SteerRM: Debiasing Reward Models via Sparse Autoencoders

Abstract

Reward models (RMs) are critical components of alignment pipelines, yet they exhibit biases toward superficial stylistic cues, preferring better-presented responses over semantically superior ones. Existing debiasing methods typically require retraining or architectural modifications, while direct activation suppression degrades performance due to representation entanglement. We propose SteerRM, the first training-free method for debiasing reward models using Sparse Autoencoder (SAE)-based interventions. SteerRM isolates stylistic effects using contrastive paired responses, identifies bias-related SAE features with a strength-stability criterion, and suppresses them at inference time. Across six reward models on RM-Bench, SteerRM improves Hard-split accuracy by 7.3 points on average while preserving overall performance. Results on a Gemma-based reward model and a controlled non-format bias further suggest generalization across RM architectures and bias types. We further find that format-related features are concentrated in shallow layers and transfer across models, revealing shared architecture-level bias encoding patterns. These results show that SAE-based interventions can mitigate reward-model biases without retraining, providing a practical and interpretable solution for alignment pipelines.

Figures (7)

• Figure 1: Overview of SteerRM. Our training-free pipeline consists of three stages: (1) synthesizing paired responses with different surface formats or styles, (2) identifying format-related SAE features, and (3) suppressing these features at inference time.
• Figure 2: SAE reconstruction quality across layers. (Left) Reconstruction MSE. (Middle) Absolute reward difference between original and reconstructed scores. (Right) L0 sparsity measured as the mean number of active features per sample. Layers 0-9 are selected based on these metrics.
• Figure 3: Distribution of top-100 candidate format-related SAE features across layers. Format-related features are concentrated in early layers (1-3), revealing that formatting information is encoded at shallow layers of the Transformer.
• Figure 4: Comparison of SAE reconstruction metrics on the Base Model (Llama-3.1-8B-Base). While both variants achieve comparable MSE, the Residual Stream SAEs (LXR) demonstrate more stable and lower L0 sparsity compared to the MLP Output SAEs (LXM), indicating a more efficient sparse representation.
• Figure 5: Reconstruction quality for SAEs trained on the MLP output (LXM) when evaluated on the Reward Model. Compared to the Residual Stream SAEs (Figure \ref{['fig:layer_selection']}), these models show higher instability in reward preservation (Reward Delta) and sparsity patterns across layers.