---
title: "Revisiting Model Interpolation for Efficient Reasoning — Wu 2026"
description: "Instead of complex merging schemes, the authors study plain weight interpolation (MI) between two… 41 references, cited by 6 papers here."
authors: ["Taiqiang Wu", "Runming Yang", "Tao Liu", "Jiahao Wang", "Ngai Wong"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.389/"
doi: "10.18653/v1/2026.acl-long.389"
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/wu-2026-revisiting-model-interpolation-efficient-reasoning/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wu-2026-revisiting-model-interpolation-efficient-reasoning/)

[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/3/) / Revisiting Model Interpolation for Efficient Reasoning

# Revisiting Model Interpolation for Efficient Reasoning

**Authors:** [Taiqiang Wu](https://md.research.elamur.ai/free/en/authors/taiqiang-wu/), [Runming Yang](https://md.research.elamur.ai/free/en/authors/runming-yang/), [Tao Liu](https://md.research.elamur.ai/free/en/authors/tao-liu/), [Jiahao Wang](https://md.research.elamur.ai/free/en/authors/jiahao-wang/) and [Ngai Wong](https://md.research.elamur.ai/free/en/authors/ngai-wong/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/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.acl-long.389/) · [PDF](https://aclanthology.org/2026.acl-long.389.pdf) · [DOI 10.18653/v1/2026.acl-long.389](https://doi.org/10.18653/v1/2026.acl-long.389) · [OpenAlex](https://openalex.org/W7166864657)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [6](https://md.research.elamur.ai/free/en/papers/wu-2026-revisiting-model-interpolation-efficient-reasoning/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper looks at efficient reasoning: how to keep the accuracy of long chain-of-thought models while producing fewer tokens. Instead of complex merging schemes, the authors study plain weight interpolation (MI) between two Qwen3 specialists, the Instruct-2507 and Thinking-2507 variants of Qwen3-4B and Qwen3-30B-A3B. They sweep the mixing coefficient λ from 0 (pure Instruct) to 1 (pure Thinking) and track accuracy, output length, and how often explicit thinking appears. Evaluation covers AIME'25, IFEval and GPQA-Diamond using OpenCompass.

The behaviour turns out to be non-linear and follows three stages:
1. Accuracy and length grow quickly, with almost no explicit reasoning.
2. Explicit thinking emerges.
3. Responses lengthen sharply while gains flatten, which signals overthinking.

This pattern lets practitioners choose λ for a target accuracy or token budget. A well-chosen interpolated model beats Task Arithmetic and TIES-Merging baselines in both accuracy and efficiency. Ablations on layers, sub-modules, decoding settings and alternative backbones help explain where the reasoning behaviour resides in the network.

## Key points
- **Relation to other methods:** MI does not depend on a shared base model and can be written as a special case of Task Arithmetic. Kimi k1.5's model averaging corresponds to λ = 0.5.
- **Experimental setup:** Stage 1 covers roughly λ \< 0.4 for Qwen3-4B. Responses were sampled 64 times per problem on AIME'25 and 8 times on IFEval and GPQA-Diamond, with metrics Pass@k, Mean@k, Vote@k, Token #N and Think #R.
- **Layer ablation (Qwen3-4B, 36 layers):** Interpolating any single third of the layers fails to trigger thinking. Interpolating only the last two-thirds reaches 100% Think #R and nearly matches full interpolation.
- **Module ablation (MI-0.8 on AIME'25):**
  - Leaving FFN sub-layers un-interpolated drops the thinking ratio from 99.95% to 0.68%.
  - Leaving MHA sub-layers un-interpolated keeps thinking intact but cuts Mean@64 from 80.47 to 71.46.
- **Alternative backbones:** Replacing the Instruct model with Qwen3-4B-Base lowers AIME'25 Mean@64 from 80.5 to 67.7. The hybrid Qwen3-4B performs on par with or slightly above Instruct. Validation is limited to Qwen3 models and to two-model merges.

