原书出处
Designing Large Language Model Applications
optimal proportions. For example, Wikipedia, academic texts, and smaller subsets were upsampled by up to three times in The Pile dataset. More involved techniques like DoReMi and RegMix are also used to calculate the right data mixture. Meta noted that for Llama 3, it empirically arrived at a data mixture where 50% of the tokens are about general knowledge, 25% are about math and reasoning,
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html
EPUB SHA-256:1b4d2347fec2f949fcddfbaf4fad597a1ae20887a096cfae64c59a619b4781bf