mteb 2.7.13__py3-none-any.whl → 2.7.15__py3-none-any.whl

mteb/abstasks/pair_classification.py

@@ -25,6 +25,8 @@ from mteb.types.statistics import (
 if TYPE_CHECKING:
     from pathlib import Path
 
+    from numpy.typing import NDArray
+
     from mteb._evaluators.pair_classification_evaluator import (
         PairClassificationDistances,
     )
@@ -36,7 +38,6 @@ if TYPE_CHECKING:
         TextStatistics,
     )
 
-
 logger = logging.getLogger(__name__)
 
 
@@ -138,7 +139,7 @@ class AbsTaskPairClassification(AbsTask):
         self, similarity_scores: PairClassificationDistances, labels: list[int]
     ) -> dict[str, float]:
         logger.info("Computing metrics...")
-        np_labels = np.asarray(labels)
+        np_labels: NDArray[np.int64] = np.asarray(labels, dtype=np.int64)
         output_scores = {}
         max_scores = defaultdict(list)
         for short_name, scores, reverse in [
@@ -281,7 +282,10 @@ class AbsTaskPairClassification(AbsTask):
         )
 
     def _compute_metrics_values(
-        self, scores: list[float], labels: np.ndarray, high_score_more_similar: bool
+        self,
+        scores: list[float],
+        labels: NDArray[np.int64],
+        high_score_more_similar: bool,
     ) -> dict[str, float]:
         """Compute the metrics for the given scores and labels.
 
@@ -315,7 +319,10 @@ class AbsTaskPairClassification(AbsTask):
         )
 
     def _find_best_acc_and_threshold(
-        self, scores: list[float], labels: np.ndarray, high_score_more_similar: bool
+        self,
+        scores: list[float],
+        labels: NDArray[np.int64],
+        high_score_more_similar: bool,
     ) -> tuple[float, float]:
         rows = list(zip(scores, labels))
         rows = sorted(rows, key=lambda x: x[0], reverse=high_score_more_similar)
@@ -323,7 +330,7 @@ class AbsTaskPairClassification(AbsTask):
         max_acc = 0
         best_threshold = -1.0
         positive_so_far = 0
-        remaining_negatives = sum(np.array(labels) == 0)
+        remaining_negatives = sum(labels == 0)
 
         for i in range(len(rows) - 1):
             score, label = rows[i]
@@ -339,10 +346,9 @@ class AbsTaskPairClassification(AbsTask):
         return max_acc, best_threshold
 
     def _find_best_f1_and_threshold(
-        self, scores, labels, high_score_more_similar: bool
+        self, scores, labels: NDArray[np.int64], high_score_more_similar: bool
     ) -> tuple[float, float, float, float]:
         scores = np.asarray(scores)
-        labels = np.asarray(labels)
 
         rows = list(zip(scores, labels))
 

mteb/models/get_model_meta.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import difflib
 import logging
+import warnings
 from typing import TYPE_CHECKING, Any
 
 from mteb.models import (
@@ -122,6 +123,11 @@ def get_model(
     return model
 
 
+_MODEL_RENAMES: dict[str, str] = {
+    "bm25s": "baseline/bm25s",
+}
+
+
 def get_model_meta(
     model_name: str,
     revision: str | None = None,
@@ -134,11 +140,17 @@ def get_model_meta(
         model_name: Name of the model to fetch
         revision: Revision of the model to fetch
         fetch_from_hf: Whether to fetch the model from HuggingFace Hub if not found in the registry
-        fill_missing: Computes missing attributes from the metadata including number of parameters and memory usage.
+        fill_missing: Fill missing attributes from the metadata including number of parameters and memory usage.
 
     Returns:
         A model metadata object
     """
+    if model_name in _MODEL_RENAMES:
+        new_name = _MODEL_RENAMES[model_name]
+        msg = f"The model '{model_name}' has been renamed to '{new_name}'. To prevent this warning use the new name."
+        warnings.warn(msg, DeprecationWarning, stacklevel=2)
+        model_name = new_name
+
     if model_name in MODEL_REGISTRY:
         model_meta = MODEL_REGISTRY[model_name]
 
