nlpxucan / WizardLM

LLMs build upon Evol Insturct: WizardLM, WizardCoder, WizardMath
9.24k stars 716 forks source link

Do you apply ablation study on wizardcoder with fine tuning with initial code alpaca? #89

Open GanjinZero opened 1 year ago

GanjinZero commented 1 year ago

In ablation study, you show multi round revision works better than single round revision. You also show wizardcoder outperforms Pretrained model starcoder. However, it is unknown how much wizardcoder benefits from code alpaca or multi round prompt revision.

JiayiFu commented 1 year ago

same question.

haorannlp commented 1 year ago

Same question ...