"Eureka-Moments in Transformers: Multi-Step Tasks Reveal Softmax Induced ..."

David T. Hoffmann et al. (2023)

Details and statistics

DOI: 10.48550/ARXIV.2310.12956

access: open

type: Informal or Other Publication

metadata version: 2023-10-27

a service of  Schloss Dagstuhl - Leibniz Center for Informatics