*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

- [Tsinghua University](https://md.research.elamur.ai/free/en/universities/tsinghua-university/) — University, Beijing, China · 333 works · 87,019 citations
- [University of Hong Kong](https://md.research.elamur.ai/free/en/universities/university-of-hong-kong/) — University, Pok Fu Lam, China · 89 works · 11,774 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 · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works

🧭 Topics (1): [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,426 works

## Cite

Cite as: Taiqiang Wu, Runming Yang, Tao Liu, Jiahao Wang and Ngai Wong. 2026. Revisiting Model Interpolation for Efficient Reasoning. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

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

```bibtex
@inproceedings{wu2026revisiting,
  title = {{Revisiting Model Interpolation for Efficient Reasoning}},
  author = {Taiqiang Wu, Runming Yang, Tao Liu, Jiahao Wang and Ngai Wong},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.389},
  url = {https://doi.org/10.18653/v1/2026.acl-long.389}
}
```

- [← previous PDTrim: Targeted Pruning for Prefill-Decode Disaggregation in Inference](https://md.research.elamur.ai/free/en/papers/zhang-2026-pdtrim-targeted-pruning-prefill-decode/)
- [next → RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization](https://md.research.elamur.ai/free/en/papers/dong-2026-rl-plus-countering-capability-boundary/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/3/)

