Recirculation Boosts LLM Reasoning Without Any Retraining

A new paper introduces Recirculation, an inference-time architecture enhancement that injects upper-layer activations into lower layers, reducing perplexity and improving reasoning accuracy without any retraining, validated on Gemma3.

2026-08-19 ~ 2026-08-20 · 3 related posts

• No-Training Depth Boost? Recirculation Paper — teortaxesTex · 2026-08-19
• Recirculation: Training-free inference architecture boosts model performance — chaumian · 2026-08-20
• Recirculation: Off-the-shelf models self-modify for instant inference boost — TheGradient · 2026-08-20