Tracing the Representation Geometry of Language Models from Pretraining to Post-training

Re-assign community