-
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Authors:
Adam Paszke,
Sam Gross,
Francisco Massa,
Adam Lerer,
James Bradbury,
Gregory Chanan,
Trevor Killeen,
Zeming Lin,
Natalia Gimelshein,
Luca Antiga,
Alban Desmaison,
Andreas Köpf,
Edward Yang,
Zach DeVito,
Martin Raison,
Alykhan Tejani,
Sasank Chilamkurthy,
Benoit Steiner,
Lu Fang,
Junjie Bai,
Soumith Chintala
Abstract:
Deep learning frameworks have often focused on either usability or speed, but not both. PyTorch is a machine learning library that shows that these two goals are in fact compatible: it provides an imperative and Pythonic programming style that supports code as a model, makes debugging easy and is consistent with other popular scientific computing libraries, while remaining efficient and supporting…
▽ More
Deep learning frameworks have often focused on either usability or speed, but not both. PyTorch is a machine learning library that shows that these two goals are in fact compatible: it provides an imperative and Pythonic programming style that supports code as a model, makes debugging easy and is consistent with other popular scientific computing libraries, while remaining efficient and supporting hardware accelerators such as GPUs.
In this paper, we detail the principles that drove the implementation of PyTorch and how they are reflected in its architecture. We emphasize that every aspect of PyTorch is a regular Python program under the full control of its user. We also explain how the careful and pragmatic implementation of the key components of its runtime enables them to work together to achieve compelling performance.
We demonstrate the efficiency of individual subsystems, as well as the overall speed of PyTorch on several common benchmarks.
△ Less
Submitted 3 December, 2019;
originally announced December 2019.
-
Thirty Meter Telescope Site Testing I: Overview
Authors:
M. Schoeck,
S. Els,
R. Riddle,
W. Skidmore,
T. Travouillon,
R. Blum,
E. Bustos,
G. Chanan,
S. G. Djorgovski,
P. Gillett,
B. Gregory,
J. Nelson,
A. Otarola,
J. Seguel,
J. Vasquez,
A. Walker,
D. Walker,
L. Wang
Abstract:
As part of the conceptual and preliminary design processes of the Thirty Meter Telescope (TMT), the TMT site testing team has spent the last five years measuring the atmospheric properties of five candidate mountains in North and South America with an unprecedented array of instrumentation. The site testing period was preceded by several years of analyses selecting the five candidates, Cerros To…
▽ More
As part of the conceptual and preliminary design processes of the Thirty Meter Telescope (TMT), the TMT site testing team has spent the last five years measuring the atmospheric properties of five candidate mountains in North and South America with an unprecedented array of instrumentation. The site testing period was preceded by several years of analyses selecting the five candidates, Cerros Tolar, Armazones and Tolonchar in northern Chile; San Pedro Martir in Baja California, Mexico and the 13 North (13N) site on Mauna Kea, Hawaii. Site testing was concluded by the selection of two remaining sites for further consideration, Armazones and Mauna Kea 13N. It showed that all five candidates are excellent sites for an extremely large astronomical observatory and that none of the sites stands out as the obvious and only logical choice based on its combined properties. This is the first article in a series discussing the TMT site testing project.
△ Less
Submitted 7 April, 2009;
originally announced April 2009.
-
Velocity Profiles in Slowly Sheared Bubble Rafts
Authors:
John Lauridsen,
Greg Chanan,
Michael Dennin
Abstract:
Measurements of average velocity profiles in a bubble raft subjected to slow, steady-shear demonstrate the coexistence between a flowing state and a jammed state similar to that observed for three-dimensional foams and emulsions [Coussot {\it et al,}, Phys. Rev. Lett. {\bf 88}, 218301 (2002)]. For sufficiently slow shear, the flow is generated by nonlinear topological rearrangements. We report o…
▽ More
Measurements of average velocity profiles in a bubble raft subjected to slow, steady-shear demonstrate the coexistence between a flowing state and a jammed state similar to that observed for three-dimensional foams and emulsions [Coussot {\it et al,}, Phys. Rev. Lett. {\bf 88}, 218301 (2002)]. For sufficiently slow shear, the flow is generated by nonlinear topological rearrangements. We report on the connection between this short-time motion of the bubbles and the long-time averages. We find that velocity profiles for individual rearrangement events fluctuate, but a smooth, average velocity is reached after averaging over only a relatively few events.
△ Less
Submitted 5 December, 2003; v1 submitted 26 November, 2003;
originally announced November 2003.
