mteb 2.3.4__py3-none-any.whl → 2.3.6__py3-none-any.whl

mteb/benchmarks/benchmarks/__init__.py

@@ -12,6 +12,7 @@ from mteb.benchmarks.benchmarks.benchmarks import (
     FA_MTEB_2,
     HUME,
     JINA_VDR,
+    JMTEB_V2,
     LONG_EMBED,
     MIEB_ENG,
     MIEB_IMG,
@@ -75,6 +76,7 @@ __all__ = [
     "HUME",
     "HUME",
     "JINA_VDR",
+    "JMTEB_V2",
     "LONG_EMBED",
     "MIEB_ENG",
     "MIEB_IMG",

mteb/benchmarks/benchmarks/benchmarks.py

@@ -2562,3 +2562,60 @@ HUME = HUMEBenchmark(
     citation=None,
     contacts=["AdnanElAssadi56", "KennethEnevoldsen", "isaac-chung", "Samoed"],
 )
+
+JMTEB_V2 = Benchmark(
+    name="JMTEB(v2)",
+    display_name="Japanese",
+    icon="https://github.com/lipis/flag-icons/raw/260c91531be024944c6514130c5defb2ebb02b7d/flags/4x3/jp.svg",
+    tasks=get_tasks(
+        languages=["jpn"],
+        tasks=[
+            # Clustering (3)
+            "LivedoorNewsClustering.v2",
+            "MewsC16JaClustering",
+            "SIB200ClusteringS2S",
+            # Classification (7)
+            "AmazonReviewsClassification",
+            "AmazonCounterfactualClassification",
+            "MassiveIntentClassification",
+            "MassiveScenarioClassification",
+            "JapaneseSentimentClassification",
+            "SIB200Classification",
+            "WRIMEClassification",
+            # STS (2)
+            "JSTS",
+            "JSICK",
+            # Retrieval (11)
+            "JaqketRetrieval",
+            "MrTidyRetrieval",
+            "JaGovFaqsRetrieval",
+            "NLPJournalTitleAbsRetrieval.V2",
+            "NLPJournalTitleIntroRetrieval.V2",
+            "NLPJournalAbsIntroRetrieval.V2",
+            "NLPJournalAbsArticleRetrieval.V2",
+            "JaCWIRRetrieval",
+            "MIRACLRetrieval",
+            "MintakaRetrieval",
+            "MultiLongDocRetrieval",
+            # Reranking (5)
+            "ESCIReranking",
+            "JQaRAReranking",
+            "JaCWIRReranking",
+            "MIRACLReranking",
+            "MultiLongDocReranking",
+        ],
+    ),
+    description="JMTEB is a benchmark for evaluating Japanese text embedding models. In v2, we have extended the benchmark to 28 datasets, enabling more comprehensive evaluation compared with v1 (MTEB(jpn, v1)).",
+    reference="https://github.com/sbintuitions/JMTEB",
+    citation=r"""
+@article{li2025jmteb,
+  author = {Li, Shengzhe and Ohagi, Masaya and Ri, Ryokan and Fukuchi, Akihiko and Shibata, Tomohide and Kawahara, Daisuke},
+  issue = {3},
+  journal = {Vol.2025-NL-265,No.3,1-15},
+  month = {sep},
+  title = {{JMTEB and JMTEB-lite: Japanese Massive Text Embedding Benchmark and Its Lightweight Version}},
+  year = {2025},
+}
+""",
+    contacts=["lsz05"],
+)

mteb/cache.py

@@ -243,7 +243,11 @@ class ResultCache:
             f"No results repository found in {results_directory}, cloning it from {remote}"
         )
 
-        subprocess.run(["git", "clone", remote, "remote"], cwd=self.cache_path)
+        subprocess.run(
+            ["git", "clone", "--depth", "1", remote, "remote"],
+            cwd=self.cache_path,
+            check=True,
+        )
 
         return results_directory
 

mteb/descriptive_stats/Classification/HebrewSentimentAnalysis.v3.json

@@ -0,0 +1,60 @@
+{
+    "test": {
+        "num_samples": 1085,
+        "number_texts_intersect_with_train": 0,
+        "text_statistics": {
+            "total_text_length": 115359,
+            "min_text_length": 8,
+            "average_text_length": 106.32165898617511,
+            "max_text_length": 2722,
+            "unique_texts": 1085
+        },
+        "image_statistics": null,
+        "label_statistics": {
+            "min_labels_per_text": 1,
+            "average_label_per_text": 1.0,
+            "max_labels_per_text": 1,
+            "unique_labels": 3,
+            "labels": {
+                "0": {
+                    "count": 868
+                },
+                "1": {
+                    "count": 190
+                },
+                "2": {
+                    "count": 27
+                }
+            }
+        }
+    },
+    "train": {
+        "num_samples": 7176,
+        "number_texts_intersect_with_train": null,
+        "text_statistics": {
+            "total_text_length": 830248,
+            "min_text_length": 5,
+            "average_text_length": 115.69788182831661,
+            "max_text_length": 4759,
+            "unique_texts": 7176
+        },
+        "image_statistics": null,
+        "label_statistics": {
+            "min_labels_per_text": 1,
+            "average_label_per_text": 1.0,
+            "max_labels_per_text": 1,
+            "unique_labels": 3,
+            "labels": {
+                "0": {
+                    "count": 4933
+                },
+                "1": {
+                    "count": 2047
+                },
+                "2": {
+                    "count": 196
+                }
+            }
+        }
+    }
+}

