1 paper
Yu-Zhe Liu, Shengda Liu, Jianyong Pi +1
We propose a topological model for capability changes during large-model training. The model represents knowledge states and preregistered capability probes as subcomplexes of a fi…