Follow
Latent Positional Information is in the Self-Attention Variance of Transformer Language Models Without Positional Embeddings. (arXiv:2305.13571v1 [cs.CL])
Latent Positional Information is in the Self-Attention Variance of Transformer Language Models Without Positional Embeddings. (arXiv:2305.13571v1 [cs.CL])