mteb 2.7.2__py3-none-any.whl → 2.7.9__py3-none-any.whl

mteb/models/model_implementations/siglip_models.py

@@ -1,13 +1,18 @@
-from typing import Any
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
 
 import torch
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput, PromptType
 
 SIGLIP_CITATION = """@misc{zhai2023sigmoid,
       title={Sigmoid Loss for Language Image Pre-Training},
@@ -131,6 +136,7 @@ siglip_so400m_patch14_224 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=877_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=3347,
     max_tokens=16,
     embed_dim=1152,
@@ -155,6 +161,7 @@ siglip_so400m_patch14_384 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=878_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=3349,
     max_tokens=64,
     embed_dim=1152,
@@ -179,6 +186,7 @@ siglip_so400m_patch16_256_i18n = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=1_130_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=4306,
     max_tokens=64,
     embed_dim=1152,
@@ -203,6 +211,7 @@ siglip_base_patch16_256_multilingual = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=371_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=1414,
     max_tokens=64,
     embed_dim=768,
@@ -227,6 +236,7 @@ siglip_base_patch16_256 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=203_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=775,
     max_tokens=64,
     embed_dim=768,
@@ -251,6 +261,7 @@ siglip_base_patch16_512 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=204_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=777,
     max_tokens=64,
     embed_dim=768,
@@ -275,6 +286,7 @@ siglip_base_patch16_384 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=203_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=776,
     max_tokens=64,
     embed_dim=768,
@@ -299,6 +311,7 @@ siglip_base_patch16_224 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=203_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=775,
     max_tokens=64,
     embed_dim=768,
@@ -323,6 +336,7 @@ siglip_large_patch16_256 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=652_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=2488,
     max_tokens=64,
     embed_dim=1024,
@@ -347,6 +361,7 @@ siglip_large_patch16_384 = ModelMeta(
     release_date="2024-01-08",
     modalities=["image", "text"],
     n_parameters=652_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=2489,
     max_tokens=64,
     embed_dim=1024,

mteb/models/model_implementations/slm_models.py

@@ -13,24 +13,27 @@ Based on:
 from __future__ import annotations
 
 import logging
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import torch
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
 from mteb._requires_package import (
     requires_image_dependencies,
     requires_package,
 )
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_implementations.colpali_models import (
     COLPALI_CITATION,
     COLPALI_TRAINING_DATA,
 )
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput, PromptType
 
 logger = logging.getLogger(__name__)
 

mteb/models/model_implementations/sonar_models.py

@@ -224,7 +224,8 @@ sonar = ModelMeta(
     use_instructions=False,  # it does take a language code as input
     revision="a551c586dcf4a49c8fd847de369412d556a7f2f2",
     release_date="2021-05-21",
-    n_parameters=None,  # it is really multiple models so not sure how to calculate this
+    n_parameters=None,
+    n_embedding_parameters=None,  # it is really multiple models so not sure how to calculate this
     max_tokens=512,  # https://github.com/facebookresearch/SONAR/blob/549d287466443bd8720f938047882630c1c5c3f7/sonar/models/sonar_text/builder.py#L139
     embed_dim=1024,
     license="mit",

mteb/models/model_implementations/spartan8806_atles_champion.py

@@ -12,6 +12,7 @@ spartan8806_atles_champion_embedding = ModelMeta(
     revision="d4c74d7000bbd25f3597fc0f2dcde59ef1386e8f",
     release_date="2025-11-15",
     n_parameters=110_000_000,
+    n_embedding_parameters=23_444_736,
     memory_usage_mb=420,
     max_tokens=512,
     embed_dim=768,

