Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909123055976448 |
|---|---|
| author | Sakarvadia, Mansi Ajith, Aswathy Khan, Arham Grzenda, Daniel Hudson, Nathaniel Bauer, André Chard, Kyle Foster, Ian |
| author_facet | Sakarvadia, Mansi Ajith, Aswathy Khan, Arham Grzenda, Daniel Hudson, Nathaniel Bauer, André Chard, Kyle Foster, Ian |
| contents | Answering multi-hop reasoning questions requires retrieving and synthesizing information from diverse sources. Large Language Models (LLMs) struggle to perform such reasoning consistently. Here we propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on LLM attention heads. First, we analyze the per-layer activations of GPT-2 models in response to single and multi-hop prompts. We then propose a mechanism that allows users to inject pertinent prompt-specific information, which we refer to as "memories," at critical LLM locations during inference. By thus enabling the LLM to incorporate additional relevant information during inference, we enhance the quality of multi-hop prompt completions. We show empirically that a simple, efficient, and targeted memory injection into a key attention layer can often increase the probability of the desired next token in multi-hop tasks, by up to 424%. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2309_05605 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Models Sakarvadia, Mansi Ajith, Aswathy Khan, Arham Grzenda, Daniel Hudson, Nathaniel Bauer, André Chard, Kyle Foster, Ian Computation and Language Artificial Intelligence Machine Learning Answering multi-hop reasoning questions requires retrieving and synthesizing information from diverse sources. Large Language Models (LLMs) struggle to perform such reasoning consistently. Here we propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on LLM attention heads. First, we analyze the per-layer activations of GPT-2 models in response to single and multi-hop prompts. We then propose a mechanism that allows users to inject pertinent prompt-specific information, which we refer to as "memories," at critical LLM locations during inference. By thus enabling the LLM to incorporate additional relevant information during inference, we enhance the quality of multi-hop prompt completions. We show empirically that a simple, efficient, and targeted memory injection into a key attention layer can often increase the probability of the desired next token in multi-hop tasks, by up to 424%. |
| title | Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Models |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2309.05605 |