Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909466158432256 |
|---|---|
| author | Yang, Dingkang Xiao, Dongling Wei, Jinjie Li, Mingcheng Chen, Zhaoyu Li, Ke Zhang, Lihua |
| author_facet | Yang, Dingkang Xiao, Dongling Wei, Jinjie Li, Mingcheng Chen, Zhaoyu Li, Ke Zhang, Lihua |
| contents | Despite their remarkable capabilities, Large Language Models (LLMs) are prone to generate responses that contradict verifiable facts, i.e., unfaithful hallucination content. Existing efforts generally focus on optimizing model parameters or editing semantic representations, which compromise the internal factual knowledge of target LLMs. In addition, hallucinations typically exhibit multifaceted patterns in downstream tasks, limiting the model's holistic performance across tasks. In this paper, we propose a Comparator-driven Decoding-Time (CDT) framework to alleviate the response hallucination. Firstly, we construct hallucinatory and truthful comparators with multi-task fine-tuning samples. In this case, we present an instruction prototype-guided mixture of experts strategy to enhance the ability of the corresponding comparators to capture different hallucination or truthfulness patterns in distinct task instructions. CDT constrains next-token predictions to factuality-robust distributions by contrasting the logit differences between the target LLMs and these comparators. Systematic experiments on multiple downstream tasks show that our framework can significantly improve the model performance and response factuality. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2408_12325 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators Yang, Dingkang Xiao, Dongling Wei, Jinjie Li, Mingcheng Chen, Zhaoyu Li, Ke Zhang, Lihua Computation and Language Despite their remarkable capabilities, Large Language Models (LLMs) are prone to generate responses that contradict verifiable facts, i.e., unfaithful hallucination content. Existing efforts generally focus on optimizing model parameters or editing semantic representations, which compromise the internal factual knowledge of target LLMs. In addition, hallucinations typically exhibit multifaceted patterns in downstream tasks, limiting the model's holistic performance across tasks. In this paper, we propose a Comparator-driven Decoding-Time (CDT) framework to alleviate the response hallucination. Firstly, we construct hallucinatory and truthful comparators with multi-task fine-tuning samples. In this case, we present an instruction prototype-guided mixture of experts strategy to enhance the ability of the corresponding comparators to capture different hallucination or truthfulness patterns in distinct task instructions. CDT constrains next-token predictions to factuality-robust distributions by contrasting the logit differences between the target LLMs and these comparators. Systematic experiments on multiple downstream tasks show that our framework can significantly improve the model performance and response factuality. |
| title | Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2408.12325 |