Files and videos

Use read_files to list files or read bytes. Use read_videos when the files are video sources that should be typed as video assets.

pipeline = mdr.read_files(
    "/data/raw/**/*",
    recursive=True,
    file_path_column="path",
    size_column="size",
)

This emits one row per file without reading file contents.

pipeline = mdr.read_files(
    "/data/prompts/*.txt",
    content_column="text",
    decode_fn=lambda data: data.decode("utf-8"),
)

Use content_column only when the contents are small enough to process as row values. For large media, keep paths and let media-aware APIs stream the data.

pipeline = (
    mdr.read_videos("/data/videos/*.mp4", file_path_column="video")
    .to_robot_rows(video_keys=("video",), fps=30.0)
)

read_videos creates a video-typed column, which lets to_robot_rows expose it through row.videos.

• GPU Video Transcoding
• Video Sources
• Media Assets and Reducers