Reward Model Python Packages

pairjudge

Pairwise LLM judges (A/B/tie): budget-aware multi-turn packing, position-bias correction, pseudo-label distillation. Generalized from the 4th-place (gold) solution to Kaggle LMSYS Chatbot Arena.

py-openjudge

OpenJudge: A Unified Framework for Holistic Evaluation and Quality Rewards

bestofn

Best-of-N 1.1 — sample a frozen language model N times and select the best answer. 45.3% to 66.5% on GSM8K with a 0.5B model, no training. Eight models measured, all 25,600 trajectories published.

rewardharness

Self-evolving agentic reward framework for image-editing evaluation — 47.4% on EditReward-Bench from only 100 preference demos, no reward-model training. arXiv 2605.08703.

rewardlint

Your RLVR reward function has a false-positive rate. Find out what it accepts that it shouldn't. Zero dependencies, no GPU, runs in a second.

reward-lens

Measurement instruments for reward models and RL runs. Every reading is either evidence or a refusal that tells you what to fix.

lightrft

LightRFT: Light, Efficient, Omni-modal & Reward-model Driven Reinforcement Fine-Tuning Framework