tpls/repro-reward-models-its

Repro - On the Power of (Approximate) Reward Models for Inference-Time Scaling: Sequential Monte Carlo and Beyond

An open experiment logbook, published with Trackio.