mteb 2.7.11__py3-none-any.whl → 2.7.13__py3-none-any.whl

mteb/models/model_implementations/querit_models.py

@@ -0,0 +1,245 @@
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING, Any
+
+import torch
+from tqdm.auto import tqdm
+
+from mteb.models.model_meta import ModelMeta
+
+from .rerankers_custom import RerankerWrapper
+
+if TYPE_CHECKING:
+    from torch.utils.data import DataLoader
+
+    from mteb.abstasks.task_metadata import TaskMetadata
+    from mteb.types import BatchedInput, PromptType
+
+logger = logging.getLogger(__name__)
+
+
+class QueritWrapper(RerankerWrapper):
+    """
+    Multi-GPU / multi-process reranker wrapper for mteb.mteb evaluation.
+    Supports flattening all query-passage pairs without explicit grouping.
+    """
+
+    def __init__(
+        self,
+        model_name: str,
+        **kwargs: Any,
+    ) -> None:
+        super().__init__(model_name, **kwargs)
+        from transformers import AutoModel, AutoTokenizer
+
+        if not self.device:
+            self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        model_args = {}
+        if self.fp_options:
+            model_args["torch_dtype"] = self.fp_options
+        self.model = AutoModel.from_pretrained(
+            model_name, trust_remote_code=True, **model_args
+        )
+        logger.info(f"Using model {model_name}")
+
+        self.model.to(self.device)
+        self.tokenizer = AutoTokenizer.from_pretrained(
+            model_name, trust_remote_code=True
+        )
+        if "[CLS]" not in self.tokenizer.get_vocab():
+            raise ValueError("Tokenizer missing required special token '[CLS]'")
+        self.cls_token_id = self.tokenizer.convert_tokens_to_ids("[CLS]")
+        self.pad_token_id = self.tokenizer.pad_token_id or 0
+
+        self.max_length = (
+            min(kwargs.get("max_length", 4096), self.tokenizer.model_max_length) - 1
+        )  # sometimes it's a v large number/max int
+        logger.info(f"Using max_length of {self.max_length}, 1 token for [CLS]")
+        self.model.eval()
+
+    def process_inputs(
+        self,
+        pairs: list[str],
+    ) -> dict[str, torch.Tensor]:
+        """
+        Encode a batch of (query, document) pairs:
+        - Concatenate prompt + Query + Content
+        - Append [CLS] at the end
+        - Left-pad to max_length
+        - Generate custom attention mask based on block types
+        """
+        # Construct input texts
+        enc = self.tokenizer(
+            pairs,
+            add_special_tokens=False,
+            truncation=True,
+            max_length=self.max_length,
+            padding=False,
+        )
+
+        input_ids_list: list[list[int]] = []
+        attn_mask_list: list[torch.Tensor] = []
+
+        for ids in enc["input_ids"]:
+            # Append [CLS] token
+            ids = ids + [self.cls_token_id]
+            block_types = [1] * (len(ids) - 1) + [2]  # content + CLS
+
+            # Pad or truncate
+            if len(ids) < self.max_length:
+                pad_len = self.max_length - len(ids)
+                ids = [self.pad_token_id] * pad_len + ids
+                block_types = [0] * pad_len + block_types
+            else:
+                ids = ids[-self.max_length :]
+                block_types = block_types[-self.max_length :]
+
+            attn = self.compute_mask_content_cls(block_types)
+            input_ids_list.append(ids)
+            attn_mask_list.append(attn)
+
+        input_ids = torch.tensor(input_ids_list, dtype=torch.long, device=self.device)
+        attention_mask = torch.stack(attn_mask_list, dim=0).to(self.device)
+
+        return {"input_ids": input_ids, "attention_mask": attention_mask}
+
+    @torch.inference_mode()
+    def predict(
+        self,
+        inputs1: DataLoader[BatchedInput],
+        inputs2: DataLoader[BatchedInput],
+        *,
+        task_metadata: TaskMetadata,
+        hf_split: str,
+        hf_subset: str,
+        prompt_type: PromptType | None = None,
+        **kwargs: Any,
+    ) -> list[float]:
+        """
+        Predict relevance scores for query-passage pairs.
+        Supports both single-process and multi-process/multi-GPU modes.
+        """
+        # Flatten all pairs from mteb.mteb DataLoaders
+        queries = [text for batch in inputs1 for text in batch["text"]]
+        passages = [text for batch in inputs2 for text in batch["text"]]
+
+        instructions = None
+        if "instruction" in inputs2.dataset.features:
+            instructions = [text for batch in inputs1 for text in batch["instruction"]]
+