@@ -149,7 +161,7 @@ def get_model_meta(
 
         if fill_missing and fetch_from_hf:
             original_meta_dict = model_meta.model_dump()
-            new_meta = ModelMeta.from_hub(model_name)
+            new_meta = ModelMeta.from_hub(model_name, fill_missing=fill_missing)
             new_meta_dict = new_meta.model_dump(exclude_none=True)
 
             updates = {

mteb/models/model_implementations/bm25.py

@@ -129,7 +129,7 @@ def bm25_loader(model_name, **kwargs) -> SearchProtocol:
 
 bm25_s = ModelMeta(
     loader=bm25_loader,
-    name="bm25s",
+    name="baseline/bm25s",
     model_type=["dense"],
     languages=["eng-Latn"],
     open_weights=True,

mteb/models/model_implementations/human.py

@@ -2,7 +2,7 @@ from mteb.models import ModelMeta
 
 human = ModelMeta(
     loader=None,
-    name="Human",
+    name="baseline/Human",
     model_type=["dense"],
     languages=["eng-Latn", "ara-Arab", "rus-Cyrl", "dan-Latn", "nob-Latn"],
     open_weights=True,

mteb/models/model_implementations/opensearch_neural_sparse_models.py

@@ -134,7 +134,7 @@ class SparseEncoderWrapper(AbsEncoder):
 
 opensearch_neural_sparse_encoding_doc_v3_gte = ModelMeta(
     name="opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte",
-    model_type=["dense"],
+    model_type=["sparse"],
     languages=["eng-Latn"],
     open_weights=True,
     revision="a8abaa916125ee512a7a8f4d706d07eb0128a8e6",
@@ -161,7 +161,7 @@ opensearch_neural_sparse_encoding_doc_v3_gte = ModelMeta(
 
 opensearch_neural_sparse_encoding_doc_v3_distill = ModelMeta(
     name="opensearch-project/opensearch-neural-sparse-encoding-doc-v3-distill",
-    model_type=["dense"],
+    model_type=["sparse"],
     languages=["eng-Latn"],
     open_weights=True,
     revision="babf71f3c48695e2e53a978208e8aba48335e3c0",
@@ -184,7 +184,7 @@ opensearch_neural_sparse_encoding_doc_v3_distill = ModelMeta(
 
 opensearch_neural_sparse_encoding_doc_v2_distill = ModelMeta(
     name="opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill",
-    model_type=["dense"],
+    model_type=["sparse"],
     languages=["eng-Latn"],
     open_weights=True,
     revision="8921a26c78b8559d6604eb1f5c0b74c079bee38f",
@@ -208,7 +208,7 @@ opensearch_neural_sparse_encoding_doc_v2_distill = ModelMeta(
 
 opensearch_neural_sparse_encoding_doc_v2_mini = ModelMeta(
     name="opensearch-project/opensearch-neural-sparse-encoding-doc-v2-mini",
-    model_type=["dense"],
+    model_type=["sparse"],
     languages=["eng-Latn"],
     open_weights=True,
     revision="4af867a426867dfdd744097531046f4289a32fdd",
@@ -231,7 +231,7 @@ opensearch_neural_sparse_encoding_doc_v2_mini = ModelMeta(
 
 opensearch_neural_sparse_encoding_doc_v1 = ModelMeta(
     name="opensearch-project/opensearch-neural-sparse-encoding-doc-v1",
-    model_type=["dense"],
+    model_type=["sparse"],
     languages=["eng-Latn"],
     open_weights=True,
     revision="98cdcbd72867c547f72f2b7b7bed9cdf9f09922d",

mteb/models/model_meta.py

@@ -71,7 +71,7 @@ FRAMEWORKS = Literal[
     "Transformers",
 ]
 
-MODEL_TYPES = Literal["dense", "cross-encoder", "late-interaction"]
+MODEL_TYPES = Literal["dense", "cross-encoder", "late-interaction", "sparse"]
 
 
 class ScoringFunction(HelpfulStrEnum):
@@ -266,7 +266,7 @@ class ModelMeta(BaseModel):
     @field_validator("name")
     @classmethod
     def _check_name(cls, v: str | None) -> str | None:
-        if v is None or v in ("bm25s", "Human"):
+        if v is None:
             return v
         if "/" not in v:
             raise ValueError(
@@ -302,26 +302,154 @@ class ModelMeta(BaseModel):
             raise ValueError("Model name is not set")
         return self.name.replace("/", "__").replace(" ", "_")
 
