Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task

Nikolay Bogoychev; Roman Grundkiewicz; Alham Fikri Aji; Maximiliana Behnke; Kenneth Heafield; Sidharth Kashyap; Emmanouil-Ioannis Farsarakis; Mateusz Chudyk

doi:10.18653/v1/2020.ngt-1.26

Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task

Nikolay Bogoychev, Roman Grundkiewicz, Alham Fikri Aji, Maximiliana Behnke, Kenneth Heafield, Sidharth Kashyap, Emmanouil-Ioannis Farsarakis, Mateusz Chudyk

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract / Description of output

We participated in all tracks of the Workshop on Neural Generation and Translation 2020 Efficiency Shared Task: single-core CPU, multicore CPU, and GPU. At the model level, we use teacher-student training with a variety of student sizes, tie embeddings and sometimes layers, use the Simpler Simple Recurrent Unit, and introduce head pruning. On GPUs, we used 16-bit floating-point tensor cores. On CPUs, we customized 8-bit quantization and multiple processes with affinity for the multicore setting. To reduce model size, we experimented with 4-bit log quantization but use floats at runtime. In the shared task, most of our submissions were Pareto optimal with respect the trade-off between time and quality.

Original language	English
Title of host publication	Proceedings of the Fourth Workshop on Neural Generation and Translation
Place of Publication	Seattle
Publisher	Association for Computational Linguistics (ACL)
Pages	218–224
Number of pages	7
ISBN (Electronic)	978-1-952148-17-0
DOIs	https://doi.org/10.18653/v1/2020.ngt-1.26
Publication status	Published - 10 Jul 2020
Event	The 4th Workshop on Neural Generation and Translation - Online workshop, Seattle, United States Duration: 10 Jul 2020 → 10 Jul 2020 https://sites.google.com/view/wngt20

Workshop

Workshop	The 4th Workshop on Neural Generation and Translation
Abbreviated title	WNGT 2020
Country/Territory	United States
City	Seattle
Period	10/07/20 → 10/07/20
Internet address	https://sites.google.com/view/wngt20

Access to Document

10.18653/v1/2020.ngt-1.26Licence: Creative Commons: Attribution (CC-BY)

Edinburghs Submissions_BOGOYCHEV_DOA11052020_AFVAccepted author manuscript, 192 KBLicence: Creative Commons: Attribution (CC-BY)
Edinburgh’s Submissions_BOGOYCHEV_DOA11052020_VOR_CC-BYFinal published version, 259 KBLicence: Creative Commons: Attribution (CC-BY)
Wngt20 Supplementary dataOther version, 10.1 KBLicence: Creative Commons: Attribution (CC-BY)

https://www.aclweb.org/anthology/2020.ngt-1.26Licence: Creative Commons: Attribution (CC-BY)

User-Focused Marian
Heafield, K.
EU government bodies
1/03/20 → 28/02/22
Project: Research
Browser-based Multilingual Translation
Heafield, K.
EU government bodies
1/01/19 → 30/06/22
Project: Research

Cite this

Bogoychev, N., Grundkiewicz, R., Aji, A. F., Behnke, M., Heafield, K., Kashyap, S., Farsarakis, E.-I., & Chudyk, M. (2020). Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task. In Proceedings of the Fourth Workshop on Neural Generation and Translation (pp. 218–224). Association for Computational Linguistics (ACL). https://doi.org/10.18653/v1/2020.ngt-1.26

@inproceedings{8666728ba696489287a4b4c6caec9115,

title = "Edinburgh{\textquoteright}s Submissions to the 2020 Machine Translation Efficiency Task",

abstract = "We participated in all tracks of the Workshop on Neural Generation and Translation 2020 Efficiency Shared Task: single-core CPU, multicore CPU, and GPU. At the model level, we use teacher-student training with a variety of student sizes, tie embeddings and sometimes layers, use the Simpler Simple Recurrent Unit, and introduce head pruning. On GPUs, we used 16-bit floating-point tensor cores. On CPUs, we customized 8-bit quantization and multiple processes with affinity for the multicore setting. To reduce model size, we experimented with 4-bit log quantization but use floats at runtime. In the shared task, most of our submissions were Pareto optimal with respect the trade-off between time and quality.",