+        num_pairs = len(queries)
+        if num_pairs == 0:
+            return []
+        final_scores: list[float] = []
+
+        batch_size = kwargs.get("batch_size", self.batch_size)
+        with tqdm(total=num_pairs, desc="Scoring", ncols=100) as pbar:
+            for start in range(0, num_pairs, batch_size):
+                end = min(start + batch_size, num_pairs)
+                batch_q = queries[start:end]
+                batch_d = passages[start:end]
+
+                batch_instructions = (
+                    instructions[start:end]
+                    if instructions is not None
+                    else [None] * len(batch_q)
+                )
+                pairs = [
+                    self.format_instruction(instr, query, doc)
+                    for instr, query, doc in zip(batch_instructions, batch_q, batch_d)
+                ]
+                enc = self.process_inputs(pairs)
+                out = self.model(**enc)
+                scores = out["score"].squeeze(-1).detach().float().cpu().tolist()
+
+                if not isinstance(scores, list):
+                    scores = [scores]
+
+                final_scores.extend(scores)
+                pbar.update(len(scores))
+
+        return final_scores
+
+    @staticmethod
+    def format_instruction(instruction: str | None, query: str, doc: str) -> str:
+        if instruction is None:
+            output = f"Judge whether the Content meets the requirements based on the Query. Query: {query}; Content: {doc}"
+        else:
+            output = f"{instruction} Query: {query}; Content: {doc}"
+        return output
+
+    @staticmethod
+    def compute_mask_content_cls(block_types: list[int]) -> torch.Tensor:
+        """
+        Create custom attention mask based on token block types:
+        - 0: padding   → ignored
+        - 1: content   → causal attention to previous content only
+        - 2: [CLS]     → causal attention to all non-padding tokens
+
+        Args:
+            block_types: List of token types for one sequence
+
+        Returns:
+            [1, seq_len, seq_len] boolean attention mask (True = allowed to attend)
+        """
+        pos = torch.tensor(block_types, dtype=torch.long)
+        n = pos.shape[0]
+        if n == 0:
+            return torch.empty((0, 0), dtype=torch.bool, device=pos.device)
+
+        row_types = pos.view(n, 1)
+        col_types = pos.view(1, n)
+
+        row_idx = torch.arange(n, device=pos.device).view(n, 1)
+        col_idx = torch.arange(n, device=pos.device).view(1, n)
+        causal_mask = col_idx <= row_idx
+
+        # Content tokens only attend to previous content
+        mask_content = (row_types == 1) & (col_types == 1) & causal_mask
+
+        # [CLS] attends to all non-pad tokens (causal)
+        mask_cls = (row_types == 2) & (col_types != 0) & causal_mask
+
+        type_mask = mask_content | mask_cls
+        return type_mask.unsqueeze(0)
+
+
+querit_reranker_training_data = {
+    "MIRACLRanking",  # https://huggingface.co/datasets/mteb/MIRACLReranking
+    "MrTidyRetrieval",  # https://huggingface.co/datasets/mteb/mrtidy
+    "ruri-v3-dataset-reranker",  # https://huggingface.co/datasets/cl-nagoya/ruri-v3-dataset-reranker
+    "MultiLongDocReranking",  # https://huggingface.co/datasets/Shitao/MLDR
+    "MindSmallReranking",  # https://huggingface.co/datasets/mteb/MindSmallReranking
+    "MSMARCO",  # https://huggingface.co/datasets/mteb/msmarco
+    "CQADupStack",  # https://huggingface.co/datasets/mteb/cqadupstack-*
+    "AskUbuntuDupQuestions",  # https://github.com/taolei87/askubuntu & The corpus and queries that overlap with mteb/askubuntudupquestions-reranking have been removed.
+    "T2Reranking",  # https://huggingface.co/datasets/THUIR/T2Ranking & The corpus and queries that overlap with mteb/T2Reranking have been removed.
+}
+
+model_meta = ModelMeta(
+    loader=QueritWrapper,
+    loader_kwargs={
+        "fp_options": "bfloat16",
+    },
+    name="Querit/Querit",
+    model_type=["cross-encoder"],
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="5ad2649cc4defb7e1361262260e9a781f14b08bc",
+    release_date="2026-01-24",
+    n_parameters=4919636992,
+    n_embedding_parameters=131907584,
+    embed_dim=1024,
+    memory_usage_mb=9383.0,
+    max_tokens=4096,
+    reference="https://huggingface.co/Querit/Querit",
+    similarity_fn_name=None,
+    training_datasets=querit_reranker_training_data,
+    license="apache-2.0",
+    framework=["PyTorch"],
+    use_instructions=None,
+    public_training_code=None,
+    public_training_data=None,
+    citation=None,
+)

