httpteapot·6 か月前·議論What do you think of the DeepSeek OCR approach where they say that vision tokens might better compress a document than its pure text representation?https://news.ycombinator.com/item?id=45640594I've spent some time feeding llm with scrapped web pages and I've found that retaining some style information (text size, visibility, decoration image content) is non trivial.
https://news.ycombinator.com/item?id=45640594
I've spent some time feeding llm with scrapped web pages and I've found that retaining some style information (text size, visibility, decoration image content) is non trivial.