mteb/models/model_implementations/stella_models.py

@@ -66,6 +66,7 @@ stella_en_400m = ModelMeta(
     revision="1bb50bc7bb726810eac2140e62155b88b0df198f",
     release_date="2024-07-12",
     n_parameters=435_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=1660,
     max_tokens=8192,
     embed_dim=4096,
@@ -101,6 +102,7 @@ stella_en_1_5b = ModelMeta(
     revision="d03be74b361d4eb24f42a2fe5bd2e29917df4604",
     release_date="2024-07-12",
     n_parameters=1_540_000_000,
+    n_embedding_parameters=232_928_256,
     memory_usage_mb=5887,
     max_tokens=131072,
     embed_dim=8960,
@@ -130,6 +132,7 @@ stella_large_zh_v3_1792d = ModelMeta(
     revision="d5d39eb8cd11c80a63df53314e59997074469f09",
     release_date="2024-02-17",
     n_parameters=None,
+    n_embedding_parameters=21_635_072,
     memory_usage_mb=None,  # can't see on model card
     embed_dim=1792,
     license="not specified",
@@ -157,6 +160,7 @@ stella_base_zh_v3_1792d = ModelMeta(
     revision="82254892a0fba125aa2abf3a4800d2dd12821343",
     release_date="2024-02-17",
     n_parameters=None,
+    n_embedding_parameters=16_226_304,
     memory_usage_mb=None,  # can't see on model card
     embed_dim=1792,
     license="mit",
@@ -185,6 +189,7 @@ stella_mrl_large_zh_v3_5_1792d = ModelMeta(
     revision="17bb1c32a93a8fc5f6fc9e91d5ea86da99983cfe",
     release_date="2024-02-27",
     n_parameters=int(326 * 1e6),
+    n_embedding_parameters=21_635_072,
     memory_usage_mb=1242,
     embed_dim=1792,
     license="mit",
@@ -209,6 +214,7 @@ zpoint_large_embedding_zh = ModelMeta(
     revision="b1075144f440ab4409c05622c1179130ebd57d03",
     release_date="2024-06-04",
     n_parameters=int(326 * 1e6),
+    n_embedding_parameters=21_635_072,
     memory_usage_mb=1242,
     embed_dim=1792,
     license="mit",

mteb/models/model_implementations/tarka_models.py

@@ -327,6 +327,7 @@ tarka_embedding_150m_v1 = ModelMeta(
     revision="b0ffecc4ef0d873e517507ed080e43b88b2704b9",
     release_date="2025-11-04",
     n_parameters=155_714_304,
+    n_embedding_parameters=None,
     embed_dim=768,
     max_tokens=2048,
     license="gemma",
@@ -361,6 +362,7 @@ tarka_embedding_350m_v1 = ModelMeta(
     revision="a850d6a329145474727424fed6b12b62096b8ba3",
     release_date="2025-11-11",
     n_parameters=354_483_968,
+    n_embedding_parameters=None,
     memory_usage_mb=676,
     embed_dim=1024,
     max_tokens=128000,

mteb/models/model_implementations/text2vec_models.py

@@ -22,6 +22,7 @@ text2vec_base_chinese = ModelMeta(
     revision="183bb99aa7af74355fb58d16edf8c13ae7c5433e",
     release_date="2022-01-23",
     n_parameters=int(102 * 1e6),
+    n_embedding_parameters=16_226_304,
     embed_dim=768,
     license="apache-2.0",
     max_tokens=512,
@@ -51,6 +52,7 @@ text2vec_base_chinese_paraphrase = ModelMeta(
     revision="e90c150a9c7fb55a67712a766d6820c55fb83cdd",
     release_date="2023-06-19",
     n_parameters=118 * 1e6,
+    n_embedding_parameters=30_720_000,
     memory_usage_mb=450,
     embed_dim=768,
     license="apache-2.0",
@@ -95,6 +97,7 @@ text2vec_base_multilingual = ModelMeta(
     # So probably best not to.
     loader=sentence_transformers_loader,
     n_parameters=117654272,
+    n_embedding_parameters=96_014_208,
     memory_usage_mb=449,
     embed_dim=384,
     license="apache-2.0",

mteb/models/model_implementations/ua_sentence_models.py

@@ -8,6 +8,7 @@ xlm_roberta_ua_distilled = ModelMeta(
     model_type=["dense"],
     loader=sentence_transformers_loader,
     n_parameters=278_000_000,
+    n_embedding_parameters=192_001_536,
     memory_usage_mb=1061,
     max_tokens=512,
     embed_dim=768,

mteb/models/model_implementations/uae_models.py

@@ -1,13 +1,18 @@
+from __future__ import annotations
+
 import logging
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import torch
-from torch.utils.data import DataLoader
 
