link
Publish Date
Number
reflection
abstract
一种以skill-mix方式且randomly sampled来作为一种新型evaluation的方式，防止一定程度上的填鸭式学习，且发现有些模型仅仅在单任务上表现好，然而缺乏general-purpose的能力
Status
Done
Type
evaluation
Author
SKILL - MIX : A FLEXIBLE AND EXPANDABLE FAMILY OF EVALUATIONS FOR AI MODELS
Words≈0 | Read Time ≈ 0 min
- Valine