Are Vision Tokens all you Need (?) 🧿 This is a fantastic blog from @tuanacelik and @LoganMarkewich explaining the recent DeepSeek OCR paper. As @karpathy pointed out, one of the key insights here is not that “VLMs are good for OCR”, its vision tokens can be a more efficient https://t.co/bgPOcHyfsa https://t.co/s7Arc6BMFR
— Jerry Liu (@jerryjliu0) Oct 27, 2025
from Twitter https://twitter.com/jerryjliu0
October 27, 2025 at 07:25PM
via IFTTT
No comments:
Post a Comment