LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913120786579456 |
|---|---|
| author | Li, Wenbing Song, Zikai Zhou, Hang Zhang, Yunyao Yu, Junqing Yang, Wei |
| author_facet | Li, Wenbing Song, Zikai Zhou, Hang Zhang, Yunyao Yu, Junqing Yang, Wei |
| contents | Recent attempts to combine low-rank adaptation (LoRA) with mixture-of-experts (MoE) for multi-task adaptation of Large Language Models (LLMs) often replace whole attention/FFN layers with switch experts or append parallel expert branches, undermining parameter efficiency and limiting task specialization. We introduce LoRA-Mixer, a modular MoE framework that routes task-specific LoRA experts into the core projection matrices of the attention module, namely input and output linear layers, rather than primarily targeting FFN blocks. The design delivers fine-grained token-level specialization by fully exploiting the attention mechanism, while remaining drop-in compatible with Transformers and state-space models (SSMs), since linear projection layers are ubiquitous. To train robust routers from limited data while promoting stable, selective decisions and high expert reuse, LoRA-Mixer employs an adaptive Routing Specialization Loss (RSL) that jointly enforces global load balance and input-aware specialization via an entropy-shaping objective. The framework supports two regimes: (i) joint optimization of adapters and router with a differentiable hard-soft top-k routing scheme, and (ii) plug-and-play routing over frozen, pre-trained LoRA modules sourced from public repositories. Across 15 benchmarks, including MedQA, GSM8K, HumanEval, and GLUE, RSL-optimized LoRA-Mixer outperforms state-of-the-art routing and LoRA-MoE baselines while using 48 percent of their trainable parameters, with gains of 3.79, 2.90, and 3.95 percentage points on GSM8K, CoLA, and ARC-C, respectively. Cross-model transfer and adapter reuse experiments further demonstrate the approach's versatility and data efficiency. Our code is available at https://github.com/hustcselwb/LoRA-Mixer. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_00029 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing Li, Wenbing Song, Zikai Zhou, Hang Zhang, Yunyao Yu, Junqing Yang, Wei Machine Learning Artificial Intelligence Recent attempts to combine low-rank adaptation (LoRA) with mixture-of-experts (MoE) for multi-task adaptation of Large Language Models (LLMs) often replace whole attention/FFN layers with switch experts or append parallel expert branches, undermining parameter efficiency and limiting task specialization. We introduce LoRA-Mixer, a modular MoE framework that routes task-specific LoRA experts into the core projection matrices of the attention module, namely input and output linear layers, rather than primarily targeting FFN blocks. The design delivers fine-grained token-level specialization by fully exploiting the attention mechanism, while remaining drop-in compatible with Transformers and state-space models (SSMs), since linear projection layers are ubiquitous. To train robust routers from limited data while promoting stable, selective decisions and high expert reuse, LoRA-Mixer employs an adaptive Routing Specialization Loss (RSL) that jointly enforces global load balance and input-aware specialization via an entropy-shaping objective. The framework supports two regimes: (i) joint optimization of adapters and router with a differentiable hard-soft top-k routing scheme, and (ii) plug-and-play routing over frozen, pre-trained LoRA modules sourced from public repositories. Across 15 benchmarks, including MedQA, GSM8K, HumanEval, and GLUE, RSL-optimized LoRA-Mixer outperforms state-of-the-art routing and LoRA-MoE baselines while using 48 percent of their trainable parameters, with gains of 3.79, 2.90, and 3.95 percentage points on GSM8K, CoLA, and ARC-C, respectively. Cross-model transfer and adapter reuse experiments further demonstrate the approach's versatility and data efficiency. Our code is available at https://github.com/hustcselwb/LoRA-Mixer. |
| title | LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2507.00029 |