mteb/leaderboard/figures.py

@@ -117,7 +117,7 @@ def _performance_size_plot(df: pd.DataFrame) -> go.Figure:
     df["Max Tokens"] = df["Max Tokens"].map(_parse_float)
     df["Log(Tokens)"] = np.log10(df["Max Tokens"])
     df["Mean (Task)"] = df["Mean (Task)"].map(_parse_float)
-    df = df.dropna(
+    df = df[df["Mean (Task)"] > 0].dropna(
         subset=["Mean (Task)", "Number of Parameters", "Embedding Dimensions"]
     )
     if not len(df.index):

mteb/models/model_implementations/kowshik24_models.py

@@ -0,0 +1,22 @@
+from mteb.models import ModelMeta, sentence_transformers_loader
+
+kowshik24_bangla_embedding_model = ModelMeta(
+    loader=sentence_transformers_loader,
+    name="Kowshik24/bangla-sentence-transformer-ft-matryoshka-paraphrase-multilingual-mpnet-base-v2",
+    languages=["ben-Beng"],  # Bengali using Bengali script
+    open_weights=True,
+    revision="6689c21e69be5950596bad084457cbaa138728d8",
+    release_date="2025-11-10",
+    n_parameters=278_000_000,
+    memory_usage_mb=1061,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=128,
+    reference="https://huggingface.co/Kowshik24/bangla-sentence-transformer-ft-matryoshka-paraphrase-multilingual-mpnet-base-v2",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=False,
+    public_training_code="https://github.com/kowshik24/Bangla-Embedding",
+    public_training_data="https://huggingface.co/datasets/sartajekram/BanglaRQA",
+    training_datasets=set(),
+)

mteb/models/model_implementations/ruri_models.py

@@ -0,0 +1,312 @@
+from mteb.models.model_meta import ModelMeta
+from mteb.models.sentence_transformer_wrapper import sentence_transformers_loader
+
+RURI_V3_PROMPTS = {
+    "Retrieval-query": "検索クエリ: ",
+    "Retrieval-document": "検索文書: ",
+    "Reranking-query": "検索クエリ: ",
+    "Reranking-document": "検索文書: ",
+    "Classification": "トピック: ",
+    "Clustering": "トピック: ",
+}
+
+RURI_V1_V2_PROMPTS = {
+    "query": "クエリ: ",
+    "document": "文章: ",
+}
+
+
+RURI_CITATION = r"""@misc{Ruri,
+  title={{Ruri: Japanese General Text Embeddings}},
+  author={Hayato Tsukagoshi and Ryohei Sasano},
+  year={2024},
+  eprint={2409.07737},
+  archivePrefix={arXiv},
+  primaryClass={cs.CL},
+  url={https://arxiv.org/abs/2409.07737},
+}"""
+
+cl_nagoya_ruri_v3_30m = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V3_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-v3-30m",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="24899e5de370b56d179604a007c0d727bf144504",
+    release_date="2025-04-07",
+    n_parameters=36_705_536,
+    memory_usage_mb=140,
+    embed_dim=256,
+    license="apache-2.0",
+    max_tokens=8192,
+    reference="https://huggingface.co/cl-nagoya/ruri-v3-30m",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    superseded_by=None,
+    training_datasets={
+        "cl-nagoya/ruri-v3-dataset-ft",
+    },
+    adapted_from="sbintuitions/modernbert-ja-30m",
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-v3-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_v3_70m = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V3_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-v3-70m",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="07a8b0aba47d29d2ca21f89b915c1efe2c23d1cc",
+    release_date="2025-04-09",
+    n_parameters=36_705_536,
+    memory_usage_mb=140,
+    embed_dim=256,
+    license="apache-2.0",
+    max_tokens=8192,
+    reference="https://huggingface.co/cl-nagoya/ruri-v3-70m",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    superseded_by=None,
+    training_datasets={"MrTidyRetrieval", "MIRACLRetrieval"},
+    adapted_from="sbintuitions/modernbert-ja-70m",
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-v3-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_v3_130m = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V3_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-v3-130m",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="e3114c6ee10dbab8b4b235fbc6dcf9dd4d5ac1a6",
+    release_date="2025-04-09",
+    n_parameters=132_140_544,
+    memory_usage_mb=504,
+    embed_dim=512,
+    license="apache-2.0",
+    max_tokens=8192,
+    reference="https://huggingface.co/cl-nagoya/ruri-v3-130m",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    superseded_by=None,
+    training_datasets={"MrTidyRetrieval", "MIRACLRetrieval"},
+    adapted_from="sbintuitions/modernbert-ja-130m",
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-v3-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_v3_310m = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V3_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-v3-310m",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="18b60fb8c2b9df296fb4212bb7d23ef94e579cd3",
+    release_date="2025-04-09",
+    n_parameters=314_611_968,
+    memory_usage_mb=1200,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=8192,
+    reference="https://huggingface.co/cl-nagoya/ruri-v3-310m",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    superseded_by=None,
+    training_datasets={"MrTidyRetrieval", "MIRACLRetrieval"},
+    adapted_from="sbintuitions/modernbert-ja-310m",
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-v3-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_small_v2 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+        trust_remote_code=True,
+    ),
+    name="cl-nagoya/ruri-small-v2",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="db18646e673b713cd0518a5bb0fefdce21e77cd9",
+    release_date="2024-12-05",
+    n_parameters=68_087_808,
+    memory_usage_mb=260,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-small-v2",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="line-corporation/line-distilbert-base-japanese",
+    superseded_by=None,
+    training_datasets={"MrTidyRetrieval", "MIRACLRetrieval"},
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-v2-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_base_v2 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-base-v2",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="8ce03882903668a01c83ca3b8111ac025a3bc734",
+    release_date="2024-12-05",
+    n_parameters=111_207_168,
+    memory_usage_mb=424,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-base-v2",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="tohoku-nlp/bert-base-japanese-v3",
+    superseded_by=None,
+    training_datasets=None,
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-v2-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_large_v2 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-large-v2",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="42898ef34a5574977380ebf0dfd28cbfbd36438b",
+    release_date="2024-12-06",
+    n_parameters=337_441_792,
+    memory_usage_mb=1287,
+    embed_dim=1024,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-large-v2",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="tohoku-nlp/bert-large-japanese-v2",
+    superseded_by=None,
+    training_datasets=None,
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-v2-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_small_v1 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+        trust_remote_code=True,
+    ),
+    name="cl-nagoya/ruri-small",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="bc56ce90cd7a979f6eb199fc52dfe700bfd94bc3",
+    release_date="2024-08-28",
+    n_parameters=68_087_808,
+    memory_usage_mb=130,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-small",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="line-corporation/line-distilbert-base-japanese",
+    superseded_by="cl-nagoya/ruri-small-v2",
+    training_datasets=None,
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+cl_nagoya_ruri_base_v1 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-base",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="1ae40b8b6c78518a499425086bab8fc16c2e4b0e",
+    release_date="2024-08-28",
+    n_parameters=111_207_168,
+    memory_usage_mb=212,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-base",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="tohoku-nlp/bert-base-japanese-v3",
+    superseded_by="cl-nagoya/ruri-base-v2",
+    training_datasets=None,
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)
+
+
+cl_nagoya_ruri_large_v1 = ModelMeta(
+    loader=sentence_transformers_loader,
+    loader_kwargs=dict(
+        model_prompts=RURI_V1_V2_PROMPTS,
+    ),
+    name="cl-nagoya/ruri-large",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="a011c39b13e8bc137ee13c6bc82191ece46c414c",
+    release_date="2024-08-28",
+    n_parameters=337_441_792,
+    memory_usage_mb=644,
+    embed_dim=1024,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/cl-nagoya/ruri-large",
+    similarity_fn_name="cosine",
+    framework=["PyTorch", "Sentence Transformers"],
+    use_instructions=True,
+    adapted_from="tohoku-nlp/bert-large-japanese-v2",
+    superseded_by="cl-nagoya/ruri-large-v2",
+    training_datasets=None,
+    public_training_code=None,
+    public_training_data="https://huggingface.co/datasets/cl-nagoya/ruri-dataset-ft",
+    citation=RURI_CITATION,
+    contacts=["hpprc"],
+)

