Accordion: Elastic-Depth Inference via Fixed-Prefix Layer Looping

Yudong Zhang, Wei Ding, Kaili Zheng, Yuchen Ding, Zhenyu Zhang

Trains one mixture-of-experts model with five inference budgets by repeating selected layers. Samples a budget during pretraining and preserves all five paths through instruction tuning and reinforcement learning, allowing callers to trade extra computation for stronger predictions without switching models.

First public: October 23, 2026 · Added to this list: October 26, 2026