判断 serving 是否为 API 类型

Check if CUDA is available for PyTorch.

加载模型和相关的工具函数

Load one or multiple diffusion pipelines and assign each to a separate GPU.

Merge captions from multiple clips into text format for LLM input. Args: dataframe: DataFrame containing clip information and ca

非极大值抑制 NMS 算法，去除重叠过多的框

解析模型生成的 BBox 文本 (x1, y1), (x2, y2)

Parse reasoning text into structured format.

Parse reasoning text into structured format.

Parse a version string into a tuple. Args: version_str (str): The version string. Returns: tuple[int | str]: The version info

消费 video_clips 列表；仅切割 filtered == False 的片段；返回成功 DataFrame 并写 CSV。 如果 filtered 列不存在，会自动添加并设为 False（不过滤）。

解析单条消息中的音频信息

解析单条消息中的音频信息

解析单条消息中的多模态信息

解析单条消息中的多模态信息

Compute greedy matching based on cosine similarity. Args: - :param: `ref_embedding` (torch.Tensor): embeddings of

去除儿化音词中的儿: 他女儿在那边儿 -> 他女儿在那边

r""" Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwa

Execute the complete long video cotqa pipeline. Args: storage: DataFlow storage object input_video_k

Execute the complete video processing pipeline. Args: storage: DataFlow storage object input_video_k

参数： - storage: DataFlowStorage - output_answer_key: 输出答案列名 - **input_keys: 模板字段名 -> DataFrame 列名 例如：descr

Reads dataframe -> cleans text -> parses -> writes back -> returns output_key

参数： - storage: DataFlowStorage - input_image_key / input_video_key: 存放图片/视频路径的列名（只允许其一存在） - output_answer_key: 输出答案列名

Main function to run the wrapper.

Execute the complete video processing pipeline. Args: storage: DataFlow storage object input_video_k

Execute the complete long video cotqa pipeline. Args: storage: DataFlow storage object input_video_k

Execute the video scene clip generation pipeline. Args: storage: DataFlow storage object input_video

Set data to be written later.

Set the file name prefix for cache files.

Given a text, normalize it by changing to lower case, removing punctuations, removing words that only contain digits and removing extra spaces

校验MyScaleDBStorage实例的关键参数有效性： - pipeline_id, input_task_id, output_task_id 必须非空，否则抛出异常。 - page_size, page_num 若未设置则赋默认值（page_

Compute greedy matching based on cosine similarity. Args: - :param: `ref_embedding` (torch.Tensor): embeddings of

Calculate Word Error Rate.

r""" Function invoked when calling the pipeline for generation. Args: image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarr