MAGIC-AI4Med / MedS-Ins

The official codes for "Towards Evaluating and Building Versatile Large Language Models for Medicine"
Other
40 stars 4 forks source link

您训练过程中用的few-shot和测试的few-shot是一样的吗?或者说训练的时候已经强化了它few-shot的推理能力,所以模型测试的结果比较好呢? #2

Closed hxyyyds closed 1 week ago

Henrychur commented 1 month ago

您好!模型在训练过程中以及测试阶段所使用的 few-shot 示例均是通过随机采样得到的,并非相同的示例。