---
title: "Dynamic Tool Dependency Retrieval for Lightweight Function"
description: "The authors, from Qualcomm AI Research and the University of Maryland, propose Dynamic Tool Dependency Retrieval… 35 references."
authors: ["Bhrij Patel", "Davide Belli", "Amir Jalalirad", "Maximilian Arnold", "Aleksandr Ermolov", "Bence Major"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.1680/"
doi: "10.18653/v1/2026.findings-acl.1680"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/patel-2026-dynamic-tool-dependency-retrieval-lightweight/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/patel-2026-dynamic-tool-dependency-retrieval-lightweight/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/28/) / Dynamic Tool Dependency Retrieval for Lightweight Function Calling

# Dynamic Tool Dependency Retrieval for Lightweight Function Calling

**Authors:** Bhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold, Aleksandr Ermolov and Bence Major · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.1680/) · [PDF](https://aclanthology.org/2026.findings-acl.1680.pdf) · [DOI 10.18653/v1/2026.findings-acl.1680](https://doi.org/10.18653/v1/2026.findings-acl.1680) · [OpenAlex](https://openalex.org/W7166887594)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
On-device function-calling agents usually depend on a retriever to narrow a large tool inventory down to a few candidates for the LLM prompt. Existing retrievers either match the query against tool descriptions or use a fixed dependency graph keyed only on the most recent call. Both approaches miss how tool needs change as a multi-step plan grows, so they let irrelevant tools into the prompt. The authors, from Qualcomm AI Research and the University of Maryland, propose Dynamic Tool Dependency Retrieval (DTDR). It learns tool dependencies from demonstration trajectories and retrieves candidates using both the user query and the full history of calls planned so far. DTDR has two lightweight variants:
- **DTDR-C** clusters demonstrations and builds an explicit dependency graph per cluster.
- **DTDR-L** is a linear classifier that captures dependencies implicitly.

The authors compare DTDR against text-, embedding- and graph-based retrievers across several datasets and LLM backbones. They also test different ways of encoding the retrieved tools into the prompt. DTDR raises function-calling success by 23–104% over static retrievers, which makes it relevant for small agents with tight memory and latency budgets.

## Key points
- DTDR is the only method in the authors' comparison that meets all five criteria they list: dependency-aware, description-free, query-aware, aware of multi-step history, and small enough for on-device use.
- For the benchmarks where tools depend on one another (TA, TB-HF, TB-MM), the dynamic variants improve end-to-end Success Rate by 300–600% over no ICL and by 15–200% over the strongest retrieval baseline.
- Weighted Hard Masking was the best of the ICL encoding strategies tested.
- Ablations show the following:
  - Using up to 3 past calls of history helps, with smaller gains beyond that.
  - DTDR-C works best when the number of clusters is about one tenth of the number of demonstrations.
  - Gains level off around 10k demonstration samples.
  - DTDR-L overfits when given fewer than 1k samples.
