pelillian·2 years ago·discussThat’s why we use top p and top k! They limit the probability space to a certain % or number of tokens ordered by likelihood
pelillian·2 years ago·discusshttps://youtube.com/@MikeBoydEasily Mike Boyd!The beauty of his channel is he doesn’t allow himself to use the internet or look anything up. It’s all trial and error, (and the editing is really well done)