mteb 2.5.2__py3-none-any.whl → 2.7.9__py3-none-any.whl

mteb/models/model_implementations/spartan8806_atles_champion.py

@@ -12,12 +12,13 @@ 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,
     license="apache-2.0",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers"],
+    framework=["Sentence Transformers", "safetensors", "Transformers"],
     reference="https://huggingface.co/spartan8806/atles-champion-embedding",
     use_instructions=False,
     training_datasets={"STSBenchmark"},

mteb/models/model_implementations/stella_models.py

@@ -66,12 +66,19 @@ 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,
     license="mit",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch", "GritLM"],
+    framework=[
+        "Sentence Transformers",
+        "PyTorch",
+        "GritLM",
+        "safetensors",
+        "Transformers",
+    ],
     reference="https://huggingface.co/NovaSearch/stella_en_400M_v5",
     training_datasets=nvidia_training_datasets,  # also distilled from gte-qwen (but training data is unknown) #2164
     public_training_code="https://github.com/NovaSearch-Team/RAG-Retrieval/blob/c40f4638b705eb77d88305d2056901ed550f9f4b/rag_retrieval/train/embedding/README.md",
@@ -95,12 +102,20 @@ 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,
     license="mit",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch", "GritLM"],
+    framework=[
+        "Sentence Transformers",
+        "PyTorch",
+        "GritLM",
+        "ONNX",
+        "safetensors",
+        "Transformers",
+    ],
     reference="https://huggingface.co/NovaSearch/stella_en_1.5B_v5",
     training_datasets=nvidia_training_datasets,  # also distilled from gte-qwen (but training data is unknown) #2164
     public_training_code="https://github.com/NovaSearch-Team/RAG-Retrieval/blob/c40f4638b705eb77d88305d2056901ed550f9f4b/rag_retrieval/train/embedding/README.md",
@@ -117,6 +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",
@@ -144,6 +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",
@@ -172,13 +189,14 @@ 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",
     max_tokens=512,
     reference="https://huggingface.co/dunzhang/stella-large-zh-v3-1792d",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=False,
     superseded_by=None,
     adapted_from="dunzhang/stella-large-zh-v3-1792d",
@@ -196,13 +214,14 @@ 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",
     max_tokens=512,
     reference="https://huggingface.co/iampanda/zpoint_large_embedding_zh",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=False,
     superseded_by=None,
     adapted_from="dunzhang/stella-mrl-large-zh-v3.5-1792d",

mteb/models/model_implementations/tarka_models.py

@@ -327,11 +327,12 @@ 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",
     reference="https://huggingface.co/Tarka-AIR/Tarka-Embedding-150M-V1",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=True,
     public_training_code=None,
     public_training_data=None,
@@ -361,13 +362,14 @@ 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,
     license=None,
     reference="https://huggingface.co/Tarka-AIR/Tarka-Embedding-350M-V1",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=True,
     public_training_code=None,
     public_training_data=None,

mteb/models/model_implementations/text2vec_models.py

@@ -22,12 +22,13 @@ 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,
     reference="https://huggingface.co/shibing624/text2vec-base-chinese",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "ONNX", "safetensors"],
     use_instructions=False,
     superseded_by=None,
     adapted_from=None,
@@ -51,13 +52,14 @@ 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",
     max_tokens=512,
     reference="https://huggingface.co/shibing624/text2vec-base-chinese-paraphrase",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors", "Transformers"],
     use_instructions=False,
     superseded_by=None,
     adapted_from=None,
@@ -95,13 +97,20 @@ 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",
     max_tokens=256,
     reference="https://huggingface.co/shibing624/text2vec-base-chinese-paraphrase",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=[
+        "Sentence Transformers",
+        "PyTorch",
+        "ONNX",
+        "safetensors",
+        "Transformers",
+    ],
     use_instructions=False,
     superseded_by=None,
     adapted_from="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",

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,
@@ -17,7 +18,7 @@ xlm_roberta_ua_distilled = ModelMeta(
     open_weights=True,
     public_training_code="https://github.com/panalexeu/xlm-roberta-ua-distilled/blob/main/researches/research_final.ipynb",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers"],
+    framework=["Sentence Transformers", "safetensors"],
     reference="https://github.com/panalexeu/xlm-roberta-ua-distilled/tree/main",
     languages=["eng-Latn", "ukr-Cyrl"],
     training_datasets=set(

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__)
 