## References

41 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. AIME. 2025. Aime problems and solutions. Accessed: 2024-06-15. [find](https://scholar.google.com/scholar?q=AIME%2E+2025%2E+Aime+problems+and+solutions%2E+Accessed%3A+2024%2D06%2D15%2E)
2. Simon A Aytes, Jinheon Baek, and Sung Ju Hwang. 2025. [Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching](https://md.research.elamur.ai/free/en/papers/aytes-2025-sketch-thought-efficient-llm-reasoning/). arXiv preprint arXiv:2503.05179.
3. Natasha Butt, Ariel Kwiatkowski, Ismail Labiad, Julia Kempe, and Yann Ollivier. 2025. Soft tokens, hard truths. arXiv preprint arXiv:2509.19170. — [arXiv:2509.19170](https://arxiv.org/abs/2509.19170) [find](https://scholar.google.com/scholar?q=Natasha+Butt%2C+Ariel+Kwiatkowski%2C+Ismail+Labiad%2C+Julia+Kempe%2C+and+Yann+Ollivier%2E+2025%2E+Soft+tokens%2C+hard+truths%2E+arXiv+preprint+arXiv%3A2509%2E19170%2E+%E2%80%94+%5BarXiv%3A2509%2E19170%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2509%2E19170%29)
4. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, and 1 others. 2021. [Evaluating large language models trained on code](https://md.research.elamur.ai/free/en/papers/chen-2021-evaluating-large-language-models-trained/). arXiv preprint arXiv:2107.03374.
5. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, and 1 others. 2024. [Do not think that much for 2+ 3=?](https://md.research.elamur.ai/free/en/papers/chen-2024-think-that-much-2-3/) on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187.
6. Tejalal Choudhary, Vipul Mishra, Anurag Goswami, and Jagannathan Sarangapani. 2020. A comprehensive survey on model compression and acceleration. Artificial Intelligence Review, 53(7):5113–5155. [find](https://scholar.google.com/scholar?q=Tejalal+Choudhary%2C+Vipul+Mishra%2C+Anurag+Goswami%2C+and+Jagannathan+Sarangapani%2E+2020%2E+A+comprehensive+survey+on+model+compression+and+acceleration%2E+Artificial+Intelligence+Review%2C+53%287%29%3A5113%E2%80%935155%2E)
7. OpenCompass Contributors. 2023. Opencompass: A universal evaluation platform for foundation models. [https://github.com/open-compass/](https://github.com/open-compass/) opencompass. [↗](https://github.com/open-compass/)
8. Ankit Gangwal and Aaryan Ajay Sharma. 2025. Merge now, regret later: The hidden cost of model merging is adversarial transferability. arXiv preprint arXiv:2509.23689. — [arXiv:2509.23689](https://arxiv.org/abs/2509.23689) [find](https://scholar.google.com/scholar?q=Ankit+Gangwal+and+Aaryan+Ajay+Sharma%2E+2025%2E+Merge+now%2C+regret+later%3A+The+hidden+cost+of+model+merging+is+adversarial+transferability%2E+arXiv+preprint+arXiv%3A2509%2E23689%2E+%E2%80%94+%5BarXiv%3A2509%2E23689%5D%28https%3A%2F%2Farxi)
9. Charles Goddard, Shamane Siriwardhana, Malikeh Ehghaghi, Luke Meyers, Vladimir Karpukhin, Brian Benedict, Mark McQuade, and Jacob Solawetz. 2024. [Arcee’s mergekit: A toolkit for merging large language models](https://md.research.elamur.ai/free/en/papers/goddard-2024-arcee-s-mergekit-toolkit-merging/). In Conference on Empirical Methods in Natural Language Processing.
10. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. [Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/). arXiv preprint arXiv:2501.12948.
11. Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, and Yuandong Tian. 2024. Training large language models to reason in a continuous latent space. arXiv preprint arXiv:2412.06769. [find](https://scholar.google.com/scholar?q=Shibo+Hao%2C+Sainbayar+Sukhbaatar%2C+DiJia+Su%2C+Xian+Li%2C+Zhiting+Hu%2C+Jason+Weston%2C+and+Yuandong+Tian%2E+2024%2E+Training+large+language+models+to+reason+in+a+continuous+latent+space%2E+arXiv+preprint+arXiv%3A2412%2E)