mteb/models/model_implementations/sarashina_embedding_models.py

@@ -0,0 +1,166 @@
+from mteb.models.instruct_wrapper import InstructSentenceTransformerModel
+from mteb.models.model_meta import ModelMeta
+from mteb.models.sentence_transformer_wrapper import sentence_transformers_loader
+from mteb.types import PromptType
+
+SARASHINA_V2_INSTRUCTIONS = {
+    "Retrieval": {
+        "query": "クエリを与えるので、もっともクエリに意味が似ている一節を探してください。",
+        "document": "text: ",
+    },
+    "Reranking": {
+        "query": "クエリを与えるので、もっともクエリに意味が似ている一節を探してください。",
+        "document": "text: ",
+    },
+    "Classification": "与えられたドキュメントを適切なカテゴリに分類してください。",
+    "Clustering": "与えられたドキュメントのトピックまたはテーマを特定してください。",
+    # optimization regarding JMTEB
+    "LivedoorNewsClustering.v2": "与えられたニュース記事のトピックを特定してください。",
+    "MewsC16JaClustering": "与えられたニュース記事のトピックを特定してください。",
+    "SIB200ClusteringS2S": "与えられたテキストのトピックを特定してください。",
+    "AmazonReviewsClassification": "与えられたAmazonレビューを適切な評価カテゴリに分類してください。",
+    "AmazonCounterfactualClassification": "与えられたAmazonのカスタマーレビューのテキストを反事実か反事実でないかに分類してください。",
+    "MassiveIntentClassification": "ユーザーの発話をクエリとして与えるので、ユーザーの意図を見つけてください。",
+    "MassiveScenarioClassification": "ユーザーの発話をクエリとして与えるので、ユーザーシナリオを見つけてください。",
+    "JapaneseSentimentClassification": "与えられたテキストの感情極性をポジティブ(1)かネガティブか(0)に分類してください。",
+    "SIB200Classification": "与えられたテキストのトピックを特定してください。",
+    "WRIMEClassification": "与えられたテキストの感情極性（-2:強いネガティブ、-1:ネガティブ、0:ニュートラル、1:ポジティブ、2:強いポジティブ）を分類してください。",
+    "JSTS": "クエリを与えるので，もっともクエリに意味が似ている一節を探してください。",
+    "JSICK": "クエリを与えるので，もっともクエリに意味が似ている一節を探してください。",
+    "JaqketRetrieval": {
+        "query": "質問を与えるので、その質問に答えるのに役立つWikipediaの文章を検索してください。",
+        "document": "text: ",
+    },
+    "MrTidyRetrieval": {
+        "query": "質問を与えるので、その質問に答えるWikipediaの文章を検索するしてください。",
+        "document": "text: ",
+    },
+    "JaGovFaqsRetrieval": {
+        "query": "質問を与えるので、その質問に答えるのに役立つ関連文書を検索してください。",
+        "document": "text: ",
+    },
+    "NLPJournalTitleAbsRetrieval.V2": {
+        "query": "論文のタイトルを与えるので、タイトルに対応する要約を検索してください。",
+        "document": "text: ",
+    },
+    "NLPJournalTitleIntroRetrieval.V2": {
+        "query": "論文のタイトルを与えるので、タイトルに対応する要約を検索してください。",
+        "document": "text: ",
+    },
+    "NLPJournalAbsIntroRetrieval.V2": {
+        "query": "論文の序論を与えるので、序論に対応する全文を検索してください。",
+        "document": "text: ",
+    },
+    "NLPJournalAbsArticleRetrieval.V2": {
+        "query": "論文の序論を与えるので、序論に対応する全文を検索してください。",
+        "document": "text: ",
+    },
+    "JaCWIRRetrieval": {
+        "query": "記事のタイトルを与えるので、そのタイトルと合っている記事の中身を検索してください。",
+        "document": "text: ",
+    },
+    "MIRACLRetrieval": {
+        "query": "質問を与えるので、その質問に答えるのに役立つ関連文書を検索してください。",
+        "document": "text: ",
+    },
+    "MintakaRetrieval": {
+        "query": "質問を与えるので、その質問に答えられるテキストを検索してください。",
+        "document": "text: ",
+    },
+    "MultiLongDocRetrieval": {
+        "query": "質問を与えるので、その質問に答えるのに役立つWikipediaの文章を検索してください。",
+        "document": "text: ",
+    },
+    "ESCIReranking": {
+        "query": "クエリを与えるので、与えられたWeb検索クエリに答える関連文章を検索してください。",
+        "document": "text: ",
+    },
+    "JQaRAReranking": {
+        "query": "質問を与えるので、その質問に答えるのに役立つWikipediaの文章を検索してください。",
+        "document": "text: ",
+    },
+    "JaCWIRReranking": {
+        "query": "記事のタイトルを与えるので、そのタイトルと合っている記事の中身を検索してください。",
+        "document": "text: ",
+    },
+    "MIRACLReranking": {