+    @classmethod
+    def _detect_cross_encoder_or_dense(
+        cls,
+        model_name: str,
+        revision: str | None,
+        sentence_transformers_loader: Callable[..., MTEBModels],
+        cross_encoder_loader: Callable[..., MTEBModels],
+    ) -> tuple[Callable[..., MTEBModels] | None, MODEL_TYPES]:
+        """Detect if model is CrossEncoder or default to dense."""
+        config = _get_json_from_hub(
+            model_name, "config.json", "model", revision=revision
+        )
+
+        if not config:
+            logger.warning(
+                f"Could not load config.json for {model_name}. "
+                "Defaulting to SentenceTransformer loader."
+            )
+            return sentence_transformers_loader, "dense"
+
+        architectures = config.get("architectures", [])
+
+        is_cross_encoder = any(
+            arch.endswith("ForSequenceClassification") for arch in architectures
+        )
+        if is_cross_encoder:
+            return cross_encoder_loader, "cross-encoder"
+
+        if cls._is_causal_lm_reranker(architectures, config, model_name):
+            return cross_encoder_loader, "cross-encoder"
+
+        logger.info(
+            f"Model {model_name} does not have modules.json or recognized architecture. "
+            "Defaulting to SentenceTransformer loader."
+        )
+        return sentence_transformers_loader, "dense"
+
+    @staticmethod
+    def _is_causal_lm_reranker(
+        architectures: list[str], config: dict[str, Any], model_name: str
+    ) -> bool:
+        """Check if model is a CausalLM-style reranker."""
+        is_causal_lm = any(arch.endswith("ForCausalLM") for arch in architectures)
+
+        if not is_causal_lm:
+            return False
+
+        num_labels = config.get("num_labels", 0)
+        model_name_lower = model_name.lower()
+
+        return (
+            num_labels > 0
+            or "rerank" in model_name_lower
+            or "cross-encoder" in model_name_lower
+        )
+
+    @classmethod
+    def _detect_model_type_and_loader(
+        cls,
+        model_name: str | None,
+        revision: str | None = None,
+    ) -> tuple[Callable[..., MTEBModels] | None, MODEL_TYPES]:
+        """Detect the model type and appropriate loader based on HuggingFace Hub configuration files.
+
+        This follows the Sentence Transformers architecture detection logic:
+        1. Check for modules.json - If present, model is a SentenceTransformer (dense encoder)
+        2. If no modules.json, check config.json for architecture:
+            - ForSequenceClassification → CrossEncoder
+            - CausalLM with reranking indicators → CrossEncoder
+        3. Default to dense (SentenceTransformer) if no clear indicators are found
+
+        Detection for CausalLM-style rerankers:
+        - Model has ForCausalLM architecture AND
+        - Has num_labels > 0 in config, OR
+        - Model name contains "rerank" or "cross-encoder"
+
+        Args:
+            model_name: The HuggingFace model name (can be None)
+            revision: The model revision
+
+        Returns:
+            A tuple of (loader_function, model_type) where:
+            - loader_function: A callable that returns MTEBModels, or None if model doesn't exist
+            - model_type: One of "dense", "cross-encoder", or "late-interaction"
+        """
+        from mteb.models import CrossEncoderWrapper, sentence_transformers_loader
+
+        if not model_name or not _repo_exists(model_name):
+            return sentence_transformers_loader, "dense"
+
+        try:
+            modules_config = _get_json_from_hub(
+                model_name, "modules.json", "model", revision=revision
+            )
+
+            if (
+                modules_config
+            ):  # SentenceTransformer/SparseEncoder (Not support for now)
+                return sentence_transformers_loader, "dense"
+            else:
+                return cls._detect_cross_encoder_or_dense(
+                    model_name,
+                    revision,
+                    sentence_transformers_loader,
+                    cross_encoder_loader=CrossEncoderWrapper,
+                )
+
+        except Exception as e:
+            logger.warning(
+                f"Error detecting model type for {model_name}: {e}. "
+                "Defaulting to SentenceTransformer loader."
+            )
+
+        return sentence_transformers_loader, "dense"
+
     @classmethod
     def _from_hub(
         cls,
         model_name: str | None,
         revision: str | None = None,
-        compute_metadata: bool = True,
+        fill_missing: bool = True,
+        compute_metadata: bool | None = None,
     ) -> Self:
         """Generates a ModelMeta from a HuggingFace model name.
 
         Args:
             model_name: The HuggingFace model name.
             revision: Revision of the model
-            compute_metadata: Add metadata based on model card
+            fill_missing: Fill missing attributes from the metadata including number of parameters and memory usage.
+            compute_metadata: Deprecated. Use fill_missing instead.
 
         Returns:
             The generated ModelMeta.
         """
-        from mteb.models import sentence_transformers_loader
+        loader: Callable[..., MTEBModels] | None
+        model_type: MODEL_TYPES
+
+        if compute_metadata is not None:
+            warnings.warn(
+                "The compute_metadata parameter is deprecated and will be removed in a future version. "
+                f"Use fill_missing instead. Setting `fill_missing={compute_metadata}`.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+            fill_missing = compute_metadata
+
+        loader, model_type = cls._detect_model_type_and_loader(model_name, revision)
 
-        loader = sentence_transformers_loader
         frameworks: list[FRAMEWORKS] = ["PyTorch"]
         model_license = None
         reference = None
@@ -331,7 +459,7 @@ class ModelMeta(BaseModel):
         embedding_dim = None
         max_tokens = None
 