12. Gabriel Ilharco, Marco Tulio Ribeiro, Mitchell Wortsman, Suchin Gururangan, Ludwig Schmidt, Hannaneh Hajishirzi, and Ali Farhadi. 2022. [Editing models with task arithmetic](https://md.research.elamur.ai/free/en/papers/ilharco-2022-editing-models-task-arithmetic/). arXiv preprint arXiv:2212.04089.
13. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, and 1 others. 2024. [Openai o1 system card](https://md.research.elamur.ai/free/en/papers/openai-2024-openai-o1-system-card/). arXiv preprint arXiv:2412.16720.
14. Zhong-Zhi Li, Xiao Liang, Zihao Tang, Lei Ji, Peijie Wang, Haotian Xu, Haizhen Huang, Weiwei Deng, Ying Nian Wu, Yeyun Gong, and 1 others. 2025. [Tl; dr: Too long, do re-weighting for effcient llm reasoning compression](https://md.research.elamur.ai/free/en/papers/li-2025-tl-dr-too-long-re/). arXiv preprint arXiv:2506.02678.
15. Xinyin Ma, Guangnian Wan, Runpeng Yu, Gongfan Fang, and Xinchao Wang. 2025. [Cot-valve: Length-compressible chain-of-thought tuning](https://md.research.elamur.ai/free/en/papers/ma-2025-cot-valve-length-compressible-chain/). In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6025–6035.
16. Daniel Marczak, Bartłomiej Twardowski, Tomasz Trz-ciński, and Sebastian Cygert. 2024. Magmax: Leveraging model merging for seamless continual learning. In European Conference on Computer Vision, pages 379–395. Springer. [find](https://scholar.google.com/scholar?q=Daniel+Marczak%2C+Bart%C5%82omiej+Twardowski%2C+Tomasz+Trz%2Dcin%CC%81ski%2C+and+Sebastian+Cygert%2E+2024%2E+Magmax%3A+Leveraging+model+merging+for+seamless+continual+learning%2E+In+European+Conference+on+Computer+Vision%2C+page)
17. Mohammed Muqeeth, Haokun Liu, and Colin Raffel. 2023. Soft merging of experts with adaptive routing. arXiv preprint arXiv:2306.03745. [find](https://scholar.google.com/scholar?q=Mohammed+Muqeeth%2C+Haokun+Liu%2C+and+Colin+Raffel%2E+2023%2E+Soft+merging+of+experts+with+adaptive+routing%2E+arXiv+preprint+arXiv%3A2306%2E03745%2E)
18. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. 2024. [Gpqa: A graduate-level google-proof q&a benchmark](https://md.research.elamur.ai/free/en/papers/rein-2023-gpqa-graduate-level-google-proof/). In First Conference on Language Modeling.
19. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. [Scaling llm test-time compute optimally can be more effective than scaling model parameters](https://md.research.elamur.ai/free/en/papers/snell-2024-scaling-llm-test-time-compute/). arXiv preprint arXiv:2408.03314.
20. Yang Sui, Yu-Neng Chuang, Guanchu Wang, Jiamu Zhang, Tianyi Zhang, Jiayi Yuan, Hongyi Liu, Andrew Wen, Shaochen Zhong, Na Zou, and 1 others. 2025. [Stop overthinking: A survey on efficient reasoning for large language models](https://md.research.elamur.ai/free/en/papers/sui-2025-stop-overthinking-survey-efficient-reasoning/). arXiv preprint arXiv:2503.16419.
21. Hanshi Sun, Momin Haider, Ruiqi Zhang, Huitao Yang, Jiahao Qiu, Ming Yin, Mengdi Wang, Peter Bartlett, and Andrea Zanette. 2024. Fast best-of-n decoding via speculative rejection. Advances in Neural Information Processing Systems, 37:32630–32652. [find](https://scholar.google.com/scholar?q=Hanshi+Sun%2C+Momin+Haider%2C+Ruiqi+Zhang%2C+Huitao+Yang%2C+Jiahao+Qiu%2C+Ming+Yin%2C+Mengdi+Wang%2C+Peter+Bartlett%2C+and+Andrea+Zanette%2E+2024%2E+Fast+best%2Dof%2Dn+decoding+via+speculative+rejection%2E+Advances+in+Neural+I)
