Research map / Large language models / Instruction and fine-tuning
PLoRA: Efficient Concurrent LoRA Training for Large Language Models
Minghao Yan, Zhuang Wang, Zhen Jia, Shivaram Venkataraman, Yida Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pack multiple LoRA configs into one job (planner only, sequential execution) (ablation) | PLoRA · Qwen2.5-3B/7B, 120 configs | makespan: around 1.8x reduction | Improved |
| add optimized packed LoRA kernels (ablation) | Sequential PLoRA · Qwen2.5-3B/7B, 120 configs | makespan: up to 3.93x further reduction | Improved |
| use Max GPU (all 8 GPUs per job) instead of Min GPU (ablation) | sequential LoRA baseline · Qwen2.5, 120 configs | makespan | Hurt |
| increase batch size from 1 to 4 (ablation) | PLoRA throughput gain · Qwen2.5 job throughput | speedup over Min GPU: 12.8x at BS 1, smaller at larger BS | Mixed |
| search 120 LoRA configs instead of default config (ablation) | LoRA fine-tuning · cola, Qwen2.5-3B | accuracy: 53.8 -> 77.2 | Improved |
| search 120 LoRA configs instead of default config (ablation) | LoRA fine-tuning · gsm8k, Qwen2.5-7B | accuracy: 72.1 -> 79.8 | Improved |
| increase LoRA batch size (ablation) | LoRA fine-tuning · mrpc/cola/wnli/gsm8k, Qwen2.5-7B | accuracy: up to 11.3% accuracy difference | Hurt |
| vary learning rate (ablation) | LoRA fine-tuning · cola, Qwen2.5-7B | accuracy: 14.2% max accuracy difference | Mixed |
| vary LoRA rank (ablation) | LoRA fine-tuning · cola, Qwen2.5-7B | accuracy: 13.1% max accuracy difference | Mixed |
| vary LoRA alpha (ablation) | LoRA fine-tuning · Qwen2.5-7B | accuracy: up to 5.9% max accuracy difference | Mixed |
| apply mrpc best config to gsm8k (ablation) | LoRA fine-tuning · Qwen2.5-3B | accuracy: -7.4% | Hurt |
| transfer cola best config from 7B to 3B (ablation) | LoRA fine-tuning · cola | accuracy: -3.6% | Hurt |
Closest papers by meaning
- Uni-LoRA: One Vector is All You Need (NeurIPS 2025 · similarity 0.93)
- LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades (NeurIPS 2025 · similarity 0.93)
- Learning Rate Matters: Vanilla LoRA May Suffice for LLM Fine-tuning (NeurIPS 2026 · similarity 0.93)
- BSLoRA: Enhancing the Parameter Efficiency of LoRA with Intra-Layer and Inter-Layer Sharing (ICML 2025 · similarity 0.92)
- Loquetier: A Virtualized Multi-LoRA Framework for Unified LLM Fine-tuning and Serving (NeurIPS 2025 · similarity 0.92)
- Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment (ICML 2025 · similarity 0.92)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
