# Streaming DiLoCo with overlapping communication: Towards a Distributed Free Lunch

arXiv: https://arxiv.org/abs/2501.18512 · HTML: https://mlsanity.com/paper/2501.18512/

## Key facts

- Cited by 28 papers in the map (data through 14 Oct 2026).
- Builds on 57 of its 68 references that are in the map.
- First posted on arXiv 30 Jan 2025.

## Builds on (most cited first, 30 of 57)

- [Communication-Efficient Learning of Deep Networks from Decentralized Data](https://mlsanity.com/paper/1602.05629/) (2016)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Advances and Open Problems in Federated Learning](https://mlsanity.com/paper/1912.04977/) (2019)
- [Training Compute-Optimal Large Language Models](https://mlsanity.com/paper/2203.15556/) (2022)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time](https://mlsanity.com/paper/2203.05482/) (2022)
- [Adaptive Federated Optimization](https://mlsanity.com/paper/2003.00295/) (2020)
- [TIES-Merging: Resolving Interference When Merging Models](https://mlsanity.com/paper/2306.01708/) (2023)
- [Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch](https://mlsanity.com/paper/2311.03099/) (2023)
- [Linear Mode Connectivity and the Lottery Ticket Hypothesis](https://mlsanity.com/paper/1912.05671/) (2019)
- [Merging Models with Fisher-Weighted Averaging](https://mlsanity.com/paper/2111.09832/) (2021)
- [Local SGD Converges Fast and Communicates Little](https://mlsanity.com/paper/1805.09767/) (2018)
- [Federated Learning with Personalization Layers](https://mlsanity.com/paper/1912.00818/) (2019)
- [Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training](https://mlsanity.com/paper/1712.01887/) (2017)
- [Git Re-Basin: Merging Models modulo Permutation Symmetries](https://mlsanity.com/paper/2209.04836/) (2022)
- [GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection](https://mlsanity.com/paper/2403.03507/) (2024)
- [Dataless Knowledge Fusion by Merging Weights of Language Models](https://mlsanity.com/paper/2212.09849/) (2022)
- [What is being transferred in transfer learning?](https://mlsanity.com/paper/2008.11687/) (2020)
- [Query-Key Normalization for Transformers](https://mlsanity.com/paper/2010.04245/) (2020)
- [Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards](https://mlsanity.com/paper/2306.04488/) (2023)
- [PowerSGD: Practical Low-Rank Gradient Compression for Distributed Optimization](https://mlsanity.com/paper/1905.13727/) (2019)
- [Don't Use Large Mini-Batches, Use Local SGD](https://mlsanity.com/paper/1808.07217/) (2018)
- [Branch-Train-Merge: Embarrassingly Parallel Training of Expert Language Models](https://mlsanity.com/paper/2208.03306/) (2022)
- [Patching open-vocabulary models by interpolating weights](https://mlsanity.com/paper/2208.05592/) (2022)
- [Small-scale proxies for large-scale Transformer training instabilities](https://mlsanity.com/paper/2309.14322/) (2023)

