Adaptive Online Replanning with Diffusion Models
Siyuan Zhou, Yilun Du, et al.
NeurIPS 2023
MEGA is a recent transformer-based architecture, which utilizes a linear recurrent operator whose parallel computation, based on the FFT, scales as O(LlogL), with L being the sequence length. We build upon their approach by replacing the linear recurrence with a special temporal convolutional network which permits larger receptive field size with shallower networks, and reduces the computational complexity to O(L). The resulting model is called TCNCA, a Temporal Convolutional Network with Chunked Attention. We evaluate TCNCA on EnWik8 language modeling, long-range-arena (LRA) sequence classification, as well as a synthetic reasoning benchmark associative recall. On EnWik8, TCNCA outperforms MEGA, reaching a lower loss with 1.37×/1.24× faster forward/backward pass during training. The dilated convolutions used in TCNCA are consistently and significantly faster operations than the FFT-based parallelized recurrence in GPUs, making them a scalable candidate for handling very large sequence lengths: they are up to 7.07×/2.86× faster in the forward/backward pass for sequences up to 131 k. Further on LRA, TCNCA achieves, on average, 1.28× speed-up during inference with similar accuracy to what MEGA achieves. On associative recall, we find that even a simplified version of TCNCA, without excessive multiplicative and additive interactions, remains superior or competitive to MEGA on a range of sequence lengths and vocabulary sizes.
Siyuan Zhou, Yilun Du, et al.
NeurIPS 2023
Erik Miehling, Rahul Nair, et al.
NeurIPS 2023
C.A. Micchelli, W.L. Miranker
Journal of the ACM
Kenneth L. Clarkson, Elad Hazan, et al.
Journal of the ACM