الملخص
We study how to build more capable general-purpose agents by extending large language models (LLMs) with native typed decision-making. We introduce Dyad, an architecture that augments a pretrained LLM with an environment-conditioned action encoder that embeds each candidate action description in parallel, then scores these embeddings against the LLM's internal state to yield a distribution over typed actions. By factorizing decision-making into representations of the evolving interaction state and environment-specific action semantics, Dyad introduces an inductive bias for learning reusable representations while keeping action scoring efficient even as the action space grows. We investigate two complementary reinforcement learning settings driven by environment interaction. With the LLM frozen, training the action encoder alone achieves consistent gains across four unseen environments, enabling modular adaptation without modifying any LLM parameters. Jointly optimizing both components outperforms conventional RL post-training across diverse interactive tasks and model scales, including a 3.80% average absolute gain on ALFWorld with a 9B model, while improving general knowledge, reasoning, and coding.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Zhan, Y., Wang, W., Liu, W., Dahlmeier, D., Qin, C., Li, J., Johansson, F. D., & Yue, Z. (2026). Dyad: Extending Large Language Models with Native Typed Decision-Making. https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making
MLA 9
Zhan, Yundaichuan, et al. "Dyad: Extending Large Language Models with Native Typed Decision-Making." https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making.
شيكاغو (المؤلف–التاريخ)
Zhan, Yundaichuan, Weishi Wang, Wenbiao Liu, Daniel Dahlmeier, Chengwei Qin, Juncheng Li, Fredrik D. Johansson, and Zhongqi Yue. 2026. "Dyad: Extending Large Language Models with Native Typed Decision-Making." https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making.
هارفارد
Zhan, Y., Wang, W., Liu, W., Dahlmeier, D., Qin, C., Li, J., Johansson, F. D. and Yue, Z. (2026) 'Dyad: Extending Large Language Models with Native Typed Decision-Making', Available at: https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making.
فانكوفر
Zhan Y, Wang W, Liu W, Dahlmeier D, Qin C, Li J, et al. Dyad: Extending Large Language Models with Native Typed Decision-Making. https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making
IEEE
Y. Zhan, W. Wang, W. Liu, D. Dahlmeier, C. Qin, J. Li, F. D. Johansson, and Z. Yue, "Dyad: Extending Large Language Models with Native Typed Decision-Making," https://omanscience.com/ar/articles/dyad-extending-large-language-models-with-native-typed-decision-making.