-        if model_name and compute_metadata and _repo_exists(model_name):
+        if model_name and fill_missing and _repo_exists(model_name):
             reference = "https://huggingface.co/" + model_name
             card = ModelCard.load(model_name)
             card_data: ModelCardData = card.data
@@ -363,6 +491,7 @@ class ModelMeta(BaseModel):
         return cls(
             loader=loader,
             name=model_name or "no_model_name/available",
+            model_type=[model_type],
             revision=revision or "no_revision_available",
             reference=reference,
             release_date=release_date,
@@ -387,14 +516,16 @@ class ModelMeta(BaseModel):
         cls,
         model: SentenceTransformer,
         revision: str | None = None,
-        compute_metadata: bool = True,
+        fill_missing: bool = False,
+        compute_metadata: bool | None = None,
     ) -> Self:
         """Generates a ModelMeta from a SentenceTransformer model.
 
         Args:
             model: SentenceTransformer model.
             revision: Revision of the model
-            compute_metadata: Add metadata based on model card
+            fill_missing: Fill missing attributes from the metadata including number of parameters and memory usage.
+            compute_metadata: Deprecated. Use fill_missing instead.
 
         Returns:
             The generated ModelMeta.
@@ -404,7 +535,9 @@ class ModelMeta(BaseModel):
             if model.model_card_data.model_name
             else model.model_card_data.base_model
         )
-        meta = cls._from_hub(name, revision, compute_metadata)
+        meta = cls._from_hub(
+            name, revision, fill_missing=fill_missing, compute_metadata=compute_metadata
+        )
         try:
             first = model[0]
 
@@ -417,7 +550,9 @@ class ModelMeta(BaseModel):
         meta.max_tokens = model.max_seq_length
         meta.embed_dim = model.get_sentence_embedding_dimension()
         meta.similarity_fn_name = ScoringFunction.from_str(model.similarity_fn_name)
-        meta.modalities = ["text"]
+        meta.modalities = ["text"]  # best guess
+        if "Sentence Transformers" not in meta.framework:
+            meta.framework.append("Sentence Transformers")
         return meta
 
     @classmethod
@@ -425,22 +560,29 @@ class ModelMeta(BaseModel):
         cls,
         model: str,
         revision: str | None = None,
-        compute_metadata: bool = True,
+        fill_missing: bool = True,
+        compute_metadata: bool | None = None,
     ) -> Self:
         """Generates a ModelMeta for model from HuggingFace hub.
 
         Args:
             model: Name of the model from HuggingFace hub. For example, `intfloat/multilingual-e5-large`
             revision: Revision of the model
-            compute_metadata: Add metadata based on model card
+            fill_missing: Fill missing attributes from the metadata including number of parameters and memory usage.
+            compute_metadata: Deprecated. Use fill_missing instead.
 
         Returns:
             The generated ModelMeta.
         """
-        meta = cls._from_hub(model, revision, compute_metadata)
+        meta = cls._from_hub(
+            model,
+            revision,
+            fill_missing=fill_missing,
+            compute_metadata=compute_metadata,
+        )
         meta.modalities = ["text"]
 
-        if model and compute_metadata and _repo_exists(model):
+        if model and fill_missing and _repo_exists(model):
             # have max_seq_length field
             sbert_config = _get_json_from_hub(
                 model, "sentence_bert_config.json", "model", revision=revision
@@ -469,21 +611,28 @@ class ModelMeta(BaseModel):
         cls,
         model: CrossEncoder,
         revision: str | None = None,
-        compute_metadata: bool = True,
+        fill_missing: bool = False,
+        compute_metadata: bool | None = None,
     ) -> Self:
         """Generates a ModelMeta from a CrossEncoder.
 
         Args:
             model: The CrossEncoder model
             revision: Revision of the model
-            compute_metadata: Add metadata based on model card
+            fill_missing: Fill missing attributes from the metadata including number of parameters and memory usage.
+            compute_metadata: Deprecated. Use fill_missing instead.
 
         Returns:
             The generated ModelMeta
         """
         from mteb.models import CrossEncoderWrapper
 
-        meta = cls._from_hub(model.model.name_or_path, revision, compute_metadata)
+        meta = cls._from_hub(
+            model.model.name_or_path,
+            revision,
+            fill_missing=fill_missing,
+            compute_metadata=compute_metadata,
+        )
         try:
             emb = model.model.get_input_embeddings()
 
