Jinhui Yuan

Jinhui Yuan

Auto-generated profile, based on publicly available information

Jinhui Yuan

OneFlow Research

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 May 2024 in IEEE Transactions on Parallel and Distributed Systems
Verified
Journal Article
Closed access
JCJinfan Chen
Jinfan Chen
SLSHIGANG LI
SHIGANG LI
RGRan Guo
Ran Guo
JYJinhui Yuan
Jinhui Yuan
THTorsten Hoefler
Torsten Hoefler
Recent advances in deep learning are driven by the growing scale of computation, data, and models. However, efficiently training large-scale models on distributed systems requires an intricate combination of data, operator, and pipeline parallelism, which exerts heavy burden on machine learning practitioners. To this end, we propose AutoDDL, a distributed training framework that automatically explores and exploits new parallelization schemes with near-optimal bandwidth cost. AutoDDL facilitates the description and implementation of different schemes by utilizing OneFlow's Split , Broadcast , and Partial Sum (SBP) abstraction. AutoDDL is equipped with an analytical performance model combined with a customized Coordinate Descent algorithm, which significantly reduces the scheme searching overhead. We conduct evaluations on Multi-Node-Single-GPU and Multi-Node-Multi-GPU machines using different models, including VGG and Transformer. Compared to the expert-optimized implementations, AutoDDL reduces the end-to-end training time by up to 31.1% and 10% for Transformer and up to 17.7% and 71.5% for VGG on the two parallel systems, respectively.

30 March 2025
Verified
Conference Paper
Open access
LPLichen Pan
Lichen Pan
JLJuncheng Liu
Juncheng Liu
YFYongquan Fu
Yongquan Fu
JYJinhui Yuan
Jinhui Yuan
RZRongkai Zhang
Rongkai Zhang
ZXZhen Xiao
Zhen Xiao

23 September 2022
Verified
Conference Paper
Closed access
ZHZhou Hongwei
Zhou Hongwei
HXHuang Xiaojie
Huang Xiaojie
KZKe Zhipeng
Ke Zhipeng
ZYZhang Yuchen
Zhang Yuchen
YJYuan Jinhui
Yuan Jinhui
Most of the existing text similarity algorithms aim to improve the accuracy, but this introduces a high overhead because of scanning repeatedly the text to collect the necessary features. In order to achieve a good balance between accuracy and performance overhead, this paper proposes a novel method based on multi-algorithm collaboration, which we call HyAlg, is the abbreviation of Hybrid Algorithm. Hy Alg consists of two stages which are general similarity determination and structure similarity determination. The former uses the general features of the text to quickly complete the similarity determination, and initially eliminates a large number of dissimilar texts, thus reducing the time cost. On the basis of the former filtering results, the latter further determines the similarity from the perspective of text structure to ensure the accuracy of similarity determination. Our experimental and analysis show that Hy Alg is able to effectively reduce the performance cost of the algorithm while ensuring the accuracy.