mteb
 import mteb.overview
@@ -114,7 +113,7 @@ def encode(self, sentences, prompt_name: str | None = None, **kwargs):
             assert prompt_name == _task_name
             return np.zeros((len(sentences), 10))
 
-    class EncoderWithoutInstructions(SentenceTransformer):
+    class EncoderWithoutInstructions(MockSentenceTransformer):
         def encode(self, sentences, **kwargs):
             assert kwargs["prompt_name"] is None
             return super().encode(sentences, **kwargs)
@@ -138,7 +137,7 @@ def encode(self, sentences, **kwargs):
         overwrite_results=True,
     )
     # Test that the task_name is not passed down to the encoder
-    model = EncoderWithoutInstructions("average_word_embeddings_levy_dependency")
+    model = EncoderWithoutInstructions()
     assert model.prompts == {}, "The encoder should not have any prompts"
     eval.run(model, output_folder=tmp_path.as_posix(), overwrite_results=True)