Applies SFT to a model by adding the sparse parameter differences to the model's parameters, and setting the value of the "abs" param

Calculates the sparse difference vector between the current parameter values and the pre-trained values. Args:

Adds a new parameter tensor to the SFT. Args: name: the parameter name, e.g. bert.embeddings.word_embeddings.weight.

Parameters ---------- predicted_indices : ``torch.Tensor``, required. A tensor of head index predictions of shape

Inverse of encode_sparse_tensor.

Compresses a sparse tensor by flattening indices into a single dimension to reduce size of saved SFTs.

Post-processes the predictions of a question-answering model to convert them to answers that are substrings of the original contexts. This is

Removes SFT from a model by subtracting the sparse parameter differences.

BuilderConfig for Masakhaner. Args: **kwargs: keyword arguments forwarded to super.

Returns SplitGenerators.

Post-processes the predictions of a question-answering model with beam search to convert them to answers that are substrings of the original

Prediction/evaluation loop, shared by `evaluate()` and `predict()`. Works both with or without labels.