Build the triplet loss over a batch of embeddings. For each anchor, we get the hardest positive and hardest negative to form a triplet.

Build the triplet loss over a batch of embeddings. For each anchor, we get the hardest positive and hardest negative to form a triplet.

Build the triplet loss over a batch of embeddings. We generate all the valid triplets and average the loss over the positive ones. Arg

Per-language MSE + TranslationEvaluator. `main_score_function` averages translation accuracies only. MSE (`negative_mse * 100`) is on a different

Corpus = deduplicated passages of the split, queries = questions, 1:1 qrels.

Build a dataset of (col_a, col_b, label) pairs with negatives.

Build a dataset of (col_a, col_b, label) pairs with negatives.

Compute the margin loss over the whole batch, from the per-mini-batch embeddings. The labels are unused: the negatives are mined from the oth

Compute the loss over the whole batch, from the per-mini-batch embeddings. When ``with_backward`` is set, back-propagate the loss (chunk by c

Calculate the cross-entropy loss and return it alongside the gradients wrt. the embeddings.

Calculate the cross-entropy loss and return it alongside the gradients wrt. the logits.

Temporarily sets the output_hidden_states to True, runs the model, and then restores the original setting. Use the all_layer_embeddin

Chunked forward for multi-vector models, returning ``(token_embeddings, attention_mask)``. The per-chunk outputs are re-padded to the widest chun

transformers caches ``warning_once`` globally, so without this a warning emitted by one test silences it in every later test. Request the fixture

Turn the inputs from the dataloader into the separate model inputs & the labels.

Turn the inputs from the dataloader into the separate model inputs & the labels. Example:: >>> list(inputs.keys()) [

Compute count vector from sparse embeddings indicating how many samples have non-zero values in each dimension. Args: embeddings: Sp

Compute the Global Orthogonal Regularization terms for a batch of embeddings. The GOR loss encourages embeddings to be well-distribu

Compute the GOR loss from pre-computed embeddings. Args: embeddings: List of embedding tensors, one for each input colum

Compute the CosineSimilarity loss from embeddings. Args: embeddings: List of embeddings labels: Labels indic

Compute the CSRReconstruction loss from embeddings. Args: outputs: List of dictionaries containing embeddings and their

Computes the evaluation metrics for the given model. Args: model (SentenceTransformer): The SentenceTransformer model to

Compute MRR, NDCG, and AP metrics using sklearn

Computes the evaluation metrics in a batched way, by batching all queries and all documents together. Args: model (Sente

Computes the evaluation metrics individually by embedding every (query, positive, negative) tuple individually. Args: mo

Run uniqueness checks concurrently with a bounded thread pool.

Return a dense text model instance for the concrete test class. Subclasses should use ``request.getfixturevalue(...)`` to obtain the

Create a sparse tensor of shape (rows, cols) with num_nonzero values per row.

:param latents: autoencoder latents (shape: [batch, hidden_dim]) :return: reconstructed data (shape: [batch, n_inputs])

Disable all adapters that are attached to the model. This leads to inferring with the base model only.

A context manager that will disable caching in the datasets library.

A context manager that will prevent any logging messages triggered during the body from being processed. Args: highest_level: th

Do forward pass on a minibatch of the input features and return corresponding embeddings.

Embed a mini-batch of inputs.

Do forward pass on all the minibatches of the input features and yield corresponding embeddings.

:param x: input data (shape: [batch, input_dim]) :param latent_slice: slice of latents to compute Example: latent_slice =

Export an optimized ONNX model from a SentenceTransformer, SparseEncoder, CrossEncoder, or MultiVectorEncoder model. The O1-O4 optimizat

Export a quantized OpenVINO model from a SentenceTransformer, SparseEncoder, CrossEncoder, or MultiVectorEncoder model. This function ap

Find a model from hf-internal-testing or tiny-random for the given architecture. If multiple models exist, prefer the one ending with 'Model'

Helper function to format documents with the template.

Format a duration in seconds to a human-readable string, e.g. "23 minutes" or "1.6 hours".

Format the evaluation metrics for the model card. The following keys will be returned: - eval_metrics: A list of dictionaries contain

Helper function to format queries with the template.

Forward pass through the transformer model. Dispatches to the appropriate model method based on the ``modality`` key in ``features``

r""" Creates a StaticEmbedding instance from a distillation process using the `model2vec` package. Args: model_name (str)

Get model_kwargs, config_kwargs, processor_kwargs, processing_kwargs for a given architecture. Returns: Tuple of (model_kwargs, config_kw