+        "query": "質問を与えるので、その質問に答えるのに役立つ関連文書を検索してください。",
+        "document": "text: ",
+    },
+    "MultiLongDocReranking": {
+        "query": "質問を与えるので、その質問に答えるのに役立つWikipediaの文章を検索してください。",
+        "document": "text: ",
+    },
+}
+
+
+def sarashina_instruction_template(
+    instruction: str, prompt_type: PromptType | None = None
+) -> str:
+    """Instruction template for Sarashina v2 model.
+
+    Returns the instruction as-is since the prompts already contain the full format.
+    For document prompts, returns the instruction directly (e.g., "text: ").
+    """
+    if not instruction:
+        return ""
+    if prompt_type == PromptType.document:
+        return "text: "
+    return f"task: {instruction}\nquery: "
+
+
+sbintuitions_sarashina_embedding_v2_1b = ModelMeta(
+    loader=InstructSentenceTransformerModel,
+    loader_kwargs=dict(
+        instruction_template=sarashina_instruction_template,
+        apply_instruction_to_passages=True,
+        prompts_dict=SARASHINA_V2_INSTRUCTIONS,
+        max_seq_length=8192,
+    ),
+    name="sbintuitions/sarashina-embedding-v2-1b",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="1f3408afaa7b617e3445d891310a9c26dd0c68a5",
+    release_date="2025-07-30",
+    n_parameters=1_224_038_144,
+    memory_usage_mb=4669,
+    embed_dim=1792,
+    license="https://huggingface.co/sbintuitions/sarashina-embedding-v2-1b/blob/main/LICENSE",
+    max_tokens=8192,
+    reference="https://huggingface.co/sbintuitions/sarashina-embedding-v2-1b",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    adapted_from="sbintuitions/sarashina2.2-1b",
+    superseded_by=None,
+    training_datasets={"NQ", "MrTidyRetrieval"},
+    public_training_code=None,
+    public_training_data="https://huggingface.co/sbintuitions/sarashina-embedding-v1-1b",
+    citation=None,
+    contacts=["Sraym1217", "akiFQC", "lsz05"],
+)
+
+sbintuitions_sarashina_embedding_v1_1b = ModelMeta(
+    loader=sentence_transformers_loader,
+    name="sbintuitions/sarashina-embedding-v1-1b",
+    languages=["jpn-Jpan"],
+    open_weights=True,
+    revision="d060fcd8984075071e7fad81baff035cbb3b6c7e",
+    release_date="2024-11-22",
+    n_parameters=1_224_038_144,
+    memory_usage_mb=4669,
+    embed_dim=1792,
+    license="https://huggingface.co/sbintuitions/sarashina-embedding-v1-1b/blob/main/LICENSE",
+    max_tokens=8192,
+    reference="https://huggingface.co/sbintuitions/sarashina-embedding-v1-1b",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=False,
+    adapted_from="sbintuitions/sarashina2.1-1b",
+    superseded_by="sbintuitions/sarashina-embedding-v2-1b",
+    training_datasets={"NQ", "MrTidyRetrieval"},
+    public_training_code=None,
+    public_training_data="https://huggingface.co/sbintuitions/sarashina-embedding-v1-1b",
+    citation=None,
+    contacts=["akiFQC", "lsz05"],
+)

mteb/models/search_wrappers.py

@@ -250,7 +250,7 @@ class SearchEncoderWrapper:
 
             # get top-k values
             cos_scores_top_k_values, cos_scores_top_k_idx = torch.topk(
-                torch.tensor(scores),
+                torch.as_tensor(scores),
                 min(
                     top_k + 1,
                     len(scores[1]) if len(scores) > 1 else len(scores[-1]),

mteb/tasks/classification/heb/__init__.py

@@ -1,6 +1,11 @@
 from .hebrew_sentiment_analysis import (
     HebrewSentimentAnalysis,
     HebrewSentimentAnalysisV2,
+    HebrewSentimentAnalysisV3,
 )
 