- On TinyAgent, DTDR-L keeps top-1 retrieval accuracy between roughly 85% and 96% across plan lengths. DTDR-C declines on plans longer than about 6 steps.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Maryland, College Park](https://md.research.elamur.ai/free/en/universities/university-of-maryland-college-park/) — University, College Park, United States · 132 works · 25,380 citations
- [Qualcomm](https://md.research.elamur.ai/free/en/companies/qualcomm/) — Company, San Diego, United States · 6 works · 144 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works

🧭 Topics (1): [🔍 Information retrieval & search](https://md.research.elamur.ai/free/en/topics/information-retrieval/) · 473 works

## Cite

Cite as: Bhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold, Aleksandr Ermolov and Bence Major. 2026. Dynamic Tool Dependency Retrieval for Lightweight Function Calling. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{patel2026dynamic,
  title = {{Dynamic Tool Dependency Retrieval for Lightweight Function Calling}},
  author = {Bhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold, Aleksandr Ermolov and Bence Major},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1680},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1680}
}
```

- [← previous Dynamic Model Switching to Mitigate Outdated Knowledge in Large Language Models](https://md.research.elamur.ai/free/en/papers/pinninti-2026-dynamic-model-switching-mitigate-outdated/)
- [next → DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections](https://md.research.elamur.ai/free/en/papers/shin-2026-dynamixsft-dynamic-mixture-optimization-instruction/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/28/)

## References

35 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Anthropic. 2025. Claude 4 system card: Opus and sonnet models. [https://www.anthropic.com/research](https://www.anthropic.com/research). Comprehensive technical and safety report for Claude 4 models. [↗](https://www.anthropic.com/research)
2. Victor Barres, Honghua Dong, Soham Ray, Xujie Si, and Karthik Narasimhan. 2025. τ 2 -bench: Evaluating conversational agents in a dual-control environment. arXiv preprint arXiv:2506.07982. [find](https://scholar.google.com/scholar?q=Victor+Barres%2C+Honghua+Dong%2C+Soham+Ray%2C+Xujie+Si%2C+and+Karthik+Narasimhan%2E+2025%2E+%CF%84+2+%2Dbench%3A+Evaluating+conversational+agents+in+a+dual%2Dcontrol+environment%2E+arXiv+preprint+arXiv%3A2506%2E07982%2E)
3. Norbert Braunschweiler, Rama Doddipatla, and Tudor-Catalin Zorila. 2025. Toolreagt: Tool retrieval for llm-based complex task solution via retrieval augmented generation. In Proceedings of the 3rd Workshop on Towards Knowledgeable Foundation Models (KnowFM), pages 75–83. [find](https://scholar.google.com/scholar?q=Norbert+Braunschweiler%2C+Rama+Doddipatla%2C+and+Tudor%2DCatalin+Zorila%2E+2025%2E+Toolreagt%3A+Tool+retrieval+for+llm%2Dbased+complex+task+solution+via+retrieval+augmented+generation%2E+In+Proceedings+of+the+3rd+Wor)
4. Wenjie Chen, Wenbin Li, Di Yao, Xuying Meng, Chang Gong, and Jingping Bi. 2025. Gtool: Graph enhanced tool planning with large language model. arXiv preprint arXiv:2508.12725. — [arXiv:2508.12725](https://arxiv.org/abs/2508.12725) [find](https://scholar.google.com/scholar?q=Wenjie+Chen%2C+Wenbin+Li%2C+Di+Yao%2C+Xuying+Meng%2C+Chang+Gong%2C+and+Jingping+Bi%2E+2025%2E+Gtool%3A+Graph+enhanced+tool+planning+with+large+language+model%2E+arXiv+preprint+arXiv%3A2508%2E12725%2E+%E2%80%94+%5BarXiv%3A2508%2E12725%5D%28htt)
