Adds a new message to the chat buffer. Removes the oldest prompt-response pair if the buffer exceeds the maximum allowed size.

Return keys and values of file key.

Return values of file key.

Extend probabilities and states with more encoded chunks. Args: x (torch.Tensor): The extended encoder output feature

Make mask tensor containing indices of padded part. Args: lengths (LongTensor or List): Batch of lengths (B,). xs (Tensor, option

r"""Update margin and related trigonometric values during training.

Search new tokens for running hypotheses and encoded speech x. Args: running_hyps (List[Hypothesis]): Running hypotheses on beam

Change the device of object recursively

argmax of frame activations Args: torch.Tensor hs_pad: 3d tensor (B, Tmax, eprojs) Returns: torch.Tensor: arg

softmax of frame activations Args: Tensor hs_pad: 3d tensor (B, Tmax, eprojs) Returns: torch.Tensor: softmax

Expand input mel-spectrogram based on the predicted duration. Args: batch (Tensor): Input tensor (T, dim). predicted

Get command line arguments.

Instantiate registered module. registry: Dictionary mapping names to functions or target paths (e.g. {'model': 'models.SequenceModel'

Make mask tensor containing indices of non-padded part. Args: lengths (LongTensor or List): Batch of lengths (B,). xs (Tensor, op

Change object to gatherable in torch.nn.DataParallel recursively The difference from to_device() is changing to torch.Tensor if float or int

Close up this step and reset state for the next step

Repeat module N times. Args: N (int): Number of repeat time. fn (Callable): Function to generate module. layer_drop_rate

Build model from the files. This method is used for inference or fine-tuning. Args: config_file: The yaml file saved whe

A helper function to instantiate Tokenizer

Get the default values of the input function. Examples: >>> def func(a, b=3): pass >>> get_default_kwargs(func) {'b': 3}

Return iterator over (id, text) pairs.

Build Text2Text instance from the pretrained model. Args: model_tag (Optional[str]): Model tag of the pretrained models.

log_softmax of frame activations Args: Tensor hs_pad: 3d tensor (B, Tmax, eprojs) Returns: torch.Tensor: log

Computes the log softmax denominator of the input activation tensor and stores the result in denom. Args: acts: Activati

Build DataLoader using iterable dataset

Initialize weights of a neural network module. Parameters are initialized using the given method or distribution. Custom initialization rout

Reserved for future development of another Trainer

Run inference. Args: text (Tensor): Input text index tensor (T_text,). feats (Tensor): Feature tensor (T_feats, aux_c

Perform training. This method performs the main process of training.

Get specified ID state from decoder hidden states. Args: states: Decoder hidden states. ((N, B, D_dec), (N, B, D_dec))

Fixture to create a YAML file under tmp_path and return its path. Usage: path = write_yaml("filename.yaml", content_as_str)

Compute the sum of log probability from the activation tensor and its denominator. Args: denom: Tensor of shape [B, T, U] flattened.

Send tensor into the device of the module. Args: m (torch.nn.Module): Torch module. x (Tensor): Torch tensor. Returns:

Score new token batch. Args: ys (torch.Tensor): torch.int64 prefix tokens (n_batch, ylen). states (List[Any]): Scorer

Run encoding. Args: x (Tensor): Input codes (T_code, N_stream). Returns: Tensor: Generated waveform (T_wav,)

str_or_none. Examples: >>> import argparse >>> parser = argparse.ArgumentParser() >>> _ = parser.add_argument('--foo', ty

Configure single-optimizer-path or named multi-path optimizers. This includes the paired scheduler configuration. Single-optimizer-p

Load an YAML config with recursive `_self_` merging via Hydra-style `defaults`. This function recursively loads and merges dependent YAML files s

Perform padding for the list of tensors. Args: xs (List): List of Tensors [(T_1, `*`), (T_2, `*`), ..., (T_B, `*`)]. pad_value (f

Helper function to build dataclass from 'args'.

Load pretrained model and adapt it for multimodal parallel processing. Args: pretrained_model_name_or_path: HF model path

Return iterator over IDs.

Step one time step as a recurrent model. Intended to be used during validation. u: (B H) state: (B H N) Returns: out

Create mask for subsequent steps (size, size). :param int size: size of mask :param str device: "cpu" or "cuda" or torch.Tensor.device :p