Follow
Weight subcloning: direct initialization of transformers using larger pretrained ones. (arXiv:2312.09299v1 [cs.LG]) http://arxiv.org/abs/2312.09299
Weight subcloning: direct initialization of transformers using larger pretrained ones. (arXiv:2312.09299v1 [cs.LG]) http://arxiv.org/abs/2312.09299