22. Anke Tang, Li Shen, Yong Luo, Shiwei Liu, Han Hu, and Bo Du. 2024. Towards efficient pareto set approximation via mixture of experts based model fusion. arXiv preprint arXiv:2406.09770. — [arXiv:2406.09770](https://arxiv.org/abs/2406.09770) [find](https://scholar.google.com/scholar?q=Anke+Tang%2C+Li+Shen%2C+Yong+Luo%2C+Shiwei+Liu%2C+Han+Hu%2C+and+Bo+Du%2E+2024%2E+Towards+efficient+pareto+set+approximation+via+mixture+of+experts+based+model+fusion%2E+arXiv+preprint+arXiv%3A2406%2E09770%2E+%E2%80%94+%5BarXiv%3A2406%2E)
23. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, and 1 others. 2025. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599. — [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://md.research.elamur.ai/free/en/papers/kimi-team-2025-kimi-k1-5-scaling-reinforcement/)
24. Joachim Utans. 1996. Weight averaging for neural networks and local resampling schemes. In Proc. AAAI-96 Workshop on Integrating Multiple Learned Models. AAAI Press, pages 133–138. Citeseer. [find](https://scholar.google.com/scholar?q=Joachim+Utans%2E+1996%2E+Weight+averaging+for+neural+networks+and+local+resampling+schemes%2E+In+Proc%2E+AAAI%2D96+Workshop+on+Integrating+Multiple+Learned+Models%2E+AAAI+Press%2C+pages+133%E2%80%93138%2E+Citeseer%2E)
25. Yiming Wang, Pei Zhang, Siyuan Huang, Baosong Yang, Zhuosheng Zhang, Fei Huang, and Rui Wang. 2025. Sampling-efficient test-time scaling: Self-estimating the best-of-n sampling in early decoding. arXiv preprint arXiv:2503.01422. [find](https://scholar.google.com/scholar?q=Yiming+Wang%2C+Pei+Zhang%2C+Siyuan+Huang%2C+Baosong+Yang%2C+Zhuosheng+Zhang%2C+Fei+Huang%2C+and+Rui+Wang%2E+2025%2E+Sampling%2Defficient+test%2Dtime+scaling%3A+Self%2Destimating+the+best%2Dof%2Dn+sampling+in+early+decoding%2E+arXi)
26. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). Advances in neural information processing systems, 35:24824– 24837.
27. Han Wu, Yuxuan Yao, Shuqi Liu, Zehua Liu, Xiaojin Fu, Xiongwei Han, Xing Li, Hui-Ling Zhen, Tao Zhong, and Mingxuan Yuan. 2025a. Unlocking efficient long-to-short llm reasoning with model merging. arXiv preprint arXiv:2503.20641. — [arXiv:2503.20641](https://arxiv.org/abs/2503.20641) [find](https://scholar.google.com/scholar?q=Han+Wu%2C+Yuxuan+Yao%2C+Shuqi+Liu%2C+Zehua+Liu%2C+Xiaojin+Fu%2C+Xiongwei+Han%2C+Xing+Li%2C+Hui%2DLing+Zhen%2C+Tao+Zhong%2C+and+Mingxuan+Yuan%2E+2025a%2E+Unlocking+efficient+long%2Dto%2Dshort+llm+reasoning+with+model+merging%2E+arX)
28. Taiqiang Wu, Zenan Xu, Bo Zhou, and Ngai Wong. 2026. The art of efficient reasoning: Data, reward, and optimization. ArXiv, abs/2602.20945. [find](https://scholar.google.com/scholar?q=Taiqiang+Wu%2C+Zenan+Xu%2C+Bo+Zhou%2C+and+Ngai+Wong%2E+2026%2E+The+art+of+efficient+reasoning%3A+Data%2C+reward%2C+and+optimization%2E+ArXiv%2C+abs%2F2602%2E20945%2E)
29. Taiqiang Wu, Runming Yang, Tao Liu, Jiahao Wang, Zenan Xu, and Ngai Wong. 2025c. Timber: Training-free instruct model refining with base via effective rank. arXiv preprint arXiv:2509.23595. — [arXiv:2509.23595](https://arxiv.org/abs/2509.23595) [find](https://scholar.google.com/scholar?q=Taiqiang+Wu%2C+Runming+Yang%2C+Tao+Liu%2C+Jiahao+Wang%2C+Zenan+Xu%2C+and+Ngai+Wong%2E+2025c%2E+Timber%3A+Training%2Dfree+instruct+model+refining+with+base+via+effective+rank%2E+arXiv+preprint+arXiv%3A2509%2E23595%2E+%E2%80%94+%5BarXiv%3A2)
30. Silei Xu, Wenhao Xie, Lingxiao Zhao, and Pengcheng He. 2025. Chain of draft: Thinking faster by writing less. arXiv preprint arXiv:2502.18600. [find](https://scholar.google.com/scholar?q=Silei+Xu%2C+Wenhao+Xie%2C+Lingxiao+Zhao%2C+and+Pengcheng+He%2E+2025%2E+Chain+of+draft%3A+Thinking+faster+by+writing+less%2E+arXiv+preprint+arXiv%3A2502%2E18600%2E)
31. Prateek Yadav, Derek Tam, Leshem Choshen, Colin Raffel, and Mohit Bansal. 2023. [Resolving interference when merging models](https://md.research.elamur.ai/free/en/papers/yadav-2023-ties-merging-resolving-interference-merging/). arXiv preprint arXiv:2306.01708, 1.
