← Back to feed

Mix Your Base and RLVR Weights: A Training-Free Knob for Diversity

AI · Studies · 2026-11-20 13:15

Researchers propose a training-free method that interpolates weights between base and RLVR models to restore sample diversity without sacrificing accuracy. This approach optimizes test-time scaling by offering a simple knob to balance precision and output variety.

Source: mix-base-rlvr.pages.dev


Auto-generated summary for reference only. RSS