Load a long CLIP model Parameters ---------- name : str A model name listed by `clip.available_models()`, or the path to a model

Load from CLIP model for fine-tuning Parameters ---------- name : str A model name listed by `clip.available_models()`, or the p

Load the model from the given path. This method is optional and can be overridden by subclasses if needed.

解析 LLM 生成的 "Describe more details about..." 文本， 并自动扩展 position 问题。

解析 LLM 生成的 "Describe more details about..." 文本， 并自动扩展 position 问题。

Read data from file. type: type that you want to read to, such as "datatrame", List[dict], etc.

Read data from Myscale/ClickHouse table.

Register the given object under the the name `obj.__name__`. Can be used as either a decorator or not. See docstring of this

这个文档和签名都会被下面的 __wrapped__ / __signature__ 覆盖成 operator.run 的

Execute OCR analysis (main interface for DataFlowStorage)

Evaluate text answers and compute rewards. Args: storage: DataFlowStorage object input_model_output_key: Colu

Runs the caption generation process, reading from the input file and saving results to output.

Execute the vision dependency filtering pipeline. Process flow: 1. Load data from storage 2. Generate respon

Execute luminance analysis and filtering. Args: storage: DataFlow storage object figure_root: Direct

运行算子 Args: storage: DataFlow存储对象 video_key: 视频路径字段名 Returns: list:

Execute the CLIP score evaluation pipeline. Process flow: 1. Load data from storage 2. Extract images and te

执行分数过滤 Args: storage: DataFlowStorage对象，用于数据读写 input_score_key: 要过滤的分数列名（如 "reward", "accuracy" 等）

Execute the DataTailor evaluation pipeline. Process flow: 1. Load data from storage 2. Extract multi-modal t

Execute the similarity-based filtering pipeline. Process flow: 1. Load data from storage 2. Extract CLIP emb

Execute the filtering pipeline. Process flow: 1. Load data from storage 2. Generate responses using rollout

运行算子 Args: storage: DataFlow存储对象 video_key: 视频路径字段名 Returns: list:

执行数据集过滤 Args: storage: DataFlowStorage对象，用于数据读写 input_video_key: 输入数据在storage中的键名 output

Execute aesthetic scoring and filtering. Args: storage: DataFlow storage object figure_root: Directo

Read dataframe from storage, compute scenes, write dataframe back. Column names can be overridden via run() params.

Execute OCR analysis and filtering. Args: storage: DataFlow storage object figure_root: Directory wh

Execute the multi-modal filtering pipeline. Args: storage: DataFlow storage object input_image_key:

Execute merged caption generation. Args: storage: DataFlow storage object caption_key: Override capt

Process dataframe with video/image CoT QA generation. This operator only generates responses and extracts answers/processes.

Main function to run the operator.

将文本拆分为句子列表

将文本拆分为句子列表

Switch to a different model with potentially different API configuration. Args: model: Model name to switch to

Test function to check the operator registry. This will print all registered operators and their signatures.

Returns the tokenized representation of given input string(s) Parameters ---------- texts : Union[str, List[str]] An input s

Write data to Myscale/ClickHouse table. data: pd.DataFrame or List[dict]，每行是data字段内容（dict）。

将音频数据写入文件或其他存储

将音频数据写入文件或其他存储