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.model_meta import ModelMeta, ScoringFunction
 from mteb.models.sentence_transformer_wrapper import SentenceTransformerEncoderWrapper
-from mteb.types import Array, BatchedInput, PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput, PromptType
 
 logger = logging.getLogger(__name__)
 
@@ -67,6 +72,7 @@ uae_large_v1 = ModelMeta(
     revision="369c368f70f16a613f19f5598d4f12d9f44235d4",
     release_date="2023-12-04",  # initial commit of hf model.
     n_parameters=int(335 * 1e6),
+    n_embedding_parameters=31_254_528,
     memory_usage_mb=1278,
     max_tokens=512,
     embed_dim=1024,

mteb/models/model_implementations/vdr_models.py

@@ -1,6 +1,12 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
 from mteb.models.instruct_wrapper import InstructSentenceTransformerModel
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import PromptType
+
+if TYPE_CHECKING:
+    from mteb.types import PromptType
 
 
 def instruction_template(
@@ -32,6 +38,7 @@ vdr_2b_multi_v1 = ModelMeta(
     release_date="2024-01-08",
     modalities=["text"],  # TODO: integrate with image
     n_parameters=2_000_000_000,
+    n_embedding_parameters=233_373_696,
     memory_usage_mb=4213,
     max_tokens=32768,
     embed_dim=1536,

mteb/models/model_implementations/vi_vn_models.py

@@ -16,6 +16,7 @@ greennode_embedding_large_vn_v1 = ModelMeta(
     loader=sentence_transformers_loader,
     open_weights=True,
     n_parameters=568_000_000,
+    n_embedding_parameters=256_002_048,
     memory_usage_mb=2167,
     embed_dim=1024,
     license="cc-by-4.0",
@@ -41,6 +42,7 @@ greennode_embedding_large_vn_mixed_v1 = ModelMeta(
     loader=sentence_transformers_loader,
     open_weights=True,
     n_parameters=568_000_000,
+    n_embedding_parameters=256_002_048,
     memory_usage_mb=2167,
     embed_dim=1024,
     license="cc-by-4.0",
@@ -66,6 +68,7 @@ aiteamvn_vietnamese_embeddings = ModelMeta(
     loader=sentence_transformers_loader,
     open_weights=True,
     n_parameters=568_000_000,
+    n_embedding_parameters=256_002_048,
     memory_usage_mb=2166,
     embed_dim=1024,
     license="cc-by-4.0",
@@ -98,6 +101,7 @@ hiieu_halong_embedding = ModelMeta(
     use_instructions=False,
     open_weights=True,
     n_parameters=278_000_000,
+    n_embedding_parameters=192_001_536,
     memory_usage_mb=1061,
     embed_dim=768,
     license="apache-2.0",
@@ -129,6 +133,7 @@ sup_simcse_vietnamese_phobert_base_ = ModelMeta(
     use_instructions=False,
     open_weights=True,
     n_parameters=135_000_000,
+    n_embedding_parameters=49_152_768,
     memory_usage_mb=517,
     max_tokens=256,
     embed_dim=768,
@@ -167,6 +172,7 @@ bkai_foundation_models_vietnamese_bi_encoder = ModelMeta(
     use_instructions=False,
     open_weights=True,
     n_parameters=135_000_000,
+    n_embedding_parameters=49_152_768,
     memory_usage_mb=515,
     max_tokens=256,
     embed_dim=768,

mteb/models/model_implementations/vista_models.py

@@ -1,14 +1,19 @@
-from typing import Any, Literal
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any, Literal
 
 import torch
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
 from mteb._requires_package import requires_image_dependencies
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput, PromptType
 
 VISTA_CITATION = """@article{zhou2024vista,
   title={VISTA: Visualized Text Embedding For Universal Multi-Modal Retrieval},
@@ -253,6 +258,7 @@ visualized_bge_base = ModelMeta(
     release_date="2024-06-06",
     modalities=["image", "text"],
     n_parameters=196_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=1631,
     max_tokens=512,
     embed_dim=768,
@@ -281,6 +287,7 @@ visualized_bge_m3 = ModelMeta(
     release_date="2024-06-06",
     modalities=["image", "text"],
     n_parameters=872_909_505,
+    n_embedding_parameters=None,
     memory_usage_mb=4263,
     max_tokens=8192,
     embed_dim=1024,

mteb/models/model_implementations/vlm2vec_models.py

@@ -1,8 +1,9 @@
+from __future__ import annotations
+
 import logging
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import torch
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
 from mteb._requires_package import (
@@ -10,10 +11,14 @@ from mteb._requires_package import (
     requires_package,
     suggest_package,
 )
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput, PromptType
 
 logger = logging.getLogger(__name__)
 
@@ -275,6 +280,7 @@ vlm2vec_lora = ModelMeta(
     release_date="2024-10-08",
     modalities=["image", "text"],
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     max_tokens=131072,
     embed_dim=3072,
@@ -299,6 +305,7 @@ vlm2vec_full = ModelMeta(
     release_date="2024-10-08",
     modalities=["image", "text"],
     n_parameters=4_150_000_000,
+    n_embedding_parameters=None,
     memory_usage_mb=7909,
     max_tokens=131072,
     embed_dim=3072,

mteb/models/model_implementations/voyage_models.py

@@ -1,16 +1,22 @@
+from __future__ import annotations
+
 import time
 from functools import wraps
-from typing import Any, Literal
+from typing import TYPE_CHECKING, Any, Literal
 
 import numpy as np
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
 from mteb._requires_package import requires_package
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+from mteb.types import PromptType
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput
 
 VOYAGE_TRAINING_DATA = set(
     # Self-reported (message from VoyageAI member)
@@ -176,6 +182,7 @@ class VoyageModel(AbsEncoder):
                     model=self._model_name,
                     input_type=input_type,
                     output_dtype=output_dtype,
+                    output_dimension=self.mteb_model_meta.embed_dim,
                 ).embeddings
             )
             pbar.update(len(batch))
@@ -209,6 +216,32 @@ model_prompts = {
     PromptType.document.value: "document",
 }
 