5. Charlie Cheng-Jie Ji, Mao Huanzhi, Yan Fanjia, G. Patil Shishir, Zhang Tianjun, Stoica Ion, and E. Gonzalez Joseph. 2024. Gorilla openfunctions v2. [find](https://scholar.google.com/scholar?q=Charlie+Cheng%2DJie+Ji%2C+Mao+Huanzhi%2C+Yan+Fanjia%2C+G%2E+Patil+Shishir%2C+Zhang+Tianjun%2C+Stoica+Ion%2C+and+E%2E+Gonzalez+Joseph%2E+2024%2E+Gorilla+openfunctions+v2%2E)
6. Sijia Cui, Aiyao He, Shuai Xu, Hongming Zhang, Yanna Wang, Qingyang Zhang, Yajing Wang, and Bo Xu. 2025. Self-guided function calling in large language models via stepwise experience recall. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 10842–10854, Suzhou, China. A… [find](https://scholar.google.com/scholar?q=Sijia+Cui%2C+Aiyao+He%2C+Shuai+Xu%2C+Hongming+Zhang%2C+Yanna+Wang%2C+Qingyang+Zhang%2C+Yajing+Wang%2C+and+Bo+Xu%2E+2025%2E+Self%2Dguided+function+calling+in+large+language+models+via+stepwise+experience+recall%2E+In+Findin)
7. Hy Dang, Tianyi Liu, Zhuofeng Wu, Jingfeng Yang, Haoming Jiang, Tao Yang, Pei Chen, Zhengyang Wang, Helen Wang, Huasheng Li, and 1 others. 2025. Improving large language models function calling and interpretability via guided-structured templates. arXiv preprint arXiv:2509.18076. [find](https://scholar.google.com/scholar?q=Hy+Dang%2C+Tianyi+Liu%2C+Zhuofeng+Wu%2C+Jingfeng+Yang%2C+Haoming+Jiang%2C+Tao+Yang%2C+Pei+Chen%2C+Zhengyang+Wang%2C+Helen+Wang%2C+Huasheng+Li%2C+and+1+others%2E+2025%2E+Improving+large+language+models+function+calling+and+in)
8. Keyan Ding, Jing Yu, Junjie Huang, Yuchen Yang, Qiang Zhang, and Huajun Chen. 2025. Scitoolagent: a knowledge-graph-driven scientific agent for multitool integration. Nature Computational Science, pages 1–11. [find](https://scholar.google.com/scholar?q=Keyan+Ding%2C+Jing+Yu%2C+Junjie+Huang%2C+Yuchen+Yang%2C+Qiang+Zhang%2C+and+Huajun+Chen%2E+2025%2E+Scitoolagent%3A+a+knowledge%2Dgraph%2Ddriven+scientific+agent+for+multitool+integration%2E+Nature+Computational+Science%2C+pag)
9. Lutfi Eren Erdogan, Nicholas Lee, Siddharth Jha, Sehoon Kim, Ryan Tabrizi, Suhong Moon, Coleman Hooper, Gopala Anumanchipalli, Kurt Keutzer, and Amir Gholami. 2024. Tinyagent: Function calling at the edge. arXiv preprint arXiv:2409.00608. [find](https://scholar.google.com/scholar?q=Lutfi+Eren+Erdogan%2C+Nicholas+Lee%2C+Siddharth+Jha%2C+Sehoon+Kim%2C+Ryan+Tabrizi%2C+Suhong+Moon%2C+Coleman+Hooper%2C+Gopala+Anumanchipalli%2C+Kurt+Keutzer%2C+and+Amir+Gholami%2E+2024%2E+Tinyagent%3A+Function+calling+at+the+)
10. Marco Federici, Davide Belli, Mart Van Baalen, Amir Jalalirad, Andrii Skliar, Bence Major, Markus Nagel, and Paul Whatmough. 2025. Efficient llm inference using dynamic input pruning and cache-aware masking. In Eighth Conference on Machine Learning and Systems. [find](https://scholar.google.com/scholar?q=Marco+Federici%2C+Davide+Belli%2C+Mart+Van+Baalen%2C+Amir+Jalalirad%2C+Andrii+Skliar%2C+Bence+Major%2C+Markus+Nagel%2C+and+Paul+Whatmough%2E+2025%2E+Efficient+llm+inference+using+dynamic+input+pruning+and+cache%2Daware+m)
11. Linfeng Gao, Yaoxiang Wang, Minlong Peng, Jialong Tang, Yuzhe Shang, Mingming Sun, and Jinsong Su. 2025. Tool graph retriever: Exploring dependency graph-based tool retrieval for large language models. arXiv preprint arXiv:2508.05152. — [arXiv:2508.05152](https://arxiv.org/abs/2508.05152) [find](https://scholar.google.com/scholar?q=Linfeng+Gao%2C+Yaoxiang+Wang%2C+Minlong+Peng%2C+Jialong+Tang%2C+Yuzhe+Shang%2C+Mingming+Sun%2C+and+Jinsong+Su%2E+2025%2E+Tool+graph+retriever%3A+Exploring+dependency+graph%2Dbased+tool+retrieval+for+large+language+models)
12. Google. 2025. Google Antigravity AI IDE. [find](https://scholar.google.com/scholar?q=Google%2E+2025%2E+Google+Antigravity+AI+IDE%2E)
13. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. [Gpt-4o system card](https://md.research.elamur.ai/free/en/papers/openai-2024-gpt-4o-system-card/). arXiv preprint arXiv:2410.21276.
14. Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, and 1 others. 2023. [Toolllm: Facilitating large language models to master 16000+ real-world apis](https://md.research.elamur.ai/free/en/papers/qin-2023-toolllm-facilitating-large-language-models/). arXiv preprint arXiv:2307.16789.
15. Qiqiang Lin, Muning Wen, Qiuying Peng, Guanyu Nie, Junwei Liao, Jun Wang, Xiaoyun Mo, Jiamu Zhou, Cheng Cheng, Yin Zhao, and 1 others. 2024. Hammer: Robust function-calling for on-device language models via function masking. arXiv preprint arXiv:2410.04587. — [arXiv:2410.04587](https://arxiv.org/abs/2410.04587) [find](https://scholar.google.com/scholar?q=Qiqiang+Lin%2C+Muning+Wen%2C+Qiuying+Peng%2C+Guanyu+Nie%2C+Junwei+Liao%2C+Jun+Wang%2C+Xiaoyun+Mo%2C+Jiamu+Zhou%2C+Cheng+Cheng%2C+Yin+Zhao%2C+and+1+others%2E+2024%2E+Hammer%3A+Robust+function%2Dcalling+for+on%2Ddevice+language+mode)
16. Ella Rabinovich and Ateret Anaby-Tavor. 2025. On the robustness of agentic function calling. arXiv preprint arXiv:2504.00914. [find](https://scholar.google.com/scholar?q=Ella+Rabinovich+and+Ateret+Anaby%2DTavor%2E+2025%2E+On+the+robustness+of+agentic+function+calling%2E+arXiv+preprint+arXiv%3A2504%2E00914%2E)
17. Xukun Liu, Zhiyuan Peng, Xiaoyuan Yi, Xing Xie, Lirong Xiang, Yuchen Liu, and Dongkuan Xu. 2024a. Toolnet: Connecting large language models with massive tools via tool graph. arXiv preprint arXiv:2403.00839. — [arXiv:2403.00839](https://arxiv.org/abs/2403.00839) [find](https://scholar.google.com/scholar?q=Xukun+Liu%2C+Zhiyuan+Peng%2C+Xiaoyuan+Yi%2C+Xing+Xie%2C+Lirong+Xiang%2C+Yuchen+Liu%2C+and+Dongkuan+Xu%2E+2024a%2E+Toolnet%3A+Connecting+large+language+models+with+massive+tools+via+tool+graph%2E+arXiv+preprint+arXiv%3A2403)
18. Zhaoyang Liu, Zeqiang Lai, Zhangwei Gao, Erfei Cui, Ziheng Li, Xizhou Zhu, Lewei Lu, Qifeng Chen, Yu Qiao, Jifeng Dai, and 1 others. 2024b. Controlllm: Augment language models with tools by searching on graphs. In European Conference on Computer Vision, pages 89–105. Springer. [find](https://scholar.google.com/scholar?q=Zhaoyang+Liu%2C+Zeqiang+Lai%2C+Zhangwei+Gao%2C+Erfei+Cui%2C+Ziheng+Li%2C+Xizhou+Zhu%2C+Lewei+Lu%2C+Qifeng+Chen%2C+Yu+Qiao%2C+Jifeng+Dai%2C+and+1+others%2E+2024b%2E+Controlllm%3A+Augment+language+models+with+tools+by+searching+)
19. OpenAI. 2025. Gpt-5 system card. [https://openai.com/research/index/publication/](https://openai.com/research/index/publication/). Describes the architecture, safety measures, and capabilities of GPT-5. [↗](https://openai.com/research/index/publication/)