@@ -498,6 +647,8 @@ class ModelMeta(BaseModel):
         meta.embed_dim = None
         meta.modalities = ["text"]
         meta.model_type = ["cross-encoder"]
+        if "Sentence Transformers" not in meta.framework:
+            meta.framework.append("Sentence Transformers")
         return meta
 
     def is_zero_shot_on(self, tasks: Sequence[AbsTask] | Sequence[str]) -> bool | None:

mteb/results/model_result.py

@@ -17,6 +17,7 @@ from .task_result import TaskError, TaskResult
 
 if TYPE_CHECKING:
     from collections.abc import Callable, Iterable
+    from pathlib import Path
 
     from mteb.abstasks.abstask import AbsTask
     from mteb.abstasks.task_metadata import (
@@ -417,3 +418,25 @@ class ModelResult(BaseModel):
         if not mods:
             mods = self.default_modalities
         return list(set(mods))
+
+    def to_disk(self, path: Path) -> None:
+        """Save ModelResult to disk as JSON.
+
+        Args:
+            path: The path to the file to save.
+        """
+        with path.open("w") as f:
+            f.write(self.model_dump_json(indent=2))
+
+    @classmethod
+    def from_disk(cls, path: Path) -> ModelResult:
+        """Load ModelResult from disk.
+
+        Args:
+            path: The path to the JSON file to load.
+
+        Returns:
+            The loaded ModelResult object.
+        """
+        with path.open("r", encoding="utf-8") as f:
+            return cls.model_validate_json(f.read())

mteb/results/task_result.py

@@ -337,16 +337,16 @@ class TaskResult(BaseModel):
             The loaded TaskResult object.
         """
         with path.open("r", encoding="utf-8") as f:
-            data = json.load(f)
+            json_str = f.read()
 
         if not load_historic_data:
             try:
-                return cls.model_validate(data)
+                return cls.model_validate_json(json_str)
             except Exception as e:
                 raise ValueError(
                     f"Error loading TaskResult from disk. You can try to load historic data by setting `load_historic_data=True`. Error: {e}"
                 )
-
+        data = json.loads(json_str)
         pre_1_11_load = (
             (
                 "mteb_version" in data
@@ -357,7 +357,7 @@ class TaskResult(BaseModel):
         )  # assume it is before 1.11.0 if the version is not present
 
         try:
-            obj: TaskResult = cls.model_validate(data)
+            obj: TaskResult = cls.model_validate_json(json_str)
         except Exception as e:
             if not pre_1_11_load:
                 raise e

mteb/tasks/pair_classification/fas/fars_tail.py

@@ -1,5 +1,3 @@
-import datasets
-
 from mteb.abstasks.pair_classification import AbsTaskPairClassification
 from mteb.abstasks.task_metadata import TaskMetadata
 
@@ -8,8 +6,8 @@ class FarsTail(AbsTaskPairClassification):
     metadata = TaskMetadata(
         name="FarsTail",
         dataset={
-            "path": "azarijafari/FarsTail",
-            "revision": "7335288588f14e5a687d97fc979194c2abe6f4e7",
+            "path": "mteb/FarsTail",
+            "revision": "0fa0863dc160869b5a2d78803b4440ea3c671ff5",
         },
         description="This dataset, named FarsTail, includes 10,367 samples which are provided in both the Persian language as well as the indexed format to be useful for non-Persian researchers. The samples are generated from 3,539 multiple-choice questions with the least amount of annotator interventions in a way similar to the SciTail dataset",
         reference="https://link.springer.com/article/10.1007/s00500-023-08959-3",
@@ -37,33 +35,3 @@ class FarsTail(AbsTaskPairClassification):
 }
 """,  # after removing neutral
     )
-
-    def load_data(self, num_proc: int = 1, **kwargs) -> None:
-        if self.data_loaded:
-            return
-        path = self.metadata.dataset["path"]
-        revision = self.metadata.dataset["revision"]
-        data_files = {
-            "test": f"https://huggingface.co/datasets/{path}/resolve/{revision}/data/Test-word.csv"
-        }
-        self.dataset = datasets.load_dataset(
-            "csv", data_files=data_files, delimiter="\t"
-        )
-        self.dataset_transform()
-        self.data_loaded = True
-
-    def dataset_transform(self, num_proc: int = 1):
-        _dataset = {}
-        self.dataset = self.dataset.filter(lambda x: x["label"] != "n")
-        self.dataset = self.dataset.map(
-            lambda example: {"label": 1 if example["label"] == "e" else 0}
-        )
-        for split in self.metadata.eval_splits:
-            _dataset[split] = [
-                {
-                    "sentence1": self.dataset[split]["premise"],
-                    "sentence2": self.dataset[split]["hypothesis"],
-                    "labels": self.dataset[split]["label"],
-                }
-            ]
-        self.dataset = _dataset

{mteb-2.7.13.dist-info → mteb-2.7.15.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mteb
-Version: 2.7.13
+Version: 2.7.15
 Summary: Massive Text Embedding Benchmark
 Author-email: MTEB Contributors <niklas@huggingface.co>, Kenneth Enevoldsen <kenneth.enevoldsen@cas.au.dk>, Nouamane Tazi <nouamane@huggingface.co>, Nils Reimers <info@nils-reimers.de>
 Maintainer-email: Kenneth Enevoldsen <kenneth.enevoldsen@cas.au.dk>, Roman Solomatin <risolomatin@gmail.com>, Isaac Chung <chungisaac1217@gmail.com>