+voyage_4_large_2048d = ModelMeta(
+    name="voyageai/voyage-4-large (embed_dim=2048)",
+    model_type=["dense"],
+    revision="1",
+    release_date="2026-01-15",
+    languages=None,  # supported languages not specified
+    loader=VoyageModel,
+    loader_kwargs=dict(
+        max_tokens=32000,
+        model_prompts=model_prompts,
+    ),
+    max_tokens=32000,
+    embed_dim=2048,
+    open_weights=False,
+    n_parameters=None,
+    memory_usage_mb=None,
+    license=None,
+    reference="https://blog.voyageai.com/2026/01/15/voyage-4/",
+    similarity_fn_name="cosine",
+    framework=["API"],
+    use_instructions=True,
+    training_datasets=VOYAGE_TRAINING_DATA,
+    public_training_code=None,
+    public_training_data=None,
+)
+
 voyage_4 = ModelMeta(
     name="voyageai/voyage-4",
     model_type=["dense"],
@@ -302,6 +335,7 @@ voyage_3_large = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2025/01/07/voyage-3-large/",
@@ -330,6 +364,7 @@ voyage_3_5 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2025/05/20/voyage-3-5/",
@@ -357,6 +392,7 @@ voyage_3_5_int8 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2025/05/20/voyage-3-5/",
@@ -384,6 +420,7 @@ voyage_3_5_binary = ModelMeta(
     embed_dim=1024,  # Same as original after unpacking from bits
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2025/05/20/voyage-3-5/",
@@ -411,6 +448,7 @@ voyage_large_2_instruct = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/05/05/voyage-large-2-instruct-instruction-tuned-and-rank-1-on-mteb/",
@@ -437,6 +475,7 @@ voyage_finance_2 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/06/03/domain-specific-embeddings-finance-edition-voyage-finance-2/",
@@ -463,6 +502,7 @@ voyage_law_2 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/04/15/domain-specific-embeddings-and-retrieval-legal-edition-voyage-law-2/",
@@ -489,6 +529,7 @@ voyage_code_2 = ModelMeta(
     embed_dim=1536,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/01/23/voyage-code-2-elevate-your-code-retrieval/",
@@ -515,6 +556,7 @@ voyage_code_3 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/12/04/voyage-code-3/",
@@ -542,6 +584,7 @@ voyage_large_2 = ModelMeta(
     embed_dim=1536,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2023/10/29/voyage-embeddings/",
@@ -568,6 +611,7 @@ voyage_2 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2023/10/29/voyage-embeddings/",
@@ -593,6 +637,7 @@ voyage_multilingual_2 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/06/10/voyage-multilingual-2-multilingual-embedding-model/",
@@ -619,6 +664,7 @@ voyage_3 = ModelMeta(
     embed_dim=1024,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/09/18/voyage-3/",
@@ -645,6 +691,7 @@ voyage_3_lite = ModelMeta(
     embed_dim=512,
     open_weights=False,
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://blog.voyageai.com/2024/09/18/voyage-3/",
@@ -673,6 +720,7 @@ voyage_3_exp = ModelMeta(
     open_weights=False,
     # from their card https://huggingface.co/voyageai/voyage-3-m-exp#model-information
     n_parameters=int(6918 * 1e6),
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     license=None,
     reference="https://huggingface.co/voyageai/voyage-3-m-exp",

mteb/models/model_implementations/voyage_v.py

@@ -4,17 +4,19 @@ import logging
 from typing import TYPE_CHECKING, Any, Literal
 
 import torch
-from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
 from mteb._requires_package import requires_image_dependencies, requires_package
-from mteb.abstasks.task_metadata import TaskMetadata
 from mteb.models.abs_encoder import AbsEncoder
 from mteb.models.model_meta import ModelMeta, ScoringFunction
-from mteb.types import Array, BatchedInput, PromptType
+from mteb.types import PromptType
 
 if TYPE_CHECKING:
     from PIL import Image
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import Array, BatchedInput
 
 logger = logging.getLogger(__name__)
 
@@ -27,6 +29,8 @@ def _downsample_image(
     Returns:
         The downsampled image.
     """
+    from PIL.Image import Resampling
+
     width, height = image.size
     pixels = width * height
 
@@ -42,15 +46,15 @@ def _downsample_image(
         logger.info(
             f"Downsampling image from {width}x{height} to {new_width}x{new_height}"
         )
-        return image.resize(new_size, Image.LANCZOS)
+        return image.resize(new_size, Resampling.LANCZOS)
     if width > height:
         if width > 10000:
             logger.error("Processing extremely wide images.")
-            return image.resize((10000, height), Image.LANCZOS)
+            return image.resize((10000, height), Resampling.LANCZOS)
     else:
         if height > 10000:
             logger.error("Processing extremely high images.")
-            return image.resize((width, 10000), Image.LANCZOS)
+            return image.resize((width, 10000), Resampling.LANCZOS)
     return image
 
 
@@ -211,6 +215,7 @@ voyage_v = ModelMeta(
     revision="1",
     release_date="2024-11-10",
     n_parameters=None,
+    n_embedding_parameters=None,
     memory_usage_mb=None,
     max_tokens=32768,
     embed_dim=1024,

mteb/models/model_implementations/xyz_models.py

@@ -31,6 +31,7 @@ xyz_embedding = ModelMeta(
     revision="4004120220b99baea764a1d3508427248ac3bccf",
     release_date="2024-09-13",
     n_parameters=326000000,
+    n_embedding_parameters=21_635_072,
     memory_usage_mb=1242,
     max_tokens=512,
     embed_dim=768,