inference-time-steering

Tag

Cards List
#inference-time-steering

LARA: Lightweight Adapters in the Residual Stream for Composable Adaptation and Alignment

arXiv cs.LG ↗ · 2026-08-03 Cached

LARA is a method for efficient adaptation that adds low-rank corrections to a frozen model's residual stream instead of modifying weights, matching LoRA's performance while enabling composable behaviors and inference-time steering.

0 favorites 0 likes
#inference-time-steering

RL^2-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models

Hugging Face Daily Papers ↗ · 2026-07-30 Cached

This paper introduces RL^2, an adaptive inference-time steering framework for Vision-Language-Action models that uses offline RL on latent representations to compose action flows, activating steering only when failure is predicted. It achieves up to +17.3% success rate improvements on SIMPLER and PolaRiS benchmarks and demonstrates real-world transfer.

0 favorites 0 likes
#inference-time-steering

GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs

arXiv cs.CL ↗ · 2026-07-13 Cached

GrAInS is a contrastive gradient-based method that uses Integrated Gradients to identify influential tokens and construct steering vectors for inference-time steering of LLMs and VLMs, improving truthfulness and reducing hallucinations without degrading fluency.

0 favorites 0 likes
#inference-time-steering

QPILOTS: Efficient Test-Time Q-Steering for Flow Policies

arXiv cs.LG ↗ · 2026-06-16 Cached

QPILOTS is a method that steers flow policies at inference time by using critic gradients projected from noisy intermediate states, achieving state-of-the-art performance on offline-to-online RL benchmarks and improving pretrained VLA models without modifying the base policy.

0 favorites 0 likes
#inference-time-steering

FineSteer: A Unified Framework for Fine-Grained Inference-Time Steering in Large Language Models

arXiv cs.CL ↗ · 2026-04-20 Cached

FineSteer is a novel inference-time steering framework that decomposes steering into conditional steering and fine-grained vector synthesis stages, using Subspace-guided Conditional Steering (SCS) and Mixture-of-Steering-Experts (MoSE) mechanisms to improve safety and truthfulness while preserving model utility. Experiments show 7.6% improvement over state-of-the-art methods on TruthfulQA with minimal utility loss.

0 favorites 0 likes
← Back to home

Submit Feedback