author = "Nikolay Bogoychev and Roman Grundkiewicz and Aji, {Alham Fikri} and Maximiliana Behnke and Kenneth Heafield and Sidharth Kashyap and Emmanouil-Ioannis Farsarakis and Mateusz Chudyk",

year = "2020",

month = jul,

day = "10",

doi = "10.18653/v1/2020.ngt-1.26",

language = "English",

pages = "218–224",

booktitle = "Proceedings of the Fourth Workshop on Neural Generation and Translation",

publisher = "Association for Computational Linguistics (ACL)",

note = "The 4th Workshop on Neural Generation and Translation, WNGT 2020 ; Conference date: 10-07-2020 Through 10-07-2020",

url = "https://sites.google.com/view/wngt20",

}

Bogoychev, N, Grundkiewicz, R, Aji, AF, Behnke, M, Heafield, K, Kashyap, S, Farsarakis, E-I & Chudyk, M 2020, Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task. in Proceedings of the Fourth Workshop on Neural Generation and Translation. Association for Computational Linguistics (ACL), Seattle, pp. 218–224, The 4th Workshop on Neural Generation and Translation, Seattle, Washington, United States, 10/07/20. https://doi.org/10.18653/v1/2020.ngt-1.26

Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task. / Bogoychev, Nikolay; Grundkiewicz, Roman; Aji, Alham Fikri et al.
Proceedings of the Fourth Workshop on Neural Generation and Translation. Seattle: Association for Computational Linguistics (ACL), 2020. p. 218–224.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task

AU - Bogoychev, Nikolay

AU - Grundkiewicz, Roman

AU - Aji, Alham Fikri

AU - Behnke, Maximiliana

AU - Heafield, Kenneth

AU - Kashyap, Sidharth

AU - Farsarakis, Emmanouil-Ioannis

AU - Chudyk, Mateusz

PY - 2020/7/10

Y1 - 2020/7/10

N2 - We participated in all tracks of the Workshop on Neural Generation and Translation 2020 Efficiency Shared Task: single-core CPU, multicore CPU, and GPU. At the model level, we use teacher-student training with a variety of student sizes, tie embeddings and sometimes layers, use the Simpler Simple Recurrent Unit, and introduce head pruning. On GPUs, we used 16-bit floating-point tensor cores. On CPUs, we customized 8-bit quantization and multiple processes with affinity for the multicore setting. To reduce model size, we experimented with 4-bit log quantization but use floats at runtime. In the shared task, most of our submissions were Pareto optimal with respect the trade-off between time and quality.

AB - We participated in all tracks of the Workshop on Neural Generation and Translation 2020 Efficiency Shared Task: single-core CPU, multicore CPU, and GPU. At the model level, we use teacher-student training with a variety of student sizes, tie embeddings and sometimes layers, use the Simpler Simple Recurrent Unit, and introduce head pruning. On GPUs, we used 16-bit floating-point tensor cores. On CPUs, we customized 8-bit quantization and multiple processes with affinity for the multicore setting. To reduce model size, we experimented with 4-bit log quantization but use floats at runtime. In the shared task, most of our submissions were Pareto optimal with respect the trade-off between time and quality.

U2 - 10.18653/v1/2020.ngt-1.26

DO - 10.18653/v1/2020.ngt-1.26

M3 - Conference contribution

SP - 218

EP - 224

BT - Proceedings of the Fourth Workshop on Neural Generation and Translation

PB - Association for Computational Linguistics (ACL)

CY - Seattle

T2 - The 4th Workshop on Neural Generation and Translation

Y2 - 10 July 2020 through 10 July 2020

ER -

Edinburgh’s Submissions to the 2020 Machine Translation Efficiency Task

Abstract / Description of output

Workshop

Access to Document

Fingerprint

Projects

User-Focused Marian

Browser-based Multilingual Translation

Cite this