mteb/tasks/bitext_mining/multilingual/bible_nlp_bitext_mining.py

@@ -914,7 +914,7 @@ class BibleNLPBitextMining(AbsTaskBitextMining):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         # Convert to standard format
         for lang in self.hf_subsets:
             l1, l2 = (l.split("_")[0] for l in lang.split("-"))

mteb/tasks/classification/ben/bengali_document_classification.py

@@ -43,7 +43,7 @@ Islam, Tanvir},
         superseded_by="BengaliDocumentClassification.v2",
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"article": "text", "category": "label"}
         )
@@ -92,7 +92,7 @@ Islam, Tanvir},
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.stratified_subsampling(
             self.dataset, seed=self.seed, splits=["test"]
         )

mteb/tasks/classification/ces/czech_product_review_sentiment_classification.py

@@ -46,7 +46,7 @@ Montoyo, Andres},
     )
     samples_per_label = 16
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"comment": "text", "rating_str": "label"}
         )
@@ -99,7 +99,7 @@ Montoyo, Andres},
     )
     samples_per_label = 16
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.stratified_subsampling(
             self.dataset, seed=self.seed, splits=["test"]
         )

mteb/tasks/classification/ces/czech_so_me_sentiment_classification.py

@@ -46,7 +46,7 @@ Montoyo, Andres},
     )
     samples_per_label = 16
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"comment": "text", "sentiment_int": "label"}
         )

mteb/tasks/classification/multilingual/hin_dialect_classification.py

@@ -60,7 +60,7 @@ class HinDialectClassification(AbsTaskClassification):
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"folksong": "text", "language": "label"}
         )

mteb/tasks/classification/multilingual/indic_lang_classification.py

@@ -137,6 +137,6 @@ Okazaki, Naoaki},
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.remove_columns(["language", "script"])
         self.dataset = self.dataset.rename_columns({"native sentence": "text"})

mteb/tasks/classification/multilingual/indic_sentiment_classification.py

@@ -52,7 +52,7 @@ class IndicSentimentClassification(AbsTaskClassification):
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         label_map = {"Negative": 0, "Positive": 1}
         # Convert to standard format
         for lang in self.hf_subsets:

mteb/tasks/classification/multilingual/language_classification.py

@@ -66,7 +66,7 @@ in Natural Language Processing},
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns({"labels": "label"})
         self.dataset = self.stratified_subsampling(
             self.dataset, seed=self.seed, splits=["test"]

mteb/tasks/classification/multilingual/south_african_lang_classification.py

@@ -49,7 +49,7 @@ class SouthAfricanLangClassification(AbsTaskClassification):
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {" text": "text", "lang_id": "label"}
         )

mteb/tasks/classification/slk/slovak_movie_review_sentiment_classification.py

@@ -35,7 +35,7 @@ class SlovakMovieReviewSentimentClassification(AbsTaskClassification):
         superseded_by="SlovakMovieReviewSentimentClassification.v2",
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns({"comment": "text"})
 
         self.dataset = self.stratified_subsampling(
@@ -76,7 +76,7 @@ class SlovakMovieReviewSentimentClassificationV2(AbsTaskClassification):
         adapted_from=["SlovakMovieReviewSentimentClassification"],
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.stratified_subsampling(
             self.dataset, seed=self.seed, splits=["test"]
         )

mteb/tasks/classification/swa/swahili_news_classification.py

@@ -37,7 +37,7 @@ class SwahiliNewsClassification(AbsTaskClassification):
         superseded_by="SwahiliNewsClassification.v2",
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"content": "text", "category": "label"}
         )
@@ -81,7 +81,7 @@ class SwahiliNewsClassificationV2(AbsTaskClassification):
         adapted_from=["SwahiliNewsClassification"],
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.stratified_subsampling(
             self.dataset, seed=self.seed, splits=["train"]
         )

mteb/tasks/clustering/deu/ten_k_gnad_clustering_p2p.py

@@ -63,7 +63,7 @@ class TenKGnadClusteringP2PFast(AbsTaskClustering):
         adapted_from=["TenKGnadClusteringP2P"],
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         ds = _convert_to_fast(
             self.dataset, self.input_column_name, self.label_column_name, self.seed
         )

mteb/tasks/clustering/deu/ten_k_gnad_clustering_s2s.py

@@ -63,7 +63,7 @@ class TenKGnadClusteringS2SFast(AbsTaskClustering):
         adapted_from=["TenKGnadClusteringS2S"],
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         ds = _convert_to_fast(
             self.dataset, self.input_column_name, self.label_column_name, self.seed
         )

mteb/tasks/clustering/nob/vg_hierarchical_clustering.py

@@ -45,7 +45,7 @@ class VGHierarchicalClusteringP2P(AbsTaskClustering):
         prompt="Identify the categories (e.g. sports) of given articles in Norwegian",
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"article": "sentences", "classes": "labels"}
         )
@@ -92,7 +92,7 @@ class VGHierarchicalClusteringS2S(AbsTaskClustering):
         prompt="Identify the categories (e.g. sports) of given articles in Norwegian",
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {"ingress": "sentences", "classes": "labels"}
         )

mteb/tasks/multilabel_classification/nld/covid_disinformation_nl_multi_label_classification.py

