# Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging

arXiv: https://arxiv.org/abs/2505.14136 · HTML: https://mlsanity.com/paper/2505.14136/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in MoE large language models, one of 12 directions in Mixture-of-experts.
- Builds on 64 of its 83 references that are in the map.
- First posted on arXiv 20 May 2025.

## Placed in

- Mixture-of-experts / [MoE large language models](https://mlsanity.com/moe/moe-large-language-models/)

## Builds on (most cited first, 30 of 64)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation](https://mlsanity.com/paper/2208.12242/) (2022)
- [REALM: Retrieval-Augmented Language Model Pre-Training](https://mlsanity.com/paper/2002.08909/) (2020)
- [Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time](https://mlsanity.com/paper/2203.05482/) (2022)
- [Improving language models by retrieving from trillions of tokens](https://mlsanity.com/paper/2112.04426/) (2021)
- [Averaging Weights Leads to Wider Optima and Better Generalization](https://mlsanity.com/paper/1803.05407/) (2018)
- [Editing Models with Task Arithmetic](https://mlsanity.com/paper/2212.04089/) (2022)
- [MPNet: Masked and Permuted Pre-training for Language Understanding](https://mlsanity.com/paper/2004.09297/) (2020)
- [DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models](https://mlsanity.com/paper/2401.06066/) (2024)
- [TIES-Merging: Resolving Interference When Merging Models](https://mlsanity.com/paper/2306.01708/) (2023)
- [Generalization through Memorization: Nearest Neighbor Language Models](https://mlsanity.com/paper/1911.00172/) (2019)
- [Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch](https://mlsanity.com/paper/2311.03099/) (2023)
- [Mixture-of-Experts with Expert Choice Routing](https://mlsanity.com/paper/2202.09368/) (2022)
- [Merging Models with Fisher-Weighted Averaging](https://mlsanity.com/paper/2111.09832/) (2021)
- [Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs](https://mlsanity.com/paper/1802.10026/) (2018)
- [LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion](https://mlsanity.com/paper/2306.02561/) (2023)
- [Git Re-Basin: Merging Models modulo Permutation Symmetries](https://mlsanity.com/paper/2209.04836/) (2022)
- [DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts](https://mlsanity.com/paper/2105.03023/) (2021)
- [Dataless Knowledge Fusion by Merging Weights of Language Models](https://mlsanity.com/paper/2212.09849/) (2022)
- [Learning to (Learn at Test Time): RNNs with Expressive Hidden States](https://mlsanity.com/paper/2407.04620/) (2024)

## Built on by (most cited first, 9 of 9)

- [DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning](https://mlsanity.com/paper/2505.19850/) (2025)
- [Learning on the Job: Test-Time Curricula for Targeted Reinforcement Learning](https://mlsanity.com/paper/2510.04786/) (2025)
- [Test-time Offline Reinforcement Learning on Goal-related Experience](https://mlsanity.com/paper/2507.18809/) (2025)
- [ACE-Merging: Data-Free Model Merging with Adaptive Covariance Estimation](https://mlsanity.com/paper/2603.02945/) (2026)
- [Specialization after Generalization: Towards Understanding Test-Time Training in Foundation Models](https://mlsanity.com/paper/2509.24510/) (2025)
- [Unveiling Language Routing Isolation in Multilingual MoE Models for Interpretable Subnetwork Adaptation](https://mlsanity.com/paper/2604.03592/) (2026)
- [Maximizing Prefix-Confidence at Test-Time Efficiently Improves Mathematical Reasoning](https://mlsanity.com/paper/2507.18122/) (2025)
- [Efficient Continual Learning in Language Models via Thalamically Routed Cortical Columns](https://mlsanity.com/paper/2602.22479/) (2026)
- [SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation](https://mlsanity.com/paper/2605.20189/) (2026)

## Abstract (arXiv)

Mixture of expert (MoE) models are a promising approach to increasing model capacity without increasing inference cost, and are core components of many state-of-the-art language models. However, current MoE models typically use only few experts due to prohibitive training and inference cost. We propose Test-Time Model Merging (TTMM) which scales the MoE paradigm to an order of magnitude more experts and uses model merging to avoid almost any test-time overhead. We show that TTMM is an approximation of test-time training (TTT), which fine-tunes an expert model for each prediction task, i.e., prompt. TTT has recently been shown to significantly improve language models, but is computationally expensive. We find that performance of TTMM improves with more experts and approaches the performance of TTT. Moreover, we find that with a 1B parameter base model, TTMM is more than 100x faster than TTT at test-time by amortizing the cost of TTT at train-time. Thus, TTMM offers a promising cost-effective approach to scale test-time training.