{mteb-2.7.13.dist-info → mteb-2.7.15.dist-info}/RECORD

@@ -40,7 +40,7 @@ mteb/abstasks/clustering.py,sha256=I8vre2f2FJFagzJEYf6hKDo3Y28xU29J_O-MhfqWqSI,1
 mteb/abstasks/clustering_legacy.py,sha256=sbx8K6paccvzDPnmhgNE_UJE83orAJnQm3NGr-Ktjfs,9184
 mteb/abstasks/dataset_card_template.md,sha256=aD6l8qc3_jxwoIGJNYLzse-jpRa8hu92AxpnUtNgges,5122
 mteb/abstasks/multilabel_classification.py,sha256=rFa_Pw2OsUzqhZS-jh2zFD7I-TNl8bVNJ-DW7EpPapU,9708
-mteb/abstasks/pair_classification.py,sha256=MpjOeIcM_RMqSpkXc8PFDAwDgmb54gEqj6ZOUcOa1aQ,14122
+mteb/abstasks/pair_classification.py,sha256=RVV5WUjs18N5PbWpyxakDNEd1UlRc4ON9I0OjD26Z78,14231
 mteb/abstasks/regression.py,sha256=ZuMZfOwU3G4hr__eHsgdagKKdrbN4-wQMLz45jr9YUc,8946
 mteb/abstasks/retrieval.py,sha256=BPyRibStAD70JfR0Z1x-VVVfzJDRVSmbOS6uREfpmok,27743
 mteb/abstasks/retrieval_dataset_loaders.py,sha256=p0y1nrWlUrt_aeoR4ocDLEQMLuD_SlMH0gBiUsOwrww,9983
@@ -1479,9 +1479,9 @@ mteb/leaderboard/table.py,sha256=U5mWtrVUTk_6t8T4KAp5qlbFgKh1PD0iKICqNMfhsoY,104
 mteb/leaderboard/text_segments.py,sha256=iMIkS04QQjPbT-SkU0x6fOcS8xRbUYevryu9HydipKM,6570
 mteb/models/__init__.py,sha256=ABTuoqiBjBtBWW3LYY7ItBHdylR6jWoy06HH0g6j6fU,910
 mteb/models/abs_encoder.py,sha256=We9HlwWP61P4cMyZ080gywvDErA1eVsU9t46PtcNrCM,16830
-mteb/models/get_model_meta.py,sha256=h6Z2Q3D_L4BeJbi1gPHTSgelbQjsQn1ALpI8uwNa0ac,6832
+mteb/models/get_model_meta.py,sha256=EUy3F2A57qfm046lDJQ2eaAaD_buHALD7XHJeYsx8vY,7234
 mteb/models/instruct_wrapper.py,sha256=XAvvbPnXiTxKhFbmusm2uS8E9BMq8QXRSzQQI1jqKzE,9781
-mteb/models/model_meta.py,sha256=5seQwce893SbgSywFsphLuZGQ9rRn1fLl9fj1SfKnXE,32887
+mteb/models/model_meta.py,sha256=B2BpzOZAxBgNwMrZ-saN0sw-Y4tjBKIbt2byDoQz3ac,38873
 mteb/models/models_protocols.py,sha256=HTB4-SYa3SeJXMMSA8o05lHTiLBbq314VW60K_PfcZY,9509
 mteb/models/search_wrappers.py,sha256=PXE1VVDWUd0LgTPJ-FxqIbGpIDWLRKo5CjrwIuu5nzw,21567
 mteb/models/sentence_transformer_wrapper.py,sha256=RsOxj-b7qzeYcxUTVJyb-lZDY4bINl4jEAEkPvKYB10,13578
@@ -1505,7 +1505,7 @@ mteb/models/model_implementations/bge_models.py,sha256=JuO1FRWrsqlsM_jslQ96oVsD3
 mteb/models/model_implementations/bica_model.py,sha256=Yx3iZrXF6ZMJS9SH5lbzNHoUWGNH3dypRtZ7dX5o7rA,1305
 mteb/models/model_implementations/blip2_models.py,sha256=C6egwozJthHmv92I0SWID3-sQCPROPJP0TzfQVKNzlo,7898
 mteb/models/model_implementations/blip_models.py,sha256=D_9e7C8GXGST8k7dMJL20x984vMeqbITu36XASi-iUU,12149
-mteb/models/model_implementations/bm25.py,sha256=-xpeMqYi0_Ne5rL_Y1qAlYvloLlQzNDUav2ESVGHAhA,5217
+mteb/models/model_implementations/bm25.py,sha256=IAKU8syYesN7seRQLII-c1ACq6BRz5Ql6nEQEXYWLwQ,5226
 mteb/models/model_implementations/bmretriever_models.py,sha256=rijCIzX6nO5kNXqxEFbZrV7bsZtmKs8RIkMqa5cPWTk,7078
