Sample data according to rank/world_size/num_workers Args: data(List): input data list Returns:

Precomputes the frequency cis.

处理带标签的字符串，将标签替换为对应的内容 Args: s: Returns: <tag> <tag>xxxx

处理带标签的字符串，将标签替换为对应的内容 Args: s: Returns: <tag> <tag>xxxx

处理带标签的字符串，提取标签并返回标签和去除标签后的内容 Args: s: 带标签的字符串，格式如 "<tag>XXXX" Returns: 元组 (标签, 去除标签后的内容)，例如 ("<tag>", "XXXX")

处理带标签的字符串，提取标签并返回标签和去除标签后的内容 Args: s: 带标签的字符串，格式如 "<tag>XXXX" Returns: 元组 (标签, 去除标签后的内容)，例如 ("<tag>", "XXXX")

Symbols in comments: B: batch_size. D: latent_dim. C: num_latent_classes per group G: num of codebook groups. Args:

[cntime] = read_cntime(file_or_fd) Reads single kaldi 'Confusion Network time info', in binary format: C++ type: vector<tuple<float,float> >.

[post] = read_post(file_or_fd) Reads single kaldi 'Posterior' in binary format. The 'Posterior' is C++ type 'vector<vector<tuple<int,float> >

generator(key,vec<vec<int,float>>) = read_post_ark(file) Returns generator of (key,posterior) tuples, read from ark file. file_or_fd : ark, gzi

generator(key,vec) = read_vec_int_ark(file_or_fd) Create generator of (key,vector<int>) tuples, which reads from the ark file/stream. file_or_f

r"""Refer to https://github.com/Outsider565/LoRA-GA/blob/ c185846309ea9012d0bcd46ebd30347dda1c592c/run_exp.py#L67 Reinitialize the lora model

Compute relative positinal encoding. Args: x (torch.Tensor): Input tensor (batch, time, size). zero_triu (bool): If tr

Export interface for c++ call, return right_context of the model

Sample data according to rank/world_size/num_workers Args: data(List): input data list Returns:

Used in ONNX export.

Static batch the data by `batch_size` Args: data: Iterable[{key, feat, label}] batch_size: batch size Retur

Static batch the data by `batch_size` Args: data: Iterable[{key, feat, label}] batch_size: batch size Retur

Static batch the data by `batch_size` Args: data: Iterable[{key, feat, label}] batch_size: batch size Retur

Static batch the data by `batch_size` Args: data: Iterable[{key, feat, label}] batch_size: batch size Retur

Read biasing list from the biasing list address, tokenize and convert it into token id

Train one epoch

write_mat(f, m, key='') Write a binary kaldi matrix to filename or stream. Supports 32bit and 64bit floats. Arguments: file_or_fd : filename o

在每个时间步处理logits，对不符合条件的token设置极小的概率。 参数： input_ids (torch.Tensor): 当前输入的token ID序列 logits (torch.Tensor): 当前时间步的logit

speech_token_num: int, speech token 的数量（token_id in [0, speech_token_num) 视为 speech token） repeat_times: int, 普通 speech token 的最大允许连续

初始化OSUM_chat_LogitsProcessor。 参数： allowed_tokens (list): 允许出现在当前时间步的token的ID列表 sequence_to_match (list): 用来判断当前时间步允许