Saves a model and its configuration files to a directory, so that it can be loaded again. Args: path (str): Path on disk

Uploads all elements of this model to a HuggingFace Hub repository, creating it if it doesn't exist. Args: repo_id (str)

Build a batched transform that resolves ID columns to their values via a join against lookup datasets. Useful for IR datasets that store query / d

Add hard negatives to a dataset of (anchor, positive) pairs to create (anchor, positive, negative) triplets or (anchor, positive, negative_1,

Infer the modality of a single input sample by inspecting its type/structure. Pure type-based detection, does not require a processor or tokenize

Parse inputs and group by modality. Analyzes a list of inputs to detect their modality (text, image, audio, video, message) and group

Returns True if the Huggingface datasets library is available.

Send a PyTorch batch (i.e., a dictionary of string keys to Tensors) to a device (e.g. "cpu", "cuda", "mps"). Args: batch (Dict[str,

Computes the MaxSim (late-interaction) score between two collections of multi-vector embeddings. For each query in ``a`` and document in ``b

Create a BaseModelCardData instance with common defaults for testing.

Generate the Python usage code snippet for the model card. Returns the code block (including \\`\\`\\` delimiters) showing how to use this mo

Decode a sparse embedding into (token, weight) pairs sorted by descending weight. Args: embeddings (torch.Tensor): Spars

Create a BaseDataCollator with a simple preprocess_fn that returns fixed tensors.

Gives a full name (package_name.class_name) for a class / object in Python. Will be used to load the correct classes from JSON files Arg

Computes the pairwise MaxSim (late-interaction) score between each query-document pair. For each ``i``, computes the MaxSim score between ``

Get the output embedding dimension from the transformer model. Returns: int: The hidden dimension size of the model's embeddings.

Disable every dropout, so that gradient-equivalence tests are deterministic. transformers<5 keeps the SDPA attention dropout as a plain float att

Create a small dummy PIL image.

Pad and concatenate homogeneous per-column feature batches into one flat batch. Running ``N`` columns through the model one by one means ``N`` fo

Guard against a vacuous comparison: an empty or all-zero gradient dict passes any assert_close loop.

Save the configuration of the module to a JSON file. Args: output_path (str): The path to the directory where the config

Patch ``model_card.VideoDecoder`` so ``_MockVideoDecoder`` instances pass ``isinstance`` checks.

Convert a batch of modality-specific inputs into the unified message format. Args: modality: The modality key (string) or tuple o

Quantizes embeddings to a lower precision. This can be used to reduce the memory footprint and increase the speed of similarity search. The s

Create a mini-batch from sentence features, handling padded, flattened, and VLM inputs. With padded inputs, this slices along the batch dimension

Compute the ``(begin, end)`` sequence ranges that split a batch into mini-batches. If ``mini_batch_num_tokens`` is None, every range spans ``mini

Given a dataset, compute the following: * Dataset Size * Dataset Columns * Dataset Stats - Strings: min,

Embed each input column, sharing one merged forward pass across columns when possible. Returns one ``sentence_embedding`` tensor per column. When

Check if the input represents a single example or a batch of examples. A list is the batch axis, with one exception: a conversation

Loads a file from a local or remote location. Args: model_name_or_path (str): The model name or path. filename (str): The na

Preprocesses the inputs for the model. Args: inputs (Sequence[SingleInput | PairInput]): A list of inputs to be preproce

Preprocess inputs into model-ready features. Args: inputs: List of inputs. Can contain strings, dicts with modality keys, PIL ima

Save non-text items in :attr:`usage_examples` as files in an ``assets/`` subdirectory. After saving, :attr:`usage_examples_display` is set wi

Starts a multi-process pool to infer with several independent processes. This method is recommended if you want to predict on multip

Stops all processes started with start_multi_process_pool. Args: pool (Dict[str, object]): A dictionary containing the i

Force :func:`merge_feature_batches` to refuse for the duration of the block. For wrapper losses that call an inner loss several times with the *s

Get torch.device from module, assuming that the whole module has one device. In case there are no PyTorch parameters, fall back to CP

Check if a string is an image URL, file path, or data URI.

Returns True if we have the required dependencies for training Sentence Transformers models, i.e. Huggingface datasets and Huggingface accele

Compute the similarity between two collections of embeddings. The output will be a matrix with the similarity scores between all embe

XTR (eXtendable Token Retrieval) contrastive scoring with global top-k token retrieval. For each query token, the top-k matches are selected glob

Converts the input `a` to a PyTorch tensor if it is not already a tensor. Lists are stacked into one tensor: a list of sparse tensors keeps i

Check if a sample is a non-text pair (2-element tuple/list with at least one non-string element). Text pairs ``(str, str)`` are handled natively

Override :attr:`usage_examples` with multimodal inputs when the model supports non-text modalities. Respects the distinction between models t

Verifies that the installed packages satisfy the version requirements declared by a model. Requirements that can't be interpreted (e.g. an i

Return the backend used for inference, which can be one of "torch", "onnx", or "openvino". Returns: str: The backend used for inf

Why the chat template cannot carry a ``query``/``document`` pair, or None when it can. Decided by rendering probe pairs and diffing the outpu

Readable config-dict rendering of a loss's scoring callable: objects exposing ``get_config_dict`` (e.g. configured metric classes) and :func:`func

Format a dataset example value for the model card examples table.

Check whether all messages in a batch contain only text content. Works with both flat format (``{"content": "hello"}``) and structured format

Convert a pair of inputs to query/document message format. Each element of the pair is wrapped in a message with role ``"query"`` (first elem

This function performs by default a cosine similarity search between a list of query embeddings and a list of corpus embeddings. It can be u

Property to get the tokenizer that is used by this model

Returns the number of dimensions in the output of :meth:`SentenceTransformer.encode`. Returns: Optional[int]: The number

Infer the modality of a batch of input samples. If all samples share the same modality, that modality is returned. If the batch contains mixe

Check if a string is a video URL or file path.

Function for Fast Community Detection. Finds in the embeddings all communities, i.e. embeddings that are close (closer than threshold).

Creates a Router model specifically for query and document modules, allowing convenient usage via `model.encode_query` and `model.enc

Infer the ``(n_patches_x, n_patches_y)`` = ``(n_cols, n_rows)`` image-patch grid that ``model``'s processor produces for an image of ``image_size`

Download a URL to a local file with a progress bar. The content is streamed in chunks and first written to a temporary ``"<path>_part"`` file

Resolve a module class reference to a class object. For class refs in the ``sentence_transformers.*`` namespace, this imports directly via

Loads the subfolder path for a given model name or path. Args: model_name_or_path (str): The name or path of the model. subf

A utility function to load the PyTorch weights of a model from a checkpoint. The checkpoint can be either a local directory or a mode

Calculate sparsity statistics for the given embeddings, including the mean number of active (non-zero) dimensions and the mean sparsi

Duck-typed stand-in for ``MultiVectorEncoder``: ``get_n_patches`` only reads ``model.processor``.