lexected@alien.topB to LocalLLaMA@poweruser.forumEnglish · 1 year agoExponentially Faster Language Modelling: 40-78x Faster Feedforward for NLU thanks to FFFsarxiv.orgexternal-linkmessage-square4fedilinkarrow-up11arrow-down10cross-posted to: programming@programming.dev
arrow-up11arrow-down1external-linkExponentially Faster Language Modelling: 40-78x Faster Feedforward for NLU thanks to FFFsarxiv.orglexected@alien.topB to LocalLLaMA@poweruser.forumEnglish · 1 year agomessage-square4fedilinkcross-posted to: programming@programming.dev
minus-squarematsu-morak@alien.topBlinkfedilinkEnglisharrow-up1·1 year agoSad part is that we need to train a generative model from scratch to use this one; i.e., we can’t fine-tune current models to use FFF. Hope someone does it soon.
Sad part is that we need to train a generative model from scratch to use this one; i.e., we can’t fine-tune current models to use FFF.
Hope someone does it soon.