EUVIMEDEUVIMED entdecken
Crossref

DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization

Hyeonjun An, Sihyun Kim, Chaerim Lim, Hyunjoon Kim, Rathijit Sen, Sangmin Jung, Hyeonsoo Lee, Dongwook Kim, Takki Yu, Jinkyu Jeong, Youngsok Kim, Kwanghyun Park

Proceedings of the ACM on Management of Data · 18.05.2026 · 4 (3) · 1-29

DOI: 10.1145/3802037 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the ACM on Management of Data, online erschienen am 18.05.2026. Verfasst von Hyeonjun An und 11 weiteren Autor:innen. Die Arbeit nennt 73 Literaturangaben. Gefördert von: National Research Foundation of Korea, Korea Basic Science Institute \/ National Research Facilities and Equipment Center, Yonsei University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multimodal Large Language Models (MLLMs) have achieved remarkable advances by integrating text, image, and audio understanding within a unified architecture. However, existing distributed training frameworks remain fundamentally data-blind: they parallelize computation without accounting for variations in input data characteristics. This data unawareness leads to severe computation skew across stages and microbatches, where heterogeneous multimodal inputs incur different processing costs. Consequently, GPU resources are unevenly utilized, synchronization delays accumulate, and overall training efficiency degrades. To address this limitation, we present DFLOP, a d ata-driven f ramework for multimodal L LM training pipeline op timization. DFLOP continuously profiles runtime behavior to capture data-induced computation variance and employs predictive scheduling to balance workloads across stages and microbatches. By coupling data characteristics with execution planning, DFLOP substantially improves GPU utilization and throughput. Extensive experiments on large-scale multimodal benchmarks show that DFLOP achieves up to 3.6× faster training compared to state-of-the-art distributed training frameworks.

Förderung

  • National Research Foundation of Korea
  • Korea Basic Science Institute \/ National Research Facilities and Equipment Center
  • Yonsei University

Zitieren

APA

An, H., Kim, S., Lim, C., Kim, H., Sen, R., Jung, S., Lee, H., Kim, D., Yu, T., Jeong, J., Kim, Y., & Park, K. (2026). DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization. Proceedings of the ACM on Management of Data, 4(3), 1-29. https://doi.org/10.1145/3802037

Vancouver

An H, Kim S, Lim C, Kim H, Sen R, Jung S, et al. DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization. Proceedings of the ACM on Management of Data. 2026;4(3):1-29. doi:10.1145/3802037

BibTeX
@article{An2026,
  author = {An, Hyeonjun and Kim, Sihyun and Lim, Chaerim and Kim, Hyunjoon and Sen, Rathijit and Jung, Sangmin and Lee, Hyeonsoo and Kim, Dongwook and Yu, Takki and Jeong, Jinkyu and Kim, Youngsok and Park, Kwanghyun},
  title = {DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization},
  journal = {Proceedings of the ACM on Management of Data},
  year = {2026},
  volume = {4},
  number = {3},
  pages = {1--29},
  doi = {10.1145/3802037}
}
Kostenlos starten →