:param value: Content for this specific node (also the deduction of the children). :param children: The children for this node.

Applies RoPE to ~inputs Args: inputs: the tensor to which RoPE is applied, the dimensions indexed by ~seq_dim indicates the s

Normal attention Args: query: positive tensor of shape (*_q, dim_qk) key: positive tensor of shape (*_k, dim_qk) scale: t

Support for batch prompts inference. Args: inputs (List[str]): A list of strings. max_out_len (int): The maximum leng

Get a suffix of prompt which is no longer than num_token tokens. Args: prompt (str): Input string. num_token (int): T

The function calculates the numerical results for each group based on the configuration in summary_groups, and updates the contents of

Calculate sequence identity between two sequences. This is a simplified implementation for sequences of equal length, compute

Performs invalid character removal and whitespace cleanup on text.

Cleans up the generated code.

Send code to the remote evaluation service using gradio_client and get the results. Args: input_data: Can be one of:

Access the code eval service. Args: file_path (str): The file path to the file to be evaluated. Returns: tup

Combine two frames Args: The following two arguments all have the transition of shape (N, 3) at the first pl

Computes per-residue pLDDT from logits. Code below adopted from https://github.com/deepmind/alphafold/blob/0be2b30b98f0da7aecb973bde0475

Compute sine and cosine tensors Args: tensor: the tensors to apply RoPE to seq_dim: the dimension indices of the spat

Encode the special tokens in the prompt. Now this is left for the future work

Evaluate the response data sample.

Evaluate the response data sample.

评估折叠类型任务，使用与 accuracy4fold_type.py 相同的计算方式

评估功能描述任务，使用 ROUGE-L

通用评估方法

评估多标签任务（EC_number, GO, Keywords）

Evaluate predictions using the configured evaluator.

评估亚细胞定位任务，使用准确率

Format table as a pandas dataframe and pivot so that columns are datasets and rows are models. Args: table (List[List]):

Generate results given a list of inputs. Args: inputs (PromptType): A string or PromptDict. The PromptDict should

d == 3 for equivalent semantics d == 0 for completely different semantics 根据大词林的信息，计算两个词的语义类别的差距

Calculate cross entropy based on given logits, targets and attention_mask for BPC loss calculation. Args: logits (np.ndar

Infer the accelerator Args: device: the device type Returns:

Token dropout scaling This computes the scaling from Rives et al. 2021 Args: mask: the mask indicating the validity of t

Determine if the LLM judge considers the answer correct. Args: llm_detail: The evaluation details from the LLM judge. Re

Get the position of the tokens given Args: shape: the shape of the tensor to be applied with RoPE device: the device on which the

计算基于词性信息的编辑距离cost :param a_pos: 单词a的词性 :param b_pos: 单词b的词性 :return: 替换编辑代价

https://kexue.fm/archives/8823 Args: num_res: [num_chunks] attn_dim Returns:

计算两个单词拼写相似度，分别由字形相似度和字音相似度组成 :param a: 单词a :param b: 单词b，且单词a的长度小于等于b :param pinyin_a: 单词a的拼音 :param pinyin_b