Saved in:
Bibliographic Details
Main Authors: Zhang, Jason, Viteri, Scott
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2409.14026
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915207080574976
author Zhang, Jason
Viteri, Scott
author_facet Zhang, Jason
Viteri, Scott
contents In this work, we examine how targeted perturbations in the activation space of Language Models (LMs) can encode complex reasoning patterns. We inject steering vectors, derived from LM activations, into LMs during inference time and study whether these vectors can induce Chain-of-Thought (CoT) reasoning in LMs without the need for natural language prompting. We demonstrate this approach on Llama3 8B Instruct and Mistral 7B v0.2 Instruct and show that activation-space interventions achieve competitive, if not superior, performance compared to traditional CoT prompting across multiple reasoning benchmarks, including GSM8k, MMLU, AGI Eval, and ARC AI2. These findings suggest that neural network activations can encode reasoning patterns, offering a new application of activation space manipulation as a tool for tuning model behavior.
format Preprint
id arxiv_https___arxiv_org_abs_2409_14026
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Uncovering Latent Chain of Thought Vectors in Language Models
Zhang, Jason
Viteri, Scott
Computation and Language
Artificial Intelligence
In this work, we examine how targeted perturbations in the activation space of Language Models (LMs) can encode complex reasoning patterns. We inject steering vectors, derived from LM activations, into LMs during inference time and study whether these vectors can induce Chain-of-Thought (CoT) reasoning in LMs without the need for natural language prompting. We demonstrate this approach on Llama3 8B Instruct and Mistral 7B v0.2 Instruct and show that activation-space interventions achieve competitive, if not superior, performance compared to traditional CoT prompting across multiple reasoning benchmarks, including GSM8k, MMLU, AGI Eval, and ARC AI2. These findings suggest that neural network activations can encode reasoning patterns, offering a new application of activation space manipulation as a tool for tuning model behavior.
title Uncovering Latent Chain of Thought Vectors in Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2409.14026