LobRA: Multi-tenant Fine-tuning over Heterogeneous Data
Summary: LobRA enables multi-tenant joint fine-tuning of LoRA adapters by tackling two data heterogeneities—sequence-length variation and skew—that hurt joint FT efficiency. It runs heterogeneous FT replicas (varying resource/parallel configs) plus a skew-aware dispatcher to balance per-step workloads, cutting GPU-seconds by ~45–60.7%. (summarized by gpt-5-mini on Feb 09 2026)
Incoming Non-self Citations Over Time
No non-self incoming citations found for this paper in this database.
Authors
- 1. Sheng Lin (Peking University)
- 2. Fangcheng Fu (Shanghai Jiao Tong University)
- 3. Haoyang Li (Peking University)
- 4. Hao Ge (Peking University)
- 5. Xuanyu Wang (Peking University)
- 6. Jiawen Niu (Peking University)
- 7. Yaofeng Tu (ZTE Corporation)
- 8. Bin Cui (Peking University)
BibTeX Citation
@article{lin_vldb25,
title = {{LobRA: Multi-tenant Fine-tuning over Heterogeneous Data}},
author = {Lin, Sheng and Fu, Fangcheng and Li, Haoyang and Ge, Hao and Wang, Xuanyu and Niu, Jiawen and Tu, Yaofeng and Cui, Bin},
journal = {PVLDB},
series = {{VLDB} '25},
volume = {18},
number = {8},
pages = {2616--2625},
doi = {10.14778/3742728.3742752},
url = {https://doi.org/10.14778/3742728.3742752},
year = {2025}
}Incoming Citations (Sorted by Pagerank)
Showing 1 of 1 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 10,588 | Hydraulis: Balancing Large Transformer Model Training via Co-designing Parallel Strategies and Data Assignment | 2026 | SIGMOD | 4.9769913e-05 |
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 9 of 9 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 523 | PyTorch Distributed: Experiences on Accelerating Data Parallel Training | 2020 | VLDB | 0.000169044 |
| 1,659 | D-Bot: Database Diagnosis System using Large Language Models | 2024 | VLDB | 9.9625133e-05 |
| 2,227 | GPTuner: A Manual-Reading Database Tuning System via GPT-Guided Bayesian Optimization | 2024 | VLDB | 8.8007923e-05 |
| 2,247 | PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel | 2023 | VLDB | 8.7597965e-05 |
| 3,450 | MiCS: Near-linear Scaling for Training Gigantic Model on Public Cloud | 2023 | VLDB | 7.2889709e-05 |
| 5,011 | Galvatron: Efficient Transformer Training over Multiple GPUs Using Automatic Parallelism | 2023 | VLDB | 6.3131083e-05 |
| 5,999 | BAGUA: Scaling up Distributed Learning with System Relaxations | 2022 | VLDB | 5.9170896e-05 |
| 8,205 | Angel-PTM: A Scalable and Economical Large-scale Pre-training System in Tencent | 2023 | VLDB | 5.3769281e-05 |
| 8,896 | mLoRA: Fine-Tuning LoRA Adapters via Highly-Efficient Pipeline Parallelism in Multiple GPUs | 2025 | VLDB | 5.2534908e-05 |
Previous
Page 1 / 1
Next
Semantically Similar Papers
| # | Overall Rank | Paper | Year | Venue |
|---|---|---|---|---|
| 1 | 11,081 | Libra: One-Shot Parameter Sensitivity Estimation for Transfer Learning in Database Performance Prediction | 2026 | VLDB |
| 2 | 7,097 | DLRover-RM: Resource Optimization for Deep Recommendation Models Training in the Cloud | 2024 | VLDB |
| 3 | 10,670 | Mixtera: A Data Plane for Foundation Model Training | 2026 | SIGMOD |
| 4 | 11,350 | CoLA: Model Collaboration for Log-based Anomaly Detection | 2025 | VLDB |
| 5 | 11,297 | Heta: Distributed Training of Heterogeneous Graph Neural Networks | 2025 | VLDB |
| 6 | 2,186 | Heterogeneity-aware Distributed Parameter Servers | 2017 | SIGMOD |
| 7 | 10,127 | RALF: Accuracy-Aware Scheduling for Feature Store Maintenance | 2024 | VLDB |
| 8 | 10,447 | DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization | 2026 | SIGMOD |
| 9 | 8,445 | LFTF: A Framework for Efficient Tensor Analytics at Scale | 2017 | VLDB |
| 10 | 8,896 | mLoRA: Fine-Tuning LoRA Adapters via Highly-Efficient Pipeline Parallelism in Multiple GPUs | 2025 | VLDB |