representation-engineering

A resource repository for representation engineering in large language models

Steering vectors for transformer language models in Pytorch / Huggingface

SOM network modified in order to control the latent space

[ICLR 2025] General-purpose activation steering library

A library for making RepE control vectors

Latent or Linguistic: Autonomous Behavior in LLMs. CS120 Final Project Submission

[🔥 ICLR 2026] - Misaligned Roles, Misplaced Images: Structural Input Perturbations Expose Multimodal Alignment Blind Spots

Official code for "Activation Steering for Accent Adaptation in Speech Foundation Models" (Interspeech 2026). Parameter-free accent adaptation via mean-shift steering vectors — no weight updates, consistent WER reductions across 8 accents.

Qwen3-0.6B activation steering: style vectors, lens contamination eval, CPRR methodology

Evaluation framework of different methods for probing and steering LLMs activations to mitigate Chain-of-Thought Unfaithfulness. Research project by Giovanni M. Occhipinti (University of Bologna), Alessandro Abate e Nandi Schoots (University of Oxford).

Phase-aware LLM activation steering and linear probing. A memory-efficient, practical implementation of Representation Engineering (RepE) for safety research.

LatentBiopsy: Geometric Anomaly Detection for LLM Residual Streams.

Representation Rerouting for Agentic Safety: Defending LLM Agents against Prompt Injection via Circuit Breakers and Triplet Loss.

[🏆 CHI26 Best Paper] CoBRA: Reproducible control of LLM agent behavior via classic social science experiments

CRSM (Continuous Reasoning State Model): An asynchronous "System 2" architecture that implements Hierarchical State Sovereignty within a Mamba backbone. Unlike traditional search wrappers, CRSM uses Forward-Projected Planning and Sparse-Gated Injection to steer latent manifolds in real-time, decoupling strategic reasoning from token generation.

Pre-generation tool-call gating via linear probes on LLM hidden states. F1 ≈ 0.91–0.94 on BFCL v4, 14–22× faster than full generation. Cross-architecture transfer across Llama / Qwen / Phi / Mistral (3B–7B) with ≥96% retention.

Mechanistic interpretability experiments on political control circuits, refusal behavior, concept steering, and late-decoder interactions in open LLMs.

Coordinate navigation for Transformer internal representations — zero training, any model

Functional emotional architecture for LLMs — 42 systems, 1994 tests, 27 psychological theories. Emergent emotions via 7 ANIMA pillars: predictive processing, global workspace, autobiographical memory, ontogenic development, motivational drives, emotional discovery, computational phenomenology.