# On Harnessing Idle Compute at the Edge for Foundation Model Training

arXiv: https://arxiv.org/abs/2512.22142 · HTML: https://mlsanity.com/paper/2512.22142/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 6 of its 10 references that are in the map.
- First posted on arXiv 13 Dec 2025.

## Builds on (most cited first, 6 of 6)

- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Decentralized Training of Foundation Models in Heterogeneous Environments](https://mlsanity.com/paper/2206.01288/) (2022)
- [NetLLM: Adapting Large Language Models for Networking](https://mlsanity.com/paper/2402.02338/) (2024)
- [AI Benchmark: Running Deep Neural Networks on Android Smartphones](https://mlsanity.com/paper/1810.01109/) (2018)
- [Asteroid: Resource-Efficient Hybrid Pipeline Parallelism for Collaborative DNN Training on Heterogeneous Edge Devices](https://mlsanity.com/paper/2408.08015/) (2024)
- [Towards Decentralized and Sustainable Foundation Model Training with the Edge](https://mlsanity.com/paper/2507.01803/) (2025)

## Abstract (arXiv)

The foundation-model ecosystem remains highly centralized because training requires immense compute resources and is therefore largely limited to large cloud operators. Edge-assisted foundation model training that harnesses spare compute on edge devices offers a more democratized alternative. However, existing edge-training approaches fall short: they struggle to match cloud-training performance, scale to larger models, fit within device memory limits, or keep communication overhead manageable. They also do not handle device heterogeneity and churn satisfactorily. We introduce Cleave, built on a structural insight: each GEMM has an asymmetric I/O pattern -- its input matrices, sent over downlink, are much larger than the partial output blocks returned over uplink -- matching edge networks where downlink bandwidth exceeds uplink by 2--10x. Exploiting this alignment with a parameter-server-centric architecture, Cleave makes per-device communication decrease as more devices join, rather than stay constant as in conventional TP. Decomposing training into independent sub-GEMM tasks yields one scheduling abstraction that unifies memory constraints, communication overhead, and fault tolerance under device churn. Our evaluation shows that Cleave achieves cloud-comparable GPU training performance and outperforms state-of-the-art edge-training methods by 4--10x in per-batch runtime at the same device counts. Beyond this shared operating range, Cleave scales to thousands of heterogeneous devices -- a regime where prior edge-training systems cannot operate -- and achieves at least 100x faster recovery from device failures.