Table of Links
Impact statement, and References
Reproducibility statement
All models were trained and evaluated on a computing cluster consisting of 128 AMD EPYC 7502 CPUs, 503 GB of RAM, and 8 NVIDIA RTX A6000 GPUs each with 49 GiB RAM. All MOMENT variants were trained on a single A6000 GPU (with any data or model parallelism). We have made MOMENT-large[6] and the Time Series Pile[7] publicly available on Huggingface. We are working on opensourcing MOMENT-base and MOMENT-small, and our research code public. The latter is currently available anonymously at https://anonymous.4open.science/ r/BETT-773F/README.md. We enlist an exhaustive list of hyper-parameters in App. E to aid reproducibility. We would like to emphasize that all datasets used in this study are publicly available.
Authors:
(1) Mononito Goswami, Auton Lab, Robotics Insititute, Carnegie Mellon University, Pittsburgh, USA ([email protected])
(2) Konrad Szafer, Auton Lab, Robotics Institute, Carnegie Mellon University, Pittsburgh, USA, with equal contribution, order decided using a random generator;
(3) Arjun Choudhry, Auton Lab, Robotics Institute, Carnegie Mellon University, Pittsburgh, USA, with equal contribution, order decided using a random generator;
(4) Yifu Cai, Auton Lab, Robotics Institute, Carnegie Mellon University, Pittsburgh, USA;
(5) Shuo Li, University of Pennsylvania, Philadelphia, USA;
(6) Artur Dubrawski, Auton Lab, Robotics Institute, Carnegie Mellon University, Pittsburgh, USA.
This paper is
[6] https://huggingface.co/AutonLab/ MOMENT-1-large
[7] https://huggingface.co/datasets/ AutonLab/Timeseries-PILE
