Summary of Kraken: Inherently Parallel Transformers For Efficient Multi-device Inference, by Rohan Baskar Prabhakar et al.
Kraken: Inherently Parallel Transformers For Efficient Multi-Device Inference
by Rohan Baskar Prabhakar, Hengrui Zhang, David Wentzlaff
First submitted to arxiv on: 14 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes Kraken, an evolution of the standard Transformer architecture designed to improve inference efficiency on multi-device systems. The current autoregressive inference approach is resource-intensive and requires parallelism for efficiency, but this introduces collective communication that is expensive and underutilizes hardware resources. To mitigate this, Kraken introduces a fixed degree of intra-layer model parallelism, allowing collective operations to be overlapped with compute, reducing latency and increasing hardware utilization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Transformer models are being used in applications where low inference latency can make a big difference, like improving the user experience. However, these models require a lot of resources and parallel processing to run efficiently. This paper presents Kraken, an improved version of the standard Transformer model that’s designed to work well on multiple devices at once. By sharing some of the calculations between different parts of the model, Kraken can speed up inference by 35.6% compared to other models. |
Keywords
» Artificial intelligence » Autoregressive » Inference » Transformer
