Ai2 测试 ManiFlow 等模型的泛化能力
In our initial ToyBlocks tests, ManiFlow, π0.5, & MolmoAct 2 generalized poorly to new spatial configurations.
摘要
在初步的 ToyBlocks 测试中，ManiFlow、π0.5 和 MolmoAct 2 对新空间配置的泛化表现不佳。 对照实验让我们能够测试训练和模型设计的改动，然后衡量哪些技术能改善泛化能力。
摘要由模型根据原文撰写，以原文为准。
正文 · 原文
In our initial ToyBlocks tests, ManiFlow, π0.5, & MolmoAct 2 generalized poorly to new spatial configurations.
Controlled experiments let us test changes to training & model design, then measure which techniques improve generalization.
来源：Ai2 · x.com
引用原文
allen_ai. In our initial ToyBlocks tests, ManiFlow, π0.5, & MolmoAct 2 generalized poorly to new spatial configurations.[EB/OL]. (2026-11-06)[访问日期]. https://x.com/allen_ai/status/2118708463194792202.