32. 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.
33. Enneng Yang, Li Shen, Guibing Guo, Xingwei Wang, Xiaochun Cao, Jie Zhang, and Dacheng Tao. 2024. Model merging in llms, mllms, and beyond: Methods, theories, applications and opportunities. arXiv preprint arXiv:2408.07666. [find](https://scholar.google.com/scholar?q=Enneng+Yang%2C+Li+Shen%2C+Guibing+Guo%2C+Xingwei+Wang%2C+Xiaochun+Cao%2C+Jie+Zhang%2C+and+Dacheng+Tao%2E+2024%2E+Model+merging+in+llms%2C+mllms%2C+and+beyond%3A+Methods%2C+theories%2C+applications+and+opportunities%2E+arXiv+prep)
34. Enneng Yang, Zhenyi Wang, Li Shen, Shiwei Liu, Guibing Guo, Xingwei Wang, and Dacheng Tao. 2023. Adamerging: Adaptive model merging for multi-task learning. arXiv preprint arXiv:2310.02575. [find](https://scholar.google.com/scholar?q=Enneng+Yang%2C+Zhenyi+Wang%2C+Li+Shen%2C+Shiwei+Liu%2C+Guibing+Guo%2C+Xingwei+Wang%2C+and+Dacheng+Tao%2E+2023%2E+Adamerging%3A+Adaptive+model+merging+for+multi%2Dtask+learning%2E+arXiv+preprint+arXiv%3A2310%2E02575%2E)
35. Le Yu, Yu Bowen, Haiyang Yu, Fei Huang, and Yongbin Li. 2023. [Language models are super mario: Absorbing abilities from homologous models as a free lunch](https://md.research.elamur.ai/free/en/papers/yu-2023-language-models-super-mario-absorbing/). ArXiv, abs/2311.03099.
36. Le Yu, Bowen Yu, Haiyang Yu, Fei Huang, and Yongbin Li. 2024. [Language models are super mario: Absorbing abilities from homologous models as a free lunch](https://md.research.elamur.ai/free/en/papers/yu-2023-language-models-super-mario-absorbing/). In Forty-first International Conference on Machine Learning.
37. Zhaojian Yu, Yinghao Wu, Yilun Zhao, Arman Cohan, and Xiao-Ping Zhang. 2025. Z1: Efficient test-time scaling with code. arXiv preprint arXiv:2504.00810. [find](https://scholar.google.com/scholar?q=Zhaojian+Yu%2C+Yinghao+Wu%2C+Yilun+Zhao%2C+Arman+Cohan%2C+and+Xiao%2DPing+Zhang%2E+2025%2E+Z1%3A+Efficient+test%2Dtime+scaling+with+code%2E+arXiv+preprint+arXiv%3A2504%2E00810%2E)
38. Jiajie Zhang, Nianyi Lin, Lei Hou, Ling Feng, and Juanzi Li. 2025a. [Adaptthink: Reasoning models can learn when to think](https://md.research.elamur.ai/free/en/papers/zhang-2025-adaptthink-reasoning-models-can-learn/). arXiv preprint arXiv:2505.13417.
39. Xuechen Zhang, Zijian Huang, Chenshun Ni, Ziyang Xiong, Jiasi Chen, and Samet Oymak. 2025b. Making small language models efficient reasoners: Intervention, supervision, reinforcement. arXiv preprint arXiv:2505.07961. [find](https://scholar.google.com/scholar?q=Xuechen+Zhang%2C+Zijian+Huang%2C+Chenshun+Ni%2C+Ziyang+Xiong%2C+Jiasi+Chen%2C+and+Samet+Oymak%2E+2025b%2E+Making+small+language+models+efficient+reasoners%3A+Intervention%2C+supervision%2C+reinforcement%2E+arXiv+preprint+a)
40. Jeffrey Zhou, Tianjian Lu, Swaroop Mishra, Siddhartha Brahma, Sujoy Basu, Yi Luan, Denny Zhou, and Le Hou. 2023. [Instruction-following evaluation for large language models](https://md.research.elamur.ai/free/en/papers/zhou-2023-instruction-following-evaluation-large-language/). arXiv preprint arXiv:2311.07911.
41. Taiqiang Wu, Runming Yang, Jiayi Li, Pengfei Hu, Ngai Wong, and Yujiu Yang. 2025b. Shadow-ft: Tuning instruct via base. arXiv preprint arXiv:2505.12716. — [arXiv:2505.12716](https://arxiv.org/abs/2505.12716) [find](https://scholar.google.com/scholar?q=Taiqiang+Wu%2C+Runming+Yang%2C+Jiayi+Li%2C+Pengfei+Hu%2C+Ngai+Wong%2C+and+Yujiu+Yang%2E+2025b%2E+Shadow%2Dft%3A+Tuning+instruct+via+base%2E+arXiv+preprint+arXiv%3A2505%2E12716%2E+%E2%80%94+%5BarXiv%3A2505%2E12716%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2505)

*Updated 31 October 2026*