 mteb/models/model_implementations/cadet_models.py,sha256=gXIfW9MkGYFhOhsrq5a_tQcPuth13Dh1dO1KySwVxyo,2305
 mteb/models/model_implementations/cde_models.py,sha256=l4E6h1hcsNY1GTXoCgQDoeG5dRcEl7JTOiiWmp6FYqg,9373
@@ -1537,7 +1537,7 @@ mteb/models/model_implementations/granite_vision_embedding_models.py,sha256=jxyR
 mteb/models/model_implementations/gritlm_models.py,sha256=756vgZGADy5FhKlFuzuD6huevC_AYD5b88V1Y5yFht8,3241
 mteb/models/model_implementations/gte_models.py,sha256=-ASkoAuAiVytVtsYMtuKonUf39i0U69HSEnJy_-PwXA,14574
 mteb/models/model_implementations/hinvec_models.py,sha256=SYWGFr8XALmM7B9tIHEQnrqq9kZOZIBkW7m7QpzerHI,1756
-mteb/models/model_implementations/human.py,sha256=MTw560BrwuvUsHUb5_Gjq7ZRfrBmaT8dGnrubFFDB_o,591
+mteb/models/model_implementations/human.py,sha256=k7vN6WTcSWyWS9wnluzr6yCOjuMi5LupQnT-4cfzNOk,600
 mteb/models/model_implementations/ibm_granite_models.py,sha256=ipLRDBerTQiL5NaoaDho410Fzy7eNFlF3jB54hGZrwI,8687
 mteb/models/model_implementations/inf_models.py,sha256=q_hNNhzMjAxbnJnAT0N6KaNegX_3XZlmz-LXY5C891I,3093
 mteb/models/model_implementations/jasper_models.py,sha256=ourAMx1_L6b2AxX046wQcxDqvYzY1Mx3gaHww0WaMA8,16476
@@ -1572,7 +1572,7 @@ mteb/models/model_implementations/nvidia_models.py,sha256=r-AW1dVQbteWjexjvZgFEt
 mteb/models/model_implementations/octen_models.py,sha256=5z-t2O-iIFiOOLdZ_AK9f7GrVRg-9_vx3JNAG9dJNPE,8562
 mteb/models/model_implementations/openai_models.py,sha256=y1wMknrrcu1L5CNwniG0mFThPVMON1c2Fj22jkKsw7Y,9730
 mteb/models/model_implementations/openclip_models.py,sha256=z2gQum16O0QhJPyxqKor3oO-_uWfnep6wSXqOFQQ2Q8,11969
-mteb/models/model_implementations/opensearch_neural_sparse_models.py,sha256=O6ewgGKZtw-kpspbz8-dXArDf5tUWy_vY6AxFs7fsZ8,8936
+mteb/models/model_implementations/opensearch_neural_sparse_models.py,sha256=J5FEvKWQUiBusL6PHcrRuRRJOQ-iMwOSu1fX0pblXhk,8941
 mteb/models/model_implementations/ops_colqwen3_models.py,sha256=5vg5d1_WfVGMgtIwkh6zf2-Paum6V35XcKEvLfRyRzs,7437
 mteb/models/model_implementations/ops_moa_models.py,sha256=Ah7L78mqC9pH8t6sf1OWXOLjouVUpAutt6lZ0np7eMM,2655
 mteb/models/model_implementations/ordalietech_solon_embeddings_mini_beta_1_1.py,sha256=xv1ftJeMND4lpeKYC3RLQB4nhdiYy0wCxrzEjUj4gSg,1114
@@ -1629,8 +1629,8 @@ mteb/models/search_encoder_index/search_indexes/__init__.py,sha256=Wm60_oUemUpFs
 mteb/models/search_encoder_index/search_indexes/faiss_search_index.py,sha256=jwC-3swhnILZnVHUrMR7Ts78TuYtVRxPusF02UV1g6E,5770
 mteb/results/__init__.py,sha256=EXQqK4Am5eIYzD52dpcGAFSdqnC38oE6JHN302oidHc,158
 mteb/results/benchmark_results.py,sha256=unBUBJ92ud0UXlkZJLn71WVcf-oUlF6XcITTccz5OBA,20318
-mteb/results/model_result.py,sha256=u1eUynaf_KVsdYdz29MACsCDR_48xODapkIGLG-lZa0,15321
