Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2505.07203 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913832281047040 |
|---|---|
| author | Du, Kuntai Wang, Bowen Zhang, Chen Cheng, Yiming Lan, Qing Sang, Hejian Cheng, Yihua Yao, Jiayi Liu, Xiaoxuan Qiao, Yifan Stoica, Ion Jiang, Junchen |
| author_facet | Du, Kuntai Wang, Bowen Zhang, Chen Cheng, Yiming Lan, Qing Sang, Hejian Cheng, Yihua Yao, Jiayi Liu, Xiaoxuan Qiao, Yifan Stoica, Ion Jiang, Junchen |
| contents | Besides typical generative applications, like ChatGPT, GitHub Copilot, and Cursor, we observe an emerging trend that LLMs are increasingly used in traditional discriminative tasks, such as recommendation, credit verification, and data labeling. The key characteristic of these emerging use cases is that the LLM generates only a single output token, rather than an arbitrarily long sequence of tokens. We call this prefill-only workload. However, since existing LLM engines assume arbitrary output lengths, they fail to leverage the unique properties of prefill-only workloads. In this paper, we present PrefillOnly, the first LLM inference engine that improves the inference throughput and latency by fully embracing the properties of prefill-only workloads. First, since it generates only one token, PrefillOnly only needs to store the KV cache of only the last computed layer, rather than of all layers. This drastically reduces the GPU memory footprint of LLM inference and allows handling long inputs without using solutions that reduces throughput, such as cross-GPU KV cache parallelization. Second, because the output length is fixed, rather than arbitrary, PrefillOnly can precisely determine the job completion time (JCT) of each prefill-only request before it starts. This enables efficient JCT-aware scheduling policies such as shortest remaining job first. PrefillOnly can process upto 4x larger queries per second without inflating average and P99 latency. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_07203 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | PrefillOnly: An Inference Engine for Prefill-only Workloads in Large Language Model Applications Du, Kuntai Wang, Bowen Zhang, Chen Cheng, Yiming Lan, Qing Sang, Hejian Cheng, Yihua Yao, Jiayi Liu, Xiaoxuan Qiao, Yifan Stoica, Ion Jiang, Junchen Distributed, Parallel, and Cluster Computing Besides typical generative applications, like ChatGPT, GitHub Copilot, and Cursor, we observe an emerging trend that LLMs are increasingly used in traditional discriminative tasks, such as recommendation, credit verification, and data labeling. The key characteristic of these emerging use cases is that the LLM generates only a single output token, rather than an arbitrarily long sequence of tokens. We call this prefill-only workload. However, since existing LLM engines assume arbitrary output lengths, they fail to leverage the unique properties of prefill-only workloads. In this paper, we present PrefillOnly, the first LLM inference engine that improves the inference throughput and latency by fully embracing the properties of prefill-only workloads. First, since it generates only one token, PrefillOnly only needs to store the KV cache of only the last computed layer, rather than of all layers. This drastically reduces the GPU memory footprint of LLM inference and allows handling long inputs without using solutions that reduces throughput, such as cross-GPU KV cache parallelization. Second, because the output length is fixed, rather than arbitrary, PrefillOnly can precisely determine the job completion time (JCT) of each prefill-only request before it starts. This enables efficient JCT-aware scheduling policies such as shortest remaining job first. PrefillOnly can process upto 4x larger queries per second without inflating average and P99 latency. |
| title | PrefillOnly: An Inference Engine for Prefill-only Workloads in Large Language Model Applications |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2505.07203 |