*Well, it's just kinda sitting there; maybe you should download it and stick in a glass jar. #DeepSeek
https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf
@bruces that paragraph is just lifted from their paper on Deepseek V3.
You should look at that paper first as it covers the hardware issues, and comes with nice pictures of their solution to problems like "managing cross entropy loss while maintaining causality"
I am glad that this paper is a computing one where such terms have specific meaning -if that phrase ever popped up in a physics paper, I would be scared.
https://github.com/deepseek-ai/DeepSeek-V3/blob/main/DeepSeek_V3.pdf