paper-with-me

홈 › Papers

Decomposing Evolutionary Mixture-of-LoRA Architectures: The Routing Lever, the Lifecycle Penalty, and a Substrate-Conditional Boundary

2026-05-11 · Ramchand Kumaresan arxiv

We decompose an evolutionary mixture-of-LoRA system on a from-scratch ~150M-parameter widened-D substrate (D=1536, V=32000; D/V approx 0.048; the "widened-1536" substrate) into three factors -- a router rewrite (parallel sigmoid gate with learnable per-adapter floor and bounded temperature anneal, fed post-stack hidden states rather than token-embedding means), a per-domain leave-one-out evaluation scope, and a lifecycle of death plus alpha-blend inheritance plus SVD mutation plus slot reallocation -- and report a 5-of-8 partial 2^3 factorial run at n=3 seeds and 25000 adaptation steps per cell. The attribution chain is sharp on this substrate: the router rewrite carries the entire +0.0426 nat balanced log-PPL improvement (Delta = log PPL_ref - log PPL_test, positive = improvement; t=12.86, p=0.006) attributed to "the full evolutionary system vs the static B3 baseline"; the headline full-system-vs-B3 balanced contrast itself is +0.015 nats, t=1.94, p=0.19 at n=3 and does not clear alpha=0.05. The per-domain evaluation scope is null at seed-resolution, and the lifecycle is a net drag of approx -0.028 nats (t=-4.46,p=0.047 in the primary chain). An auxiliary alpha=0 inheritance counterfactual at n=3 seeds is sign-inconsistent at the headline metric and underpowered for either an equivalence or load-bearing conclusion (corrected from an earlier arithmetic-mean aggregator that erroneously cleared inheritance; see Appendix B.11). A base-perturbation probe directionally refutes a "genomic-context" reframe of the lifecycle role. A controllable synthetic sandbox locates a substrate-conditional regime boundary: evolutionary search on the routing channel is load-bearing only when adapters are pre-aligned to the task; in every other regime tested it underperforms, ties, or actively degrades the gradient solution.

📄 PDF Abstract BibTeX arXiv:2605.11153

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging

2026-02-28 · Jie Cao, Zhenxuan Fan, Zhuonan Wang, Tianwei Lin 외 arxiv

Large language models (LLMs) achieve remarkable performance on diverse downstream and domain-specific tasks via parameter-efficient fine-tuning (PEFT). However, existing PEFT methods, particularly MoE-LoRA architectures,…

parameter-efficient fine-tuning

TalkLoRA: Communication-Aware Mixture of Low-Rank Adaptation for Large Language Models

2026-04-07 · Lin Mu, Haiyang Wang, Li Ni, Lei Sang 외 arxiv

Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning of Large Language Models (LLMs), and recent Mixture-of-Experts (MoE) extensions further enhance flexibility by dynamically combining multiple LoRA expert…

parameter-efficient fine-tuning

ReMix: Reinforcement routing for mixtures of LoRAs in LLM finetuning

2026-03-10 · Ruizhong Qiu, Hanqing Zeng, Yinglong Xia, Yiwen Meng 외 arxiv

Low-rank adapters (LoRAs) are a parameter-efficient finetuning technique that injects trainable low-rank matrices into pretrained models to adapt them to new tasks. Mixture-of-LoRAs models expand neural networks efficien…

Reinforcement Learning

MoE-Sieve: Routing-Guided LoRA for Efficient MoE Fine-Tuning

2026-03-25 · Andrea Manzoni arxiv

Standard LoRA fine-tuning of Mixture-of-Experts (MoE) models applies adapters to every expert, yet our profiling shows that per-layer expert routing is highly skewed: a small subset of experts handles most tokens in each…

Learning to Select, Not Relearn: Hard-Routed Mixtures of Reasoning LoRAs

2026-06-30 · Seyed Alireza Molavi, Zhan Su, Yan Hu, Peyman Sheikholharam Mashhadi 외 arxiv

Composing independently trained LoRA adapters into a single large language model is useful for multi-domain adaptation, especially when the original training data cannot be shared. A common approach is to use MoE-style r…

Reinforcement LearningDomain Adaptation