Summary of Trex- Reusing Vision Transformer’s Attention For Efficient Xbar-based Computing, by Abhishek Moitra et al.
TReX- Reusing Vision Transformer’s Attention for Efficient Xbar-based Computing
by Abhishek Moitra, Abhiroop Bhattacharjee, Youngeun Kim, Priyadarshini Panda
First submitted to arxiv on: 22 Aug 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Hardware Architecture (cs.AR)
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 The proposed framework, TReX, optimizes the performance of Vision Transformers (ViTs) in edge-computing scenarios by reusing attention blocks and achieving optimal accuracy-energy-delay-area tradeoffs. Building on previous works that focused on algorithm-hardware co-design and architecture improvements for energy-efficient deployment, TReX addresses the neglected overhead and co-dependence of attention blocks on IMC-implemented ViTs. By optimally choosing transformer encoders for attention reuse, TReX achieves near iso-accuracy performance while meeting user-specified delay requirements. Compared to state-of-the-art token pruning and weight sharing approaches, TReX demonstrates improved accuracy at lower energy consumption levels. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary TReX is a new way to make computer vision models work better on devices that don’t have powerful processors. It does this by reusing parts of the model instead of recalculating everything from scratch each time. This makes the model run faster and use less power, which is important for things like self-driving cars or smart home security cameras that need to process images quickly. The authors tested TReX on a big dataset and found that it can reduce energy consumption by 2.3 times while still keeping the same level of accuracy. |
Keywords
» Artificial intelligence » Attention » Pruning » Token » Transformer
