"""Instruction variants. Picked deterministically per sample id (see schema.pick).""" CAPTION = [ "Describe this image in detail.", "Write a caption for this image.", "What is shown in this image? Describe it.", "Give a short description of this image.", ] ARTICLE = [ "Write the article that goes with these images.", "Write the text of the web page these images come from.", ] OCR_PAGE = [ "Read all the text in this image.", "Transcribe all the text in this document page.", "Extract the text from this image exactly as written.", "What does the text in this image say? Write it out in full.", ] OCR_CROP = [ "Read the text in this image.", "What does the text say?", "Transcribe the text in this image.", ] SCENE_TEXT_BOXES = [ "Detect and read all the text in this image. For each piece of text give the text and its " "bounding box as [x1, y1, x2, y2], normalized to 0-999.", ]