-__all__ = ["HebrewSentimentAnalysis", "HebrewSentimentAnalysisV2"]
+__all__ = [
+    "HebrewSentimentAnalysis",
+    "HebrewSentimentAnalysisV2",
+    "HebrewSentimentAnalysisV3",
+]

mteb/tasks/classification/heb/hebrew_sentiment_analysis.py

@@ -9,7 +9,12 @@ class HebrewSentimentAnalysis(AbsTaskClassification):
             "path": "mteb/HebrewSentimentAnalysis",
             "revision": "03eb0996c8234e0d8cd7206bf4763815deda12ed",
         },
-        description="HebrewSentiment is a data set consists of 12,804 user comments to posts on the official Facebook page of Israel’s president, Mr. Reuven Rivlin. In October 2015, we used the open software application Netvizz (Rieder, 2013) to scrape all the comments to all of the president’s posts in the period of June – August 2014, the first three months of Rivlin’s presidency.2 While the president’s posts aimed at reconciling tensions and called for tolerance and empathy, the sentiment expressed in the comments to the president’s posts was polarized between citizens who warmly thanked the president, and citizens that fiercely critiqued his policy.",
+        description=(
+            "HebrewSentiment is a data set consists of 12,804 user comments to posts on the official Facebook page of Israel’s president, Mr. Reuven Rivlin. "
+            "In October 2015, we used the open software application Netvizz (Rieder, 2013) to scrape all the comments to all of the president’s posts in the period of June – August 2014, "
+            "the first three months of Rivlin’s presidency.2 While the president’s posts aimed at reconciling tensions and called for tolerance and empathy, "
+            "the sentiment expressed in the comments to the president’s posts was polarized between citizens who warmly thanked the president, and citizens that fiercely critiqued his policy. "
+        ),
         reference="https://huggingface.co/datasets/hebrew_sentiment",
         type="Classification",
         category="t2c",
@@ -37,7 +42,7 @@ class HebrewSentimentAnalysis(AbsTaskClassification):
   year = {2018},
 }
 """,
-        superseded_by="HebrewSentimentAnalysis.v2",
+        superseded_by="HebrewSentimentAnalysis.v3",
     )
 
 
@@ -49,7 +54,61 @@ class HebrewSentimentAnalysisV2(AbsTaskClassification):
             "revision": "7ecd049fc8ac0d6f0a0121c8ff9fe44ea5bd935b",
             "name": "morph",
         },
-        description="HebrewSentiment is a data set consists of 12,804 user comments to posts on the official Facebook page of Israel’s president, Mr. Reuven Rivlin. In October 2015, we used the open software application Netvizz (Rieder, 2013) to scrape all the comments to all of the president’s posts in the period of June – August 2014, the first three months of Rivlin’s presidency.2 While the president’s posts aimed at reconciling tensions and called for tolerance and empathy, the sentiment expressed in the comments to the president’s posts was polarized between citizens who warmly thanked the president, and citizens that fiercely critiqued his policy. This version corrects errors found in the original data. For details, see [pull request](https://github.com/embeddings-benchmark/mteb/pull/2900)",
+        description=(
+            "HebrewSentiment is a data set consists of 12,804 user comments to posts on the official Facebook page of Israel’s president, Mr. Reuven Rivlin. "
+            "In October 2015, we used the open software application Netvizz (Rieder, 2013) to scrape all the comments to all of the president’s posts in the period of June – August 2014, "
+            "the first three months of Rivlin’s presidency.2 While the president’s posts aimed at reconciling tensions and called for tolerance and empathy, "
+            "the sentiment expressed in the comments to the president’s posts was polarized between citizens who warmly thanked the president, and citizens that fiercely critiqued his policy. "
+            "This version corrects errors found in the original data. For details, see [pull request](https://github.com/embeddings-benchmark/mteb/pull/2900)"
+        ),
+        reference="https://huggingface.co/datasets/hebrew_sentiment",
+        type="Classification",
+        category="t2c",
+        modalities=["text"],
+        eval_splits=["test"],
+        eval_langs=["heb-Hebr"],
+        main_score="accuracy",
+        date=("2015-10-01", "2015-10-31"),
+        domains=["Reviews", "Written"],
+        task_subtypes=["Sentiment/Hate speech"],
+        license="mit",
+        annotations_creators="expert-annotated",
+        dialect=[],
+        sample_creation="found",
+        bibtex_citation=r"""
+@inproceedings{amram-etal-2018-representations,
+  address = {Santa Fe, New Mexico, USA},
+  author = {Amram, Adam and Ben David, Anat and Tsarfaty, Reut},
+  booktitle = {Proceedings of the 27th International Conference on Computational Linguistics},
+  month = aug,
+  pages = {2242--2252},
+  publisher = {Association for Computational Linguistics},
+  title = {Representations and Architectures in Neural Sentiment Analysis for Morphologically Rich Languages: A Case Study from {M}odern {H}ebrew},
+  url = {https://www.aclweb.org/anthology/C18-1190},
+  year = {2018},
+}
+""",
+        adapted_from=["HebrewSentimentAnalysis"],
+        superseded_by="HebrewSentimentAnalysis.v3",
+    )
+
+
+class HebrewSentimentAnalysisV3(AbsTaskClassification):
+    label_column_name = "labels"
+    metadata = TaskMetadata(
+        name="HebrewSentimentAnalysis.v3",
+        dataset={
+            "path": "mteb/HebrewSentimentAnalysisV4",
+            "revision": "aa0b83c4b16cd28daf7c41ef3402e3ffe9c70c59",
+        },
+        description=(
+            "HebrewSentiment is a data set consists of 12,804 user comments to posts on the official Facebook page of Israel’s president, Mr. Reuven Rivlin. "
+            "In October 2015, we used the open software application Netvizz (Rieder, 2013) to scrape all the comments to all of the president’s posts in the period of June – August 2014, "
+            "the first three months of Rivlin’s presidency.2 While the president’s posts aimed at reconciling tensions and called for tolerance and empathy, "
+            "the sentiment expressed in the comments to the president’s posts was polarized between citizens who warmly thanked the president, and citizens that fiercely critiqued his policy. "
+            "This version corrects texts (took pre-tokenized) [more details in this thread](https://huggingface.co/datasets/mteb/HebrewSentimentAnalysis/discussions/2). "
+            "This version corrects errors found in the original data. For details, see [pull request](https://github.com/embeddings-benchmark/mteb/pull/2900)"
+        ),
         reference="https://huggingface.co/datasets/hebrew_sentiment",
         type="Classification",
         category="t2c",

