Reward ModelsRLHFProcess Supervision

Reward Models: Outcome vs Process Supervision, Learned vs Rule-Based

A practical taxonomy of outcome models, process models, verifiers, judges, and hybrid reward systems.

TL;DR

Reward systems differ along two independent axes: where feedback is assigned (final outcome or intermediate process) and how it is produced (learned model, deterministic rule, human label, or model judge). Calling all four ?reward models? hides the engineering trade-offs.

A Useful Taxonomy

SignalStrengthMain risk
Deterministic verifierReproducible and cheapNarrow specification; exploitable checker
Learned preference modelHandles subjective qualitiesDistribution shift and proxy optimization
LLM judgeFlexible, little training setupBias, inconsistency, self-preference
Process reward modelBetter local creditExpensive/noisy step labels

Outcome supervision scores the final answer. Process supervision scores intermediate reasoning steps. A process score can improve credit assignment, but only if ?step correctness? is meaningful and labels are trustworthy. It can also reward verbose decompositions or penalize valid unconventional reasoning.

Training a Pairwise Reward Model

Given chosen and rejected responses with scalar outputs (r_w,r_l), a Bradley?Terry objective is -log(sigmoid(r_w-r_l)). Accuracy on held-out pairs is necessary but insufficient: RL actively searches for inputs that maximize the model, including regions absent from its validation set.

import torch.nn.functional as F

def pairwise_reward_loss(chosen_reward, rejected_reward):
    return -F.logsigmoid(chosen_reward - rejected_reward).mean()

Production Design

Version reward components independently, log their raw values before aggregation, test calibration on new policy outputs, and retain adversarial holdouts. Use deterministic RLVR where possible, learned feedback where judgment is genuinely required, and human review for consequential ambiguity.

Do not interpret a weighted sum as a stable objective when components have different variances. Normalize and monitor each channel, then check whether improving the combined score harms any constituent objective.

References

  1. Christiano, P. et al. (2017). Deep Reinforcement Learning from Human Preferences.
  2. Ouyang, L. et al. (2022). Training Language Models to Follow Instructions with Human Feedback.
  3. Lightman, H. et al. (2023). Let’s Verify Step by Step.
  4. Wang, P. et al. (2023). Math-Shepherd.