CoQMoE: Co-Designed Quantization and Computation Orchestration for Mixture-of-Experts Vision Transformer on FPGA
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916787763806208 |
|---|---|
| author | Dong, Jiale Wu, Hao Wang, Zihao Lou, Wenqi Zheng, Zhendong Gong, Lei Wang, Chao Zhou, Xuehai |
| author_facet | Dong, Jiale Wu, Hao Wang, Zihao Lou, Wenqi Zheng, Zhendong Gong, Lei Wang, Chao Zhou, Xuehai |
| contents | Vision Transformers (ViTs) exhibit superior performance in computer vision tasks but face deployment challenges on resource-constrained devices due to high computational/memory demands. While Mixture-of-Experts Vision Transformers (MoE-ViTs) mitigate this through a scalable architecture with sub-linear computational growth, their hardware implementation on FPGAs remains constrained by resource limitations. This paper proposes a novel accelerator for efficiently implementing quantized MoE models on FPGAs through two key innovations: (1) A dual-stage quantization scheme combining precision-preserving complex quantizers with hardware-friendly simplified quantizers via scale reparameterization, with only 0.28 $\%$ accuracy loss compared to full precision; (2) A resource-aware accelerator architecture featuring latency-optimized streaming attention kernels and reusable linear operators, effectively balancing performance and resource consumption. Experimental results demonstrate that our accelerator achieves nearly 155 frames per second, a 5.35$\times$ improvement in throughput, and over $80\%$ energy reduction compared to state-of-the-art (SOTA) FPGA MoE accelerators, while maintaining $<1\%$ accuracy loss across vision benchmarks. Our implementation is available at https://github.com/DJ000011/CoQMoE. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_08496 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | CoQMoE: Co-Designed Quantization and Computation Orchestration for Mixture-of-Experts Vision Transformer on FPGA Dong, Jiale Wu, Hao Wang, Zihao Lou, Wenqi Zheng, Zhendong Gong, Lei Wang, Chao Zhou, Xuehai Hardware Architecture Vision Transformers (ViTs) exhibit superior performance in computer vision tasks but face deployment challenges on resource-constrained devices due to high computational/memory demands. While Mixture-of-Experts Vision Transformers (MoE-ViTs) mitigate this through a scalable architecture with sub-linear computational growth, their hardware implementation on FPGAs remains constrained by resource limitations. This paper proposes a novel accelerator for efficiently implementing quantized MoE models on FPGAs through two key innovations: (1) A dual-stage quantization scheme combining precision-preserving complex quantizers with hardware-friendly simplified quantizers via scale reparameterization, with only 0.28 $\%$ accuracy loss compared to full precision; (2) A resource-aware accelerator architecture featuring latency-optimized streaming attention kernels and reusable linear operators, effectively balancing performance and resource consumption. Experimental results demonstrate that our accelerator achieves nearly 155 frames per second, a 5.35$\times$ improvement in throughput, and over $80\%$ energy reduction compared to state-of-the-art (SOTA) FPGA MoE accelerators, while maintaining $<1\%$ accuracy loss across vision benchmarks. Our implementation is available at https://github.com/DJ000011/CoQMoE. |
| title | CoQMoE: Co-Designed Quantization and Computation Orchestration for Mixture-of-Experts Vision Transformer on FPGA |
| topic | Hardware Architecture |
| url | https://arxiv.org/abs/2506.08496 |