nlpxucan / WizardLM

LLMs build upon Evol Insturct: WizardLM, WizardCoder, WizardMath
9.06k stars 709 forks source link

decoding not stop in WizardMath #213

Open Ricardokevins opened 9 months ago

Ricardokevins commented 9 months ago

In wizard math 7b, i use the model.generate to inference, but model will generate a lot of And to my surprise, it's not the special token.

How to prevent this when using model.generate? ( generating a lot of waste a lot of time)