Yifei Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Interests
Yifei Li has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Bole Zhou
Independent, China
2 shared publications - Jiejing Zhang
Alibaba Group, China
2 shared publications - Xuechao Wei
Peking University, Beijing, Beijing, CHINA
2 shared publications - Yingda Chen
ModelScope Team, Alibaba Group Inc., hangzhou, Zhejiang, China
2 shared publications - Yinghan Li
Alibaba Group, China
2 shared publications
Latest Publications
- Open Access
- 30 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 537-548
RadiK: Scalable and Optimized GPU-Parallel Radix Top-K Selection
Top-k selection, which identifies the largest or smallest k elements from a data set, is a fundamental operation in data-intensive domains such as databases and deep learning, so its scalability and efficiency are critical for these high-performance systems. However, previous studies on its efficient GPU implementation are mostly merge-based and rely heavily on fast but size-limited on-chip memory, thereby limiting scalability with a restricted upper bound on k. This work introduces RadiK, a scalable and optimized GPU-parallel radix top-k selection that supports significantly larger k values than existing methods without compromising efficiency, regardless of input length and batch size. RadiK incorporates a novel optimization framework tailored for high memory bandwidth and resource utilization, achieving up to 2.5 × speedup over the prior art for non-batch queries and up to 4.8 × speedup for batch queries. In addition, we propose an adaptive scaling technique that strengthens robustness, which further provides up to 2.7 × speedup on highly adversarial input distributions.
- 30 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 537-548
Open AccessRadiK: Scalable and Optimized GPU-Parallel Radix Top-K Selection
Top-k selection, which identifies the largest or smallest k elements from a data set, is a fundamental operation in data-intensive domains such as databases and deep learning, so its scalability and efficiency are critical for these high-performance systems. However, previous studies on its efficient GPU implementation are mostly merge-based and rely heavily on fast but size-limited on-chip memory, thereby limiting scalability with a restricted upper bound on k. This work introduces RadiK, a scalable and optimized GPU-parallel radix top-k selection that supports significantly larger k values than existing methods without compromising efficiency, regardless of input length and batch size. RadiK incorporates a novel optimization framework tailored for high memory bandwidth and resource utilization, achieving up to 2.5 × speedup over the prior art for non-batch queries and up to 4.8 × speedup for batch queries. In addition, we propose an adaptive scaling technique that strengthens robustness, which further provides up to 2.7 × speedup on highly adversarial input distributions.
- Open Access
- 20 February 2024
- Published by Association for Computing Machinery (ACM)
- p. 472-474
POSTER: RadiK: Scalable Radix Top-K Selection on GPUs
- 20 February 2024
- Published by Association for Computing Machinery (ACM)
- p. 472-474
Open AccessPOSTER: RadiK: Scalable Radix Top-K Selection on GPUs