Re:Cerebras-GPT vs. LLaMA AI Model Performance Comparison
by nl on Thursday March 30, 2023 @02:09AM
(#35368707)
http://nlg.csie.ntu.edu.tw/~cjchen/papers/eacl2023.pdf (which was pointed to downthread) indicates tokenizing and converting to scientific notation gives a 5% increase in performance on a "comparing numbers" benchmark.
[ Parent
| on HN ]
There are no comments on this comment yet.