{mteb-2.3.4.dist-info → mteb-2.3.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mteb
-Version: 2.3.4
+Version: 2.3.6
 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-2.3.4.dist-info → mteb-2.3.6.dist-info}/RECORD

@@ -5,7 +5,7 @@ mteb/_helpful_enum.py,sha256=jh73N1jlcpg7RGz4bj8UpctiMNvqvHpp9wrB7SYEzIU,510
 mteb/_log_once.py,sha256=-tUKzxGQzf2LZSuQXi97oYFXMta1B6GEYXd7BPqssvY,1095
 mteb/_requires_package.py,sha256=eHg_TD9BVZRzNCcQQrUP17d8M1DF_vOd_tVx54AmAnM,3017
 mteb/_set_seed.py,sha256=HPlPRl__Pe6IG-4UgJqTfplcivJ_wA2kaClbXoHQedM,1178
-mteb/cache.py,sha256=77GtlwqHZxqkoCYcu76KCFL4AnHHkH8w-JY3oglMrbc,20102
+mteb/cache.py,sha256=XiFuhjZ2C-o0LgP1YM8g9As_vigJCUNfTrOb9-EiFlM,20177
 mteb/deprecated_evaluator.py,sha256=t13Eluvm5ByVIOqgT7fqiVfLb8Ud3A4bbF2djRfs8iA,26901
 mteb/evaluate.py,sha256=B60CkqRHzkI-3zIfHyocp-YUeWrzeoOvX_RN5vSlGqE,19363
 mteb/filter_tasks.py,sha256=5XE1OYmgDDoJYnXwFf4ma_PIT_Lekzs420sQF_kpCiY,7240
@@ -55,8 +55,8 @@ mteb/benchmarks/__init__.py,sha256=MQEVeli-zLaJ7Xg0z7RhXQwsdmm7Ht_W2Ln0rZo1Szc,2
 mteb/benchmarks/_create_table.py,sha256=OAiR44ynJ2fMzoBmVITQtOTYQzxIu9KUdS_HzlBlAck,20195
 mteb/benchmarks/benchmark.py,sha256=70RlMyyg_wkWTlU_IbfLl-KaqRWXGCKTd8fWe9X-AQE,4173
 mteb/benchmarks/get_benchmark.py,sha256=-n_O-gitRKZi48gJKNgGuI36hsP7yLVSiwulnMHN7Gw,3935
-mteb/benchmarks/benchmarks/__init__.py,sha256=0ySgD14Mu3Y1nJzazR_eUir81ia3x6E23N57SzQNkF0,2150
-mteb/benchmarks/benchmarks/benchmarks.py,sha256=Ob2cHVXwFk328xbV-2ZmUibiVAMtT2RN1ygGgiP6UNQ,92662
+mteb/benchmarks/benchmarks/__init__.py,sha256=Ig5dSFunzI-F-OamruuKJVSstbG3xQNkXCxRY3Bj_Ck,2180
+mteb/benchmarks/benchmarks/benchmarks.py,sha256=qHHmJfisT75VRVoZfPcHhShCG0jY6vSWZEx-D01XxKU,94757
 mteb/benchmarks/benchmarks/rteb_benchmarks.py,sha256=QnCSrTTaBfcRlAQp2Nu81tgv1idMXqiM16Fp2zKJ5Ys,10607
 mteb/cli/__init__.py,sha256=v-csUr3eUZElIvrGB6QGtaIdndDfNWEe9oZchsGsJpg,64
 mteb/cli/_display_tasks.py,sha256=7A06dT9sSoTz6shyMvskPxuc5eHY_H7PGPlROzMP0yw,2196
@@ -252,6 +252,7 @@ mteb/descriptive_stats/Classification/HeadlineClassification.json,sha256=VfTqah7
 mteb/descriptive_stats/Classification/HeadlineClassification.v2.json,sha256=n-KiCmlKXb5QOzNG9QTdjwYR-cRV3Qvn96KOF2so7Cs,2110
 mteb/descriptive_stats/Classification/HebrewSentimentAnalysis.json,sha256=j517xsonPntbr4k5Pa9ftGGIZqzaOLDvPqeGjWngdyI,1659
 mteb/descriptive_stats/Classification/HebrewSentimentAnalysis.v2.json,sha256=FaxgCr_lbS-ppsd_cEp_87kbCRCGqFUDMZv375XGvdk,1658
+mteb/descriptive_stats/Classification/HebrewSentimentAnalysis.v3.json,sha256=75u2ZWek2BT8JQElPqBHvCqTEXbeTwlrXcyQC5NK-wU,1658
 mteb/descriptive_stats/Classification/HinDialectClassification.json,sha256=HK13-QJaWc0uC4wOqhZHKzT6i05_wJhURdUJ6Upq8OQ,4745
 mteb/descriptive_stats/Classification/HindiDiscourseClassification.json,sha256=4XMoJL46RHYKYKdndxD5lZ2b3hGjL2CArx3eRalO8eg,1049
 mteb/descriptive_stats/Classification/HindiDiscourseClassification.v2.json,sha256=MGz2Ntd2JxyHtlP3GhkhbMQ_yH2Eladd4vZSFOme2K8,2088
@@ -1426,7 +1427,7 @@ mteb/languages/programming_languages.py,sha256=zxAakT3OSUnAuTnQ34VyeFIECnNXMlleZ
