search

nlpxucan / WizardLM

LLMs build upon Evol Insturct: WizardLM, WizardCoder, WizardMath
9.19k stars 712 forks source link

OOM at the half way finetuning #138

Open yy9996 opened 1 year ago

yy9996 commented 1 year ago

training break down at 1.57 epoch, why? and how to avoid it?