PP-OCRv6
Collection
From 1.5M to 34.5M Parameters, Surpassing Billion-Scale VLMs on OCR Tasks • 19 items • Updated • 78
I see at least one dataset on the huggingface repo that you seem to have ignored for two months: https://huggingface.co/spaces/hf-audio/open_asr_leaderboard/discussions/57
there's more on the github, including a suggestion to use more accurate versions of the existing voxpopuli and earnings22: https://github.com/huggingface/open_asr_leaderboard/issues/153