Congrats!
#3
by CyborgPaloma - opened
To be getting so much traction so soon after deepseek's OCR release just goes to show how good this paper is and how useful and needed it still is. Hopefully that is inspiring the confidence in your team that it aught to. Thank you for the apache-2.0!!
Thanks for the kind words!
Yeah, true!
This kind of "playing around" with the tokenizer is such a great idea, all the following matrixes get smaller, leading to such incredible performance gains.