calvinmccarter / itlumm

Paper

"Look-ups are not (yet) all you need for deep learning inference" was presented at the Sparsity in Neural Networks Workshop 2022. See the arXiv preprint and please cite as follows:

@misc{https://doi.org/10.48550/arxiv.2207.05808,
  doi = {10.48550/ARXIV.2207.05808},
  url = {https://arxiv.org/abs/2207.05808},
  author = {McCarter, Calvin and Dronen, Nicholas},
  title = {Look-ups are not (yet) all you need for deep learning inference},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}

Code

This repo is mostly copypasta from the Bolt repo which contains code for MADDNESS -- see our bolt directory. The maddness-old directory is deprecated and can be safely ignored. In the code, ITLUMM and Pluto refer to the same method; Vingilote refers to an early, inferior version of ITLUMM. The driveit directory contains code related to acceleration of full NNs with replacement of linear layers and fine-tuning. The snn2022 directory contains materials for the SNN2022 paper.