Return the approximate number of batches. .. note:: This is an upper-bound estimate. The actual number of batches yi

Return the number of batches in the sampler.

Delegate assignments of the wrapped loss's hyperparameters to the wrapped loss. Without this, ``loss.scale = 5.0`` would hit the setterless p

A backward hook to backpropagate the cached gradients mini-batch by mini-batch. ``loss_obj`` only needs an ``embed_minibatch_iter(sentence_featur

If the tensor `a` is 1-dimensional, it is unsqueezed to add a batch dimension. Args: a (Tensor): The input tensor. Returns:

Decorator that converts legacy ``pooling_mode_*`` bool kwargs to the new ``pooling_mode`` format. When any of the old boolean keyword arguments (

Compute total/unique/dup counts across query/doc columns.

Returns the first module of this sequential embedder

Get architectures that support any-to-any (multimodal LM).

Get the default modality configuration for the current transformer task. Returns: tuple[MODALITY_CONFIG, str]: A tuple of (modali

Get the default modality configuration for the current transformer task. Returns: tuple[MODALITY_CONFIG, str]: A tuple of (modali

Get architectures that support fill-mask (masked LM).

Detect the model type. Adds Stanford-NLP and PyLate-v3 detection on top of the base behaviour. - If ``config.json`` has ``architectures == ["

Inject load-time defaults for module configs, applied with setdefault priority (saved config values always win): legacy top-level multi-vector

Get architectures that support sequence-classification.

Get architectures that support text-generation (causal LM).

Module-level (pickle-friendly) pool function used by the pickle smoke test.

Return the list of keys to ignore when saving the model.

Returns the last module of this sequential embedder

Convert a SentenceTransformer (and similar) checkpoint into a MultiVectorEncoder. If a final :class:`~sentence_transformers.base.modules.dens

Converts a non-SparseEncoder model into a SparseEncoder by appending a SparseAutoEncoder. If ``model_type`` is ``"SentenceTransformer"``, loa

Creates a simple Transformer + Mean Pooling model and returns the modules, except for CausalLM-based models which use Last Token pool

Build the default module sequence for a fresh MultiVectorEncoder. Three paths: 1. Stanford-NLP ColBERT (``architectures == ["HF_ColBE

Creates a simple transformer-based model and returns the modules. For models with a ForMaskedLM architecture, uses SpladePooling with

Internal working process to encode inputs in multi-process setup. Workers are terminated externally via ``stop_multi_process_pool``.

Internal working process to predict input pairs in a multi-process setup.

Internal working process to encode sentences in multi-process setup. Workers are terminated externally via ``stop_multi_process_pool``.

Worker function for multi-process inference. Must be overridden by subclasses. This is called as the target function in each spawned process

Return the list of modules that should not be split when using model parallelism.