-mteb/results/task_result.py,sha256=TL9V7P7cXqLljnCHdzzaQ9vZZSXh9qXYFnPRG049alw,34247
+mteb/results/model_result.py,sha256=h894O5-RSCOF8XNpXMuhBCqnj43T-1K5Y1el_fyrzP4,15954
+mteb/results/task_result.py,sha256=Wi5MRQBkb0Qddhc4nLzrrjm1nGlCh8aq4_VCQoxGmNg,34300
 mteb/tasks/__init__.py,sha256=izAxU0ip1F_YUwx0dFCuN35BaktdmePh6vlDiHC0kLo,503
 mteb/tasks/aggregated_tasks/__init__.py,sha256=Ufgbh1AirxCQkojO3AUhUFWM8zQG10cfdVTkj_PeyLI,104
 mteb/tasks/aggregated_tasks/eng/__init__.py,sha256=HgaSyAX8Is5CGE006RgJkLQQVxrx2FmMnm6NHQBDi-4,358
@@ -2113,7 +2113,7 @@ mteb/tasks/pair_classification/eng/twitter_sem_eval2015_pc.py,sha256=0YjKK4C47Uu
 mteb/tasks/pair_classification/eng/twitter_url_corpus_pc.py,sha256=M9B3JCFt4L6yEnd8S-o-W-FtCSMdl0h1KST3fqApEVA,1796
 mteb/tasks/pair_classification/fas/__init__.py,sha256=1Bbr5ZKSjpPuJb9zvk7OSd2Krdh1bpxJjVNLNPFT4Ck,440
 mteb/tasks/pair_classification/fas/fa_mteb_pair_classification.py,sha256=1i8phewQffpIxzWtBWQFUisHu3XhBpk9Sf-IkwM8jNg,10932
-mteb/tasks/pair_classification/fas/fars_tail.py,sha256=NPtN4mAQ6TcBJmX_GD7WZ4mE-_Bw3D9Aj-GEliZDIMo,2848
+mteb/tasks/pair_classification/fas/fars_tail.py,sha256=jb-6UW0Lk7YxdMMCZsMavY6CRiv3T6MFrbvlPd0vPPk,1676
 mteb/tasks/pair_classification/hye/__init__.py,sha256=hU4xSf6kyKhD4o4CuNMQNE1w9FKv8tkkqvYvhpMV5Kg,93
 mteb/tasks/pair_classification/hye/armenian_paraphrase_pc.py,sha256=Ezi604W-cHOLDm8O9j3yq9z-GzDt9OWI9jgyqVjY9M4,1437
 mteb/tasks/pair_classification/ind/__init__.py,sha256=iXGvZ6eNgGhyD2wgbkvV-bpPPCJNxlE5eq_qvF2Y_UI,53
@@ -2646,9 +2646,9 @@ mteb/types/_metadata.py,sha256=NN-W0S6a5TDV7UkpRx1pyWtGF4TyyCyoPUfHOwdeci8,2290
 mteb/types/_result.py,sha256=UKNokV9pu3G74MGebocU512aU_fFU9I9nPKnrG9Q0iE,1035
 mteb/types/_string_validators.py,sha256=PY-dYq4E8O50VS3bLYdldPWp400fl_WzUjfVSkNWe8U,523
 mteb/types/statistics.py,sha256=gElgSShKBXpfcqaZHhU_d2UHln1CyzUj8FN8KFun_UA,4087
-mteb-2.7.13.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-mteb-2.7.13.dist-info/METADATA,sha256=Fy5yLv2nllzE1LSF6S33tHCnWsaiGZzI5HTEUC02ufU,14348
-mteb-2.7.13.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-mteb-2.7.13.dist-info/entry_points.txt,sha256=8IJoEJFKoDHmVnNev-qJ9pp4Ln7_1-ma9QsXnzVCzGU,39
-mteb-2.7.13.dist-info/top_level.txt,sha256=OLVIjcQAlWBz0bdmutKlWHLF42FF0hp4uVAg3ZyiG4U,5
-mteb-2.7.13.dist-info/RECORD,,
+mteb-2.7.15.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+mteb-2.7.15.dist-info/METADATA,sha256=EoUeroRRdre5jYbplBGCJuWs-6M7cZGpzwLqSQyJKgI,14348
+mteb-2.7.15.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+mteb-2.7.15.dist-info/entry_points.txt,sha256=8IJoEJFKoDHmVnNev-qJ9pp4Ln7_1-ma9QsXnzVCzGU,39
+mteb-2.7.15.dist-info/top_level.txt,sha256=OLVIjcQAlWBz0bdmutKlWHLF42FF0hp4uVAg3ZyiG4U,5
+mteb-2.7.15.dist-info/RECORD,,