@@ -66,7 +66,7 @@ Yih, Scott Wen-tau},
         },
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         labels = [
             "q2_label",
             "q3_label",

mteb/tasks/pair_classification/multilingual/pub_chem_wiki_pair_classification.py

@@ -60,7 +60,7 @@ class PubChemWikiPairClassification(AbsTaskPairClassification):
 """,
     )
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         _dataset = {}
         for lang in self.hf_subsets:
             _dataset[lang] = {}

mteb/tasks/retrieval/code/code_rag.py

@@ -59,7 +59,7 @@ class CodeRAGProgrammingSolutionsRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = Dict[doc_id, Dict[str, str]] #id => dict with document data like title and text
@@ -116,7 +116,7 @@ class CodeRAGOnlineTutorialsRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = Dict[doc_id, Dict[str, str]] #id => dict with document data like title and text
@@ -176,7 +176,7 @@ class CodeRAGLibraryDocumentationSolutionsRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = Dict[doc_id, Dict[str, str]] #id => dict with document data like title and text
@@ -233,7 +233,7 @@ class CodeRAGStackoverflowPostsRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = Dict[doc_id, Dict[str, str]] #id => dict with document data like title and text

mteb/tasks/retrieval/dan/dan_fever_retrieval.py

@@ -55,7 +55,7 @@ Derczynski, Leon},
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = dict[doc_id, dict[str, str]] #id => dict with document data like title and text

mteb/tasks/retrieval/dan/tv2_nordretrieval.py

@@ -68,7 +68,7 @@ Piperidis, Stelios},
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = dict[doc_id, dict[str, str]] #id => dict with document data like title and text

mteb/tasks/retrieval/dan/twitter_hjerne_retrieval.py

@@ -44,7 +44,7 @@ class TwitterHjerneRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = dict[doc_id, dict[str, str]] #id => dict with document data like title and text

mteb/tasks/retrieval/nob/norquad.py

@@ -58,7 +58,7 @@ Fishel, Mark},
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = dict[doc_id, dict[str, str]] #id => dict with document data like title and text

mteb/tasks/retrieval/nob/snl_retrieval.py

@@ -45,7 +45,7 @@ class SNLRetrieval(AbsTaskRetrieval):
         self.dataset_transform()
         self.data_loaded = True
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         """And transform to a retrieval dataset, which have the following attributes
 
         self.corpus = dict[doc_id, dict[str, str]] #id => dict with document data like title and text

mteb/tasks/sts/multilingual/sem_rel24_sts.py

@@ -66,6 +66,6 @@ Seid Muhie Yimam and Saif M. Mohammad},
     min_score = 0
     max_score = 1
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         for lang, subset in self.dataset.items():
             self.dataset[lang] = subset.rename_column("label", "score")

mteb/tasks/sts/multilingual/sts_benchmark_multilingual_sts.py

@@ -56,6 +56,6 @@ class STSBenchmarkMultilingualSTS(AbsTaskSTS):
     min_score = 0
     max_score = 5
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         for lang, subset in self.dataset.items():
             self.dataset[lang] = subset.rename_column("similarity_score", "score")

mteb/tasks/sts/por/assin2_sts.py

@@ -39,7 +39,7 @@ class Assin2STS(AbsTaskSTS):
     min_score = 1
     max_score = 5
 
-    def dataset_transform(self) -> None:
+    def dataset_transform(self, num_proc: int = 1, **kwargs) -> None:
         self.dataset = self.dataset.rename_columns(
             {
                 "premise": "sentence1",

{mteb-2.7.11.dist-info → mteb-2.7.13.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mteb
-Version: 2.7.11
+Version: 2.7.13
 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>
@@ -32,8 +32,6 @@ Requires-Dist: rich>=0.0.0
 Requires-Dist: pytrec-eval-terrier>=0.5.6
 Requires-Dist: pydantic>=2.0.0
 Requires-Dist: polars>=0.20.22
-Requires-Dist: torch; python_full_version < "3.14"
-Requires-Dist: torch>=2.9.0; python_full_version >= "3.14"
 Provides-Extra: image
 Requires-Dist: torchvision>0.2.1; extra == "image"
 Requires-Dist: transformers[torch-vision,vision]; extra == "image"