Timezone: »

Terra: Imperative-Symbolic Co-Execution of Imperative Deep Learning Programs
Taebum Kim · Eunji Jeong · Geon-Woo Kim · Yunmo Koo · Sehoon Kim · Gyeongin Yu · Byung-Gon Chun

Thu Dec 09 08:30 AM -- 10:00 AM (PST) @

Imperative programming allows users to implement their deep neural networks (DNNs) easily and has become an essential part of recent deep learning (DL) frameworks. Recently, several systems have been proposed to combine the usability of imperative programming with the optimized performance of symbolic graph execution. Such systems convert imperative Python DL programs to optimized symbolic graphs and execute them. However, they cannot fully support the usability of imperative programming. For example, if an imperative DL program contains a Python feature with no corresponding symbolic representation (e.g., third-party library calls or unsupported dynamic control flows) they fail to execute the program. To overcome this limitation, we propose Terra, an imperative-symbolic co-execution system that can handle any imperative DL programs while achieving the optimized performance of symbolic graph execution. To achieve this, Terra builds a symbolic graph by decoupling DL operations from Python features. Then, Terra conducts the imperative execution to support all Python features, while delegating the decoupled operations to the symbolic execution. We evaluated Terra’s performance improvement and coverage with ten imperative DL programs for several DNN architectures. The results show that Terra can speed up the execution of all ten imperative DL programs, whereas AutoGraph, one of the state-of-the-art systems, fails to execute five of them.

Author Information

Taebum Kim (Seoul National University)
Eunji Jeong (Seoul National University)
Geon-Woo Kim (Seoul National University)
Yunmo Koo (Seoul National University)
Sehoon Kim (University of California Berkeley)
Gyeongin Yu (Seoul National University)
Byung-Gon Chun (Seoul National University)

More from the Same Authors

  • 2023 : SPEED: Speculative Pipelined Execution for Efficient Decoding »
    Coleman Hooper · Sehoon Kim · Hiva Mohammadzadeh · Hasan Genc · Kurt Keutzer · Amir Gholami · Sophia Shao
  • 2023 Poster: Speculative Decoding with Big Little Decoder »
    Sehoon Kim · Karttikeya Mangalam · Suhong Moon · Jitendra Malik · Michael Mahoney · Amir Gholami · Kurt Keutzer
  • 2022 Poster: A Fast Post-Training Pruning Framework for Transformers »
    Woosuk Kwon · Sehoon Kim · Michael Mahoney · Joseph Hassoun · Kurt Keutzer · Amir Gholami
  • 2022 Poster: Squeezeformer: An Efficient Transformer for Automatic Speech Recognition »
    Sehoon Kim · Amir Gholami · Albert Shaw · Nicholas Lee · Karttikeya Mangalam · Jitendra Malik · Michael Mahoney · Kurt Keutzer
  • 2019 Poster: Knowledge Extraction with No Observable Data »
    Jaemin Yoo · Minyong Cho · Taebum Kim · U Kang
  • 2018 : Posters (all accepted papers) + Break »
    Jianyu Wang · Denis Gudovskiy · Ziheng Jiang · Michael Kaufmann · Andreea Anghel · James Bradbury · Nikolas Ioannou · Nitin Agrawal · Emma Tosch · Gyeongin Yu · Keno Fischer · Jarrett Revels · Giuseppe Siracusano · Yaoqing Yang · Jeff Johnson · Yang You · Hector Yuen · Chris Ying · Honglei Liu · Nikoli Dryden · Xiangxi Mo · Yangzihao Wang · Amit Juneja · Micah Smith · Qian Yu · pramod gupta · Deepak Narayanan · Keshav Santhanam · Tim Capes · Abdul Dakkak · Norman Mu · Ke Deng · Liam Li · Joao Carreira · Luis Remis · Deepti Raghavan · Una-May O'Reilly · Amanpreet Singh · Mahmoud (Mido) Assran · Eugene Wu · Eytan Bakshy · Jinliang Wei · Michael Innes · Viral Shah · Haibin Lin · Conrad Sanderson · Ryan Curtin · Marcus Edel