@inproceedings{scanandsnapunderstandingtrainingdynamics, title = {Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformer}, year = {2023}, booktitle = {NeurIPS 2023 11}, eprint = {2305.16380}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2305.16380v4}, }