20. Varatheepan Paramanayakam, Andreas Karatzas, Iraklis Anagnostopoulos, and Dimitrios Stamoulis. 2025. Less is more: Optimizing function calling for llm execution on edge devices. In 2025 Design, Automation & Test in Europe Conference (DATE), pages 1–7. IEEE. [find](https://scholar.google.com/scholar?q=Varatheepan+Paramanayakam%2C+Andreas+Karatzas%2C+Iraklis+Anagnostopoulos%2C+and+Dimitrios+Stamoulis%2E+2025%2E+Less+is+more%3A+Optimizing+function+calling+for+llm+execution+on+edge+devices%2E+In+2025+Design%2C+Automa)
21. Bhargavi Paranjape, Scott Lundberg, Sameer Singh, Hannaneh Hajishirzi, Luke Zettlemoyer, and Marco Tulio Ribeiro. 2023. Art: Automatic multi-step reasoning and tool-use for large language models. arXiv preprint arXiv:2303.09014. [find](https://scholar.google.com/scholar?q=Bhargavi+Paranjape%2C+Scott+Lundberg%2C+Sameer+Singh%2C+Hannaneh+Hajishirzi%2C+Luke+Zettlemoyer%2C+and+Marco+Tulio+Ribeiro%2E+2023%2E+Art%3A+Automatic+multi%2Dstep+reasoning+and+tool%2Duse+for+large+language+models%2E+arXi)
22. Bhrij Patel, Ashish Jagmohan, and Aditya Vempaty. 2025. Learning api functionality from in-context demonstrations for tool-based agents. Empirical Methods of Natural Language Processing. [find](https://scholar.google.com/scholar?q=Bhrij+Patel%2C+Ashish+Jagmohan%2C+and+Aditya+Vempaty%2E+2025%2E+Learning+api+functionality+from+in%2Dcontext+demonstrations+for+tool%2Dbased+agents%2E+Empirical+Methods+of+Natural+Language+Processing%2E)
23. Shishir G Patil, Huanzhi Mao, Fanjia Yan, Charlie Cheng-Jie Ji, Vishnu Suresh, Ion Stoica, and Joseph E Gonzalez. 2025. The berkeley function calling leaderboard (bfcl): From tool use to agentic evaluation of large language models. In Forty-second International Conference on Machine Learning. [find](https://scholar.google.com/scholar?q=Shishir+G+Patil%2C+Huanzhi+Mao%2C+Fanjia+Yan%2C+Charlie+Cheng%2DJie+Ji%2C+Vishnu+Suresh%2C+Ion+Stoica%2C+and+Joseph+E+Gonzalez%2E+2025%2E+The+berkeley+function+calling+leaderboard+%28bfcl%29%3A+From+tool+use+to+agentic+evalu)
24. Shishir G Patil, Tianjun Zhang, Xin Wang, and Joseph E Gonzalez. 2024. [Gorilla: Large language model connected with massive apis](https://md.research.elamur.ai/free/en/papers/patil-2023-gorilla-large-language-model-connected/). Advances in Neural Information Processing Systems, 37:126544–126565.
