chunked/nltk_chunking/model_doc_align/chunk_10.txt

In this paper, we leverage a noisy dataset of over one billion image alt-text pairs, obtained without expensive filtering or post-processing steps in the Conceptual Captions dataset.