Latent Phase-Shift Rollback: Inference-Time Error Correction via Residual Stream Monitoring and KV-Cache Steering

arXiv:2604.18567v1 Announce Type: cross Abstract: Large language models frequently commit unrecoverable reasoning errors mid-generation: once a wrong step is taken, subsequent tokens compound the mistake rather than correct it. We introduce extbf{Latent Phase-Shift Rollback} (LPSR): at each generation step, we monitor the residual stream at a critical layer lcrit, detect abrupt directional reversals (phase shifts) via a cosine-similarity + entropy dual gate, and respond by rolling back the KV-cache and injecting a pre-computed steering vector. No fine-tuning, gradient computation, or additional forward passes are required. LPSR achieves mathbf{44.0%} on MATH-500 with an 8B model versus 28.8% for standard AR (+15.2 pp; McNemar hi^2 = 66.96, p < 10^{-15}). Critically, prompted self-correction, the most natural inference-time baseline, scores only 19.8%, below standard AR; LPSR exceeds it by +24.2 pp (hi^2 = 89.4, p approx 0). LPSR also outperforms Best-of-16 (+7.8 pp) at 5.4imes lower token cost, and surpasses a standard 70B model (35.2%) with 8.75imes fewer parameters at {sim}3imes the token budget. A 32-layer sweep reveals a novel extbf{detection-correction dissociation}: error-detection AUC peaks at layer14 (0.718) but task accuracy peaks at layer16 (44.0% vs. 29.2%), demonstrating that optimal monitoring depth differs for detection and correction.

Related

• YOCO++: Enhancing YOCO with KV Residual Connections for Efficient LLM Inference
• From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
• SeLaR: Selective Latent Reasoning in Large Language Models
• PILOT: Planning via Internalized Latent Optimization Trajectories for Large Language Models

Source: arXiv cs.CL | 2026-04-21