25. Shuofei Qiao, Honghao Gui, Chengfei Lv, Qianghuai Jia, Huajun Chen, and Ningyu Zhang. 2023. Making language models better tool learners with execution feedback. arXiv preprint arXiv:2305.13068. [find](https://scholar.google.com/scholar?q=Shuofei+Qiao%2C+Honghao+Gui%2C+Chengfei+Lv%2C+Qianghuai+Jia%2C+Huajun+Chen%2C+and+Ningyu+Zhang%2E+2023%2E+Making+language+models+better+tool+learners+with+execution+feedback%2E+arXiv+preprint+arXiv%3A2305%2E13068%2E)
26. Nils Reimers and Iryna Gurevych. 2019. [Sentence-bert: Sentence embeddings using siamese bert-networks](https://md.research.elamur.ai/free/en/papers/reimers-2019-sentence-bert-sentence-embeddings-siamese/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
27. Stephen Robertson, Hugo Zaragoza, and 1 others. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval, 3(4):333–389. [find](https://scholar.google.com/scholar?q=Stephen+Robertson%2C+Hugo+Zaragoza%2C+and+1+others%2E+2009%2E+The+probabilistic+relevance+framework%3A+Bm25+and+beyond%2E+Foundations+and+Trends%C2%AE+in+Information+Retrieval%2C+3%284%29%3A333%E2%80%93389%2E)
28. Vishnu Sarukkai, Zhiqiang Xie, and Kayvon Fatahalian. 2025. Self-generated in-context examples improve llm agents for sequential decision-making tasks. arXiv preprint arXiv:2505.00234. — [arXiv:2505.00234](https://arxiv.org/abs/2505.00234) [find](https://scholar.google.com/scholar?q=Vishnu+Sarukkai%2C+Zhiqiang+Xie%2C+and+Kayvon+Fatahalian%2E+2025%2E+Self%2Dgenerated+in%2Dcontext+examples+improve+llm+agents+for+sequential+decision%2Dmaking+tasks%2E+arXiv+preprint+arXiv%3A2505%2E00234%2E+%E2%80%94+%5BarXiv%3A2505%2E0)
29. Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. [Toolformer: Language models can teach themselves to use tools](https://md.research.elamur.ai/free/en/papers/schick-2023-toolformer-language-models-can-teach/). Advances in Neural Information Processing Systems, 36:68539–68551.
30. Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. 2023. [Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face](https://md.research.elamur.ai/free/en/papers/shen-2023-hugginggpt-solving-ai-tasks-chatgpt/). Advances in Neural Information Processing Systems, 36:38154–38180.
31. Yongliang Shen, Kaitao Song, Xu Tan, Wenqi Zhang, Kan Ren, Siyu Yuan, Weiming Lu, Dongsheng Li, and Yueting Zhuang. 2024. Taskbench: Benchmarking large language models for task automation. Advances in Neural Information Processing Systems, 37:4540–4574. [find](https://scholar.google.com/scholar?q=Yongliang+Shen%2C+Kaitao+Song%2C+Xu+Tan%2C+Wenqi+Zhang%2C+Kan+Ren%2C+Siyu+Yuan%2C+Weiming+Lu%2C+Dongsheng+Li%2C+and+Yueting+Zhuang%2E+2024%2E+Taskbench%3A+Benchmarking+large+language+models+for+task+automation%2E+Advances+in)
32. Qingyu Song, Peiyu Liao, Wenqian Zhao, Yiwen Wang, Shoubo Hu, Hui-Ling Zhen, Ning Jiang, and Mingxuan Yuan. 2025. Harnessing on-device large language model: Empirical results and implications for ai pc. arXiv preprint arXiv:2505.15030. — [arXiv:2505.15030](https://arxiv.org/abs/2505.15030) [find](https://scholar.google.com/scholar?q=Qingyu+Song%2C+Peiyu+Liao%2C+Wenqian+Zhao%2C+Yiwen+Wang%2C+Shoubo+Hu%2C+Hui%2DLing+Zhen%2C+Ning+Jiang%2C+and+Mingxuan+Yuan%2E+2025%2E+Harnessing+on%2Ddevice+large+language+model%3A+Empirical+results+and+implications+for+ai+p)
33. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. [Qwen3 technical report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/). arXiv preprint arXiv:2505.09388.
34. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. [React: Synergizing reasoning and acting in language models](https://md.research.elamur.ai/free/en/papers/yao-2022-react-synergizing-reasoning-acting-language/). In International Conference on Learning Representations (ICLR).
35. Aohan Zeng, Xin Lv, Qinkai Zheng, Zhenyu Hou, Bin Chen, Chengxing Xie, Cunxiang Wang, Da Yin, Hao Zeng, Jiajie Zhang, and 1 others. 2025. [Glm-4.5: Agentic, reasoning, and coding (arc) foundation models](https://md.research.elamur.ai/free/en/papers/team-2025-4-glm-4-5-agentic/). arXiv preprint arXiv:2508.06471.

*Updated 1 November 2026*