 mteb/leaderboard/__init__.py,sha256=991roXmtRwEQysV-37hWEzWpkvPgMCGRqZTHR-hm2io,88
 mteb/leaderboard/app.py,sha256=29MxFLKEVT-roULHG5boHmsQVhld1rDGNS94r7MWlz8,33118
 mteb/leaderboard/benchmark_selector.py,sha256=uH66SI0iT1J4_fnebViWa83dQwhPi7toBv7PRL_epDw,7784
-mteb/leaderboard/figures.py,sha256=cfOK82rRf-7sCjyP7GBxh4ezhOIt0OhD0_86mKtzLrg,7530
+mteb/leaderboard/figures.py,sha256=mPO0go_23QEhAm1RJdLiBxPFCoUiA74_ztyl6yimc7k,7553
 mteb/leaderboard/table.py,sha256=6SnrYC5GcBlvVSO6vOk6ObuqtoveBLv3JUuXqdKueG8,8333
 mteb/leaderboard/text_segments.py,sha256=iMIkS04QQjPbT-SkU0x6fOcS8xRbUYevryu9HydipKM,6570
 mteb/models/__init__.py,sha256=ABTuoqiBjBtBWW3LYY7ItBHdylR6jWoy06HH0g6j6fU,910
@@ -1435,7 +1436,7 @@ mteb/models/get_model_meta.py,sha256=GeofphZ8wFtwAHYQipgQlZzxNIFAVFGzo_E2sMzjZTc
 mteb/models/instruct_wrapper.py,sha256=Ty4nfEvioycL_uATkhd0PGuyeB5Xc9xrRd6HOGgb-tc,9005
 mteb/models/model_meta.py,sha256=b-Nel9nX5bJk4cgJnqkBzEKyMY7uXvxlCBSxmmH1Ios,14769
 mteb/models/models_protocols.py,sha256=D2hYWn_UBGMaKtRwBx3u0B0ni6lHJjSzTxX21XFNwIc,8917
-mteb/models/search_wrappers.py,sha256=AcMhjQyKdeitUjnaqgnP3_zTeVSum8rz1sjBRddHUVQ,20328
+mteb/models/search_wrappers.py,sha256=zpCvxUVNQWekyC4Fiz7mvlI0VPdSrFq41A0GrCDvBK4,20331
 mteb/models/sentence_transformer_wrapper.py,sha256=n5CMsM6Lpg_CFHH0NkpJusMsaLUTt-L9vRmFINQ961k,12338
 mteb/models/cache_wrappers/__init__.py,sha256=1w1TnMwulWJSzNkLXjbh5MY3sqgHWc6vUntYn49i9X8,169
 mteb/models/cache_wrappers/cache_backend_protocol.py,sha256=TR7kD7KbN1J4piszIecpegtLZYGy7sRHZt3SDWlImKk,1665
@@ -1494,6 +1495,7 @@ mteb/models/model_implementations/jina_clip.py,sha256=CfiIxbhKspjQajNtObCfGPHOWP
 mteb/models/model_implementations/jina_models.py,sha256=HrHm2Io3g9gHwxU5icAaudy_E8rAVkAAIFSzVYWF-dM,34859
 mteb/models/model_implementations/kalm_models.py,sha256=FmW7Z5Qs6WYBLuKvql3u4IJW36kj4k-Ypah8qTBEBkg,59837
 mteb/models/model_implementations/kennethenevoldsen_models.py,sha256=DF-9nmsewYO9ikZ0kV81ujKGr7Ot36-9iPoxN7KX2mY,2993
+mteb/models/model_implementations/kowshik24_models.py,sha256=HoQpybjhquK2XSnawlq0aiSWFI5M7l6N4DNY4MQ-P10,976
 mteb/models/model_implementations/lens_models.py,sha256=fC7_NB1F8vBAlXD0p0-hALf6eZTPFJwpz57dy71OlwI,1696
 mteb/models/model_implementations/lgai_embedding_models.py,sha256=S83pbfkMH3YUNl4skusgbK-Rn-uLuScQVxgXwegR_N4,2333
 mteb/models/model_implementations/linq_models.py,sha256=EtvUyiNbjU-GJd1kS0Z0gBACkP2pFOjk0KfGMZz4K9Y,1872
@@ -1534,8 +1536,10 @@ mteb/models/model_implementations/rerankers_custom.py,sha256=ro73A9-hHudy3_qIMrh
 mteb/models/model_implementations/rerankers_monot5_based.py,sha256=rxVwzapNnHl4gCw79XVCaTXj3-wbToyj7XVL97tpAF4,34302
 mteb/models/model_implementations/richinfoai_models.py,sha256=llvYa0JUjyOOMbuTgOYoJ2qeqZ5rLHX1ZjZIYlYbdvA,989
 mteb/models/model_implementations/ru_sentence_models.py,sha256=GuZFwbzaooufvSMGNjIsL0DDLrqHjhdSsAQHHZo5H08,40480
+mteb/models/model_implementations/ruri_models.py,sha256=-BTYkZ8dEWZUbGqx3YB5yFSrzMwZtXX7sMUHzrlB8ws,10043
 mteb/models/model_implementations/salesforce_models.py,sha256=KslTK-IKeLvNG-vQir9k6swkaOgjk6eyozm_BOVgTpY,5160