## Built on by (most cited first, 28 of 28)

- [Communication-Efficient Language Model Training Scales Reliably and Robustly: Scaling Laws for DiLoCo](https://mlsanity.com/paper/2503.09799/) (2025)
- [MuLoCo: Muon is a practical inner optimizer for DiLoCo](https://mlsanity.com/paper/2505.23725/) (2025)
- [Eager Updates For Overlapped Communication and Computation in DiLoCo](https://mlsanity.com/paper/2502.12996/) (2025)
- [Weight Factorization and Centralization for Continual Learning in Speech Recognition](https://mlsanity.com/paper/2506.16574/) (2025)
- [Covenant-72B: Pre-Training a 72B LLM with Trustless Peers Over-the-Internet](https://mlsanity.com/paper/2603.08163/) (2026)
- [NoLoCo: No-all-reduce Low Communication Training Method for Large Models](https://mlsanity.com/paper/2506.10911/) (2025)
- [Smoothing DiLoCo with Primal Averaging for Faster Training of LLMs](https://mlsanity.com/paper/2512.17131/) (2025)
- [TAH-QUANT: Effective Activation Quantization in Pipeline Parallelism over Slow Network](https://mlsanity.com/paper/2506.01352/) (2025)
- [Decoupled DiLoCo for Resilient Distributed Pre-training](https://mlsanity.com/paper/2604.21428/) (2026)
- [Distributed and Decentralised Training: Technical Governance Challenges in a Shifting AI Landscape](https://mlsanity.com/paper/2507.07765/) (2025)
- [AsyncMesh: Fully Asynchronous Optimization for Data and Pipeline Parallelism](https://mlsanity.com/paper/2601.22442/) (2026)
- [MT-DAO: Multi-Timescale Distributed Adaptive Optimizers with Local Updates](https://mlsanity.com/paper/2510.05361/) (2025)
- [Factored Gossip DiLoCo: Reducing Blocking Communication in DiLoCo](https://mlsanity.com/paper/2606.22768/) (2026)
- [Partial Parameter Updates for Efficient Distributed Training](https://mlsanity.com/paper/2509.22418/) (2025)
- [LoRDO: Distributed Low-Rank Optimization with Infrequent Communication](https://mlsanity.com/paper/2602.04396/) (2026)
- [LOSCAR-SGD: Local SGD with Communication-Computation Overlap and Delay-Corrected Sparse Model Averaging](https://mlsanity.com/paper/2605.20866/) (2026)
- [Unifying Local Communications and Local Updates for LLM Pretraining](https://mlsanity.com/paper/2606.11081/) (2026)
- [ResBM: Residual Bottleneck Models for Low-Bandwidth Pipeline Parallelism](https://mlsanity.com/paper/2604.11947/) (2026)
- [Outer-Momentum Restarting in High-Dimensional Two-Phase Optimization](https://mlsanity.com/paper/2605.28585/) (2026)
- [Does Distributed Training Undermine Compute Governance?](https://mlsanity.com/paper/2605.29359/) (2026)
- [Echelon: Auditable Aggregate-Only Language-Model Adaptation Across Privacy Boundaries](https://mlsanity.com/paper/2606.02958/) (2026)
- [Mixtures of Subspaces for Bandwidth Efficient Context Parallel Training](https://mlsanity.com/paper/2606.16384/) (2026)
- [FoMoE: Breaking the Full-Replica Barrier with a Federation of MoEs](https://mlsanity.com/paper/2606.19025/) (2026)
- [ShardMeter: Sharded and Geo-Distributed Training Without the Guesswork](https://mlsanity.com/paper/2608.23840/) (2026)
- [Distributed Training using an Intelligent Network](https://mlsanity.com/paper/2608.26453/) (2026)
- [Communication-Efficient LLM Adaptation over Decentralized GPU Meshes](https://mlsanity.com/paper/2609.14339/) (2026)
- [AutoLoCo: Communication Efficient Distributed LLM Training via Adaptive Synchronization](https://mlsanity.com/paper/2609.36662/) (2026)
- [How Inner Optimizers Shape Replica Scaling](https://mlsanity.com/paper/2610.19064/) (2026)

## Abstract (arXiv)

Training of large language models (LLMs) is typically distributed across a large number of accelerators to reduce training time. Since internal states and parameter gradients need to be exchanged at each and every single gradient step, all devices need to be co-located using low-latency high-bandwidth communication links to support the required high volume of exchanged bits. Recently, distributed algorithms like DiLoCo have relaxed such co-location constraint: accelerators can be grouped into ``workers'', where synchronizations between workers only occur infrequently. This in turn means that workers can afford being connected by lower bandwidth communication links without affecting learning quality. However, in these methods, communication across workers still requires the same peak bandwidth as before, as the synchronizations require all parameters to be exchanged across all workers. In this paper, we improve DiLoCo in three ways. First, we synchronize only subsets of parameters in sequence, rather than all at once, which greatly reduces peak bandwidth. Second, we allow workers to continue training while synchronizing, which decreases wall clock time. Third, we quantize the data exchanged by workers, which further reduces bandwidth across workers. By properly combining these modifications, we show experimentally that we can distribute training of billion-scale parameters and reach similar quality as before, but reducing required bandwidth by two orders of magnitude.