ReBound: Readout-Aware Boundary Pruning for Efficient Mamba Prefill

Mamba language models make autoregressive decoding efficient, but prefill still processes every prompt token through every layer. We propose \methodname, a training-free readout-aware boundary pruning method that runs the selected Mamba layer at full length and then removes low-impact boundary representations before downstream layers. Its score combines channel-mean $\Delta$, selective-update energy, and proximity to the final autoregressive readout, while a calibration-only KL sweep selects the earliest behaviorally safe pruning layer. At 10% FLOP reduction on Mamba-1 2.8B, \methodname reaches 68.0% average lm-evaluation-harness accuracy, only 0.62 points below baseline, and raises WikiText-103 continuation perplexity by only 0.61%. A complete seven-way score ablation shows that the full $\Delta +E+P$ score is best, and calibration generalizes across architectures, with sharp safe-layer transitions at L44 for Mamba-1 2.8B and L52 for Mamba-2 2.7B. On five LongBench tasks, \methodname preserves long-context quality substantially better than $\Delta$-only DTP on both architectures. Measured hardware performance realizes these savings: at L44 with 50% token keep, latency improves by $1.15\times$ and throughput by $1.157\times$, close to the $1.174\times$ FLOP-based ceiling. These results show that Mamba token compression should preserve the final readout rather than rely on local token importance alone.