 mteb/models/model_implementations/samilpwc_models.py,sha256=oMwKNwCxoH1jZgCy04oo2oVlBZWu253QMpnEEC6emz8,2021
+mteb/models/model_implementations/sarashina_embedding_models.py,sha256=TSmr2FEX79mJTA9mbEV3meEZYSelGv58Veiw__TTGFM,8415
 mteb/models/model_implementations/searchmap_models.py,sha256=XvVl99emIgnNUCxkTuFQXW6py2R8vgsArfpyHveCugw,1904
 mteb/models/model_implementations/seed_1_6_embedding_models.py,sha256=8J3htEddltyGTydIbnMUudgAV97FdD43-SQKaSA_Iuc,18534
 mteb/models/model_implementations/seed_models.py,sha256=SgK4kPVO6V33G3F1zSq06zSkWarPLEwBt1SWp4TUoVw,14142
@@ -1722,8 +1726,8 @@ mteb/tasks/classification/fra/french_book_reviews.py,sha256=Fsx8UznQVNDNUhcdsTeN
 mteb/tasks/classification/fra/movie_review_sentiment_classification.py,sha256=ov-fbReWP9T_RqhxFtS-gjNaZmbM9J8gQdBQyci5yqU,3290
 mteb/tasks/classification/guj/__init__.py,sha256=HZfimpBCywBLi5VGof_A9Ua6bqtMUoWGRhM1eqAEWKE,186
 mteb/tasks/classification/guj/gujarati_news_classification.py,sha256=VEdbzqlw8b8N8R3TQc275iiCxqGLAMAM8Nf_N7FnUGA,2303
-mteb/tasks/classification/heb/__init__.py,sha256=Wa9-nATstuSXN2OFsO3tF0BZdlk3AzM0g9R1VzZUTc0,171
-mteb/tasks/classification/heb/hebrew_sentiment_analysis.py,sha256=1DPoxWdGAZJB7LwEQWJzpR8KEg2fpbgGXe_cGv1bo5M,4523
+mteb/tasks/classification/heb/__init__.py,sha256=xQNtDxjUsCXQhA_ZYO_4kpBtHWQSBhWhQwAuWcTo6GE,246
+mteb/tasks/classification/heb/hebrew_sentiment_analysis.py,sha256=2wmKwq4Z4YKTG3QieuxmUE56ofiSOV63kV-jh8Zomh4,7281
 mteb/tasks/classification/hin/__init__.py,sha256=KdScMtYYmjsali0InoHP0PKQ8yCbaD-j2tLqc7_lhlo,356
 mteb/tasks/classification/hin/hindi_discourse_classification.py,sha256=HXLJZFEpnI7UjwA-NuadhrCplqaLiSb8Npla_6oXC48,3717
 mteb/tasks/classification/hin/sentiment_analysis_hindi.py,sha256=YIeEuNa2UTr8Jwh_wx15broRuD3NHNansGq7Bl9Vjl0,3101
@@ -2569,9 +2573,9 @@ mteb/types/_metadata.py,sha256=NN-W0S6a5TDV7UkpRx1pyWtGF4TyyCyoPUfHOwdeci8,2290
 mteb/types/_result.py,sha256=CRAUc5IvqI3_9SyXDwv-PWLCXwXdZem9RePeYESRtuw,996
 mteb/types/_string_validators.py,sha256=PY-dYq4E8O50VS3bLYdldPWp400fl_WzUjfVSkNWe8U,523
 mteb/types/statistics.py,sha256=YwJsxTf1eaCI_RE-J37a-gK5wDeGAsmkeZKoZCFihSo,3755
-mteb-2.3.4.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-mteb-2.3.4.dist-info/METADATA,sha256=IbQhbIhtR1KD2NPrrGl9vm0Ua3QQoeGVViqj4sUNRR8,13923
-mteb-2.3.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mteb-2.3.4.dist-info/entry_points.txt,sha256=8IJoEJFKoDHmVnNev-qJ9pp4Ln7_1-ma9QsXnzVCzGU,39
-mteb-2.3.4.dist-info/top_level.txt,sha256=OLVIjcQAlWBz0bdmutKlWHLF42FF0hp4uVAg3ZyiG4U,5
-mteb-2.3.4.dist-info/RECORD,,
+mteb-2.3.6.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+mteb-2.3.6.dist-info/METADATA,sha256=urz0_67bNhVt17rvN3pZdvMFt_mvxI7MFvamWkNoNjM,13923
+mteb-2.3.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+mteb-2.3.6.dist-info/entry_points.txt,sha256=8IJoEJFKoDHmVnNev-qJ9pp4Ln7_1-ma9QsXnzVCzGU,39
+mteb-2.3.6.dist-info/top_level.txt,sha256=OLVIjcQAlWBz0bdmutKlWHLF42FF0hp4uVAg3ZyiG4U,5
+mteb-2.3.6.dist-info/RECORD,,