Disable deepspeed ulysses (sequence parallelism) if it is enabled

Encode a given input tensor with the specified sample rate at the given bandwidth. The RVQ encode method sets the appropriate number of quanti

Encode a content from a local file to base64 format.

Extract the generation prompt and reference answer from the input tokens. For example: Input Text = '<|begin_of_text|><|start_header_id|>use

Args: hidden_states (`torch.Tensor` of shape `(batch_size, seq_len, hidden_size)`): Hidden states from the LLM co

r""" Args: input_features (`torch.LongTensor` of shape `(batch_size, feature_size, sequence_length)`): Float value

Args: hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` attention_mask (

Forward pass for the Higgs-Audio model. Args: input_ids (:obj:`torch.LongTensor`): The input ids of the prompt. I

Forward pass for the split embedding wrapper. :param input_ids: Tensor of shape [batch_size, seq_len] with indices in [0..original_vo

Args: x (Tensor): Float tensor variable with the shape (B, C, T). Returns: Tensor: Float tensor variable wit

Args: x (Tensor): Float tensor variable with the shape (B, C, T). Returns: Tensor: Float tensor variable wit

Residual vector quantization on the given input tensor. Args: x (torch.Tensor): Input tensor. sample_rate (int): Sampl

Model forward pass Parameters ---------- audio_data : Tensor[B x 1 x T] Audio data to encode sample_rate

Quantized the input tensor using a fixed codebook and returns the corresponding codebook vectors Parameters ----------

Quantized the input tensor using a fixed set of `n` codebooks and returns the corresponding codebook vectors Parameters ------

Freeze the final text head

Given the unquantized latents, reconstruct the continuous representation after quantization. Parameters ---------- la

Convert full-width punctuation to half-width in a given string.

The generate function in huggingface generally follows these steps: for sample_step in 1, 2, 3, 4, 5, ... ...

Generate audio from a chatml sample. Args: chat_ml_sample: A chatml sample. max_new_tokens: The maximum numbe

Merges a list of ChatMLDatasetSample instances, inserting eos_token_id and ignore_index between them, and adjusting offsets for audio_ids_start and au

Count the number of parameters per component in the model. HiggsAudio has the following main components: audio_tower: For mapping

Slice the inputs to create chuncks per the sequence parallel rank. This is used for the context parallel training. Args: sp_size (`i