Author
Listed:
- Mingxuan Zhang
- Bo Yuan
- Hanzhe Li
- Kangming Xu
Abstract
This paper introduces LLM-CloudComplete, a novel cloud-based system for efficient and scalable code completion leveraging large language models (LLMs). We address the challenges of deploying LLMs for real-time code completion by implementing a distributed inference architecture, adaptive resource allocation, and multi-level caching mechanisms. Our system utilizes a pipeline parallelism technique to distribute LLM layers across multiple GPU nodes, achieving near-linear scaling in throughput. We propose an adaptive resource allocation algorithm using reinforcement learning to optimize GPU utilization under varying workloads. A similarity-based retrieval mechanism is implemented within a three-tier caching system to reduce computational load and improve response times. Additionally, we introduce several latency reduction strategies, including predictive prefetching, incremental completion generation, and sparse attention optimization. Extensive evaluations on diverse programming languages demonstrate that LLM-CloudComplete outperforms existing state-of-the-art code completion systems, achieving a 7.4% improvement in Exact Match accuracy while reducing latency by 76.2% and increasing throughput by 320%. Our ablation studies reveal the significant contributions of each system component to overall performance. LLM-CloudComplete represents a substantial advancement in cloud-based AI-assisted software development, paving the way for more efficient and responsive coding tools. We discuss limitations and future research directions, including privacy-preserving techniques and adaptability to diverse programming paradigms.
Suggested Citation
Download full text from publisher
Corrections
All material on this site has been provided by the respective publishers and authors. You can help correct errors and omissions. When requesting a correction, please mention this item's handle: RePEc:das:njaigs:v:5:y:2024:i:1:p:295-326:id:200. See general information about how to correct material in RePEc.
If you have authored this item and are not yet registered with RePEc, we encourage you to do it here. This allows to link your profile to this item. It also allows you to accept potential citations to this item that we are uncertain about.
We have no bibliographic references for this item. You can help adding them by using this form .
If you know of missing items citing this one, you can help us creating those links by adding the relevant references in the same way as above, for each refering item. If you are a registered author of this item, you may also want to check the "citations" tab in your RePEc Author Service profile, as there may be some citations waiting for confirmation.
For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: Open Knowledge (email available below). General contact details of provider: https://newjaigs.com/index.php/JAIGS/ .
Please note that corrections may take a couple of weeks to filter through
the various RePEc services.