Prune-then-Loop Retrofitting Turns Pretrained VLMs into Recurrent Models

A new paper proposes prune-then-loop, retrofitting a pretrained VLM by deleting layers of its language model and looping the remaining ones. The approach saves 23.3% of language parameters while outperforming the original baseline.

2026-10-31 ~ 2026-10-31 · 3 related posts

• Prune-then-loop retrofit turns a pretrained VLM recurrent, beating baseline with 23.3% fewer language params — HildeKuehne · 2026-10-31
• Prune-then-Loop: Turning a Pretrained VLM into a Recurrent Model — HildeKuehne · 2026-10-31
• Recurrent retrofitting cuts VLM language params 23.3% while beating the unpruned baseline — HildeKuehne · 2026-10-31