Pith. sign in

REVIEW 7 cited by

SlowMo: Improving Communication-Efficient Distributed SGD with Slow Momentum

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1910.00643 v2 pith:6LIYUUFC submitted 2019-10-01 cs.LG cs.DCmath.OCstat.ML

Jianyu Wang , Vinayak Tantia , Nicolas Ballas , Michael Rabbat This is my paper · ORCID
classification cs.LGcs.DCmath.OCstat.ML
keywords slowmobasebmufdistributedmethodsmomentummultipleoptimization
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Distributed optimization is essential for training large models on large datasets. Multiple approaches have been proposed to reduce the communication overhead in distributed training, such as synchronizing only after performing multiple local SGD steps, and decentralized methods (e.g., using gossip algorithms) to decouple communications among workers. Although these methods run faster than AllReduce-based methods, which use blocking communication before every update, the resulting models may be less accurate after the same number of updates. Inspired by the BMUF method of Chen & Huo (2016), we propose a slow momentum (SlowMo) framework, where workers periodically synchronize and perform a momentum update, after multiple iterations of a base optimization algorithm. Experiments on image classification and machine translation tasks demonstrate that SlowMo consistently yields improvements in optimization and generalization performance relative to the base optimizer, even when the additional overhead is amortized over many updates so that the SlowMo runtime is on par with that of the base optimizer. We provide theoretical convergence guarantees showing that SlowMo converges to a stationary point of smooth non-convex losses. Since BMUF can be expressed through the SlowMo framework, our results also correspond to the first theoretical convergence guarantees for BMUF.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Overcoming the Communication-Performance Tradeoff in LLM Pretraining

    cs.LG 2025-08 conditional novelty 7.0 of 10

    SparseLoCo combines error feedback with Top-k sparsification and 2-bit quantization to send 1-3% of the pseudo-gradient during LLM pre-training while matching or beating DiLoCo's dense updates.

  2. FedACT: Federated Adaptive Coordinate Trust Modulation for Robust Transformer Training under Data Heterogeneity

    cs.LG 2026-07 conditional novelty 6.5 of 10

    Global-aware coordinate trust modulation after corrected AdamW updates improves federated Transformer and LLM training under data heterogeneity over strong adaptive baselines.

  3. Decentralized SGD with Controlled Disagreement Finds Flatter Minima

    cs.LG 2026-02 conditional novelty 6.0 of 10

    Keeping consensus errors alive in decentralized SGD via a learning-rate-scaled mixing term improves test accuracy and flatter minima over both DSGD and synchronous SGD.

  4. FedWCM: Unleashing the Potential of Momentum-based Federated Learning in Long-Tailed Scenarios

    cs.LG 2025-07 reject novelty 6.0 of 10

    FedWCM uses per-client data-distribution scores to adapt momentum and aggregation weights in federated learning, showing empirical gains over FedAvg and FedCM on long-tailed non-IID datasets, but its convergence proof...

  5. HALoS: Hierarchical Asynchronous Local SGD over Slow Networks for Geo-Distributed Large Language Model Training

    cs.LG 2025-06 conditional novelty 6.0 of 10

    A hierarchical asynchronous local SGD method with regional parameter servers and global model merging is claimed to train small LLMs up to 7.5x faster than DiLoCo in simulated geo-distributed settings.

  6. MuLoCo: Muon is a practical inner optimizer for DiLoCo

    cs.LG 2025-05 conditional novelty 5.0 of 10

    Using Muon instead of AdamW inside DiLoCo improves worker scaling and critical batch size for LLM pre-training across 150M to 15B parameters.

  7. Federated Continual Learning: Concepts, Challenges, and Solutions

    cs.LG 2025-02 conditional novelty 1.0 of 10

    A literature review that categorizes challenges and solutions in federated continual learning and adds an experimental comparison of aggregation strategies.

Pith tools