Speculative Decoding Surprisingly Excels on Sparse MoE Models

References

1. Ranjan et al. — Why Speculative Decoding Works Better Than Predicted on Sparse MoE Models (2026)
2. Leviathan, Kalman, Matias — Fast Inference from Transformers via Speculative Decoding (2022; ICML 2023)
3. Chen et al. — Accelerating LLM Decoding with Speculative Sampling (2023)
4. Huang et al. — MoESD (2025)
5. Li et al. — EAGLE-3 (2025)
6. Yao et al. — Exploiting Inter-Layer Expert Affinity for MoE Inference (2024)
7. Eliseev, Mazur — Fast Inference of MoE Language Models with Offloading (2023)
8. Dai et al. — DeepSeekMoE (2024)
9. Jiang et al. — Mixtral of Experts (2024)
10. Deming & Stephan (1940); Csiszár (1975); Jaynes (1957); Cuturi — Sinkhorn Distances (2013): iterative proportional fitting / max-entropy background
11. Williams, Waterman, Patterson — Roofline (2009)
12. Pope et al. — Efficiently Scaling Transformer Inference (2022)
13. Gale et al. — MegaBlocks (2023)
14. Luo et al. — Dissecting the NVIDIA Hopper Architecture (2025)
15. Saxena et al. — Utility-driven Speculative Decoding for MoE (2025)
16. Sadhukhan et al. — MagicDec (2025)
17. He et al. — ExpertFlow (2026); Olson et al. — Probing Semantic Routing (2025); Bershatsky, Oseledets — Spatial Structure of MoE (2025); Ludziejewski et al. — Scaling Laws for Fine-Grained MoE (2024)