@@ -67,12 +72,19 @@ 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,
     license="mit",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=[
+        "Sentence Transformers",
+        "PyTorch",
+        "ONNX",
+        "safetensors",
+        "Transformers",
+    ],
     reference="https://huggingface.co/WhereIsAI/UAE-Large-V1",
     use_instructions=True,
     citation="""

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(
@@ -32,13 +38,14 @@ 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,
     license="apache-2.0",
     reference="https://huggingface.co/llamaindex/vdr-2b-multi-v1",
     similarity_fn_name=ScoringFunction.COSINE,
-    framework=["PyTorch", "Sentence Transformers"],
+    framework=["PyTorch", "Sentence Transformers", "safetensors", "Transformers"],
     use_instructions=True,
     public_training_code=None,
     public_training_data="https://huggingface.co/datasets/llamaindex/vdr-multilingual-train",

mteb/models/model_implementations/vi_vn_models.py

@@ -16,13 +16,14 @@ 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",
     max_tokens=8194,
     reference="https://huggingface.co/GreenNode/GreenNode-Embedding-Large-VN-V1",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=False,
     public_training_code=None,
     public_training_data="https://huggingface.co/datasets/GreenNode/GreenNode-Table-Markdown-Retrieval-VN",
@@ -41,13 +42,14 @@ 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",
     max_tokens=8194,
     reference="https://huggingface.co/GreenNode/GreenNode-Embedding-Large-VN-Mixed-V1",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     use_instructions=False,
     public_training_code=None,
     public_training_data="https://huggingface.co/datasets/GreenNode/GreenNode-Table-Markdown-Retrieval-VN",
@@ -66,13 +68,14 @@ 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",
     max_tokens=8194,
     reference="https://huggingface.co/AITeamVN/Vietnamese_Embedding",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "ONNX", "safetensors"],
     use_instructions=False,
     public_training_code=None,
     public_training_data=None,
@@ -98,13 +101,14 @@ 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",
     max_tokens=514,
     reference="https://huggingface.co/hiieu/halong_embedding",
     similarity_fn_name="cosine",
-    framework=["Sentence Transformers", "PyTorch"],
+    framework=["Sentence Transformers", "PyTorch", "safetensors"],
     public_training_code=None,
     public_training_data=None,
     training_datasets=None,
@@ -129,13 +133,14 @@ 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,
     license="apache-2.0",
     public_training_code=None,
     public_training_data=None,
-    framework=["PyTorch", "Sentence Transformers"],
+    framework=["PyTorch", "Sentence Transformers", "Transformers", "safetensors"],
     reference="https://huggingface.co/VoVanPhuc/sup-SimCSE-VietNamese-phobert-base",
     similarity_fn_name="cosine",
     training_datasets=None,
@@ -167,13 +172,14 @@ 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,
     license="apache-2.0",
     public_training_code=None,
     public_training_data=None,
-    framework=["PyTorch", "Sentence Transformers"],
+    framework=["PyTorch", "Sentence Transformers", "safetensors", "Transformers"],
     reference="https://huggingface.co/bkai-foundation-models/vietnamese-bi-encoder",
     similarity_fn_name="cosine",
     training_datasets=None,

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__)
 
@@ -41,7 +46,7 @@ class VLM2VecWrapper(AbsEncoder):
             model_name,
             "pip install flash-attn --no-build-isolation",
         ):
-            import flash_attn  # noqa
+            pass
 
         requires_package(self, "peft", model_name, "pip install 'mteb[peft]'")
         from peft import LoraConfig, PeftModel
@@ -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,
@@ -282,7 +288,7 @@ vlm2vec_lora = ModelMeta(
     open_weights=True,
     public_training_code="https://github.com/TIGER-AI-Lab/VLM2Vec",
     public_training_data="https://huggingface.co/datasets/TIGER-Lab/MMEB-train",
-    framework=["PyTorch"],
+    framework=["PyTorch", "Transformers"],
     reference="https://huggingface.co/TIGER-Lab/VLM2Vec-LoRA",
     similarity_fn_name=ScoringFunction.COSINE,
     use_instructions=True,
@@ -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,
@@ -306,7 +313,7 @@ vlm2vec_full = ModelMeta(
     open_weights=True,
     public_training_code="https://github.com/TIGER-AI-Lab/VLM2Vec",
     public_training_data="https://huggingface.co/TIGER-Lab/VLM2Vec-Full",
-    framework=["PyTorch"],
+    framework=["PyTorch", "Transformers", "safetensors"],
     reference="https://huggingface.co/TIGER-Lab/VLM2Vec-Full",
     similarity_fn_name=ScoringFunction.COSINE,
     use_instructions=True,