The paper introduces Domain‑Specialized Upcycling (DS‑Upcycling), a method that expands a dense Transformer with domain‑specific experts and a router to improve multi‑domain adaptation. Unlike prior LoRA‑based adapters that either weight all domains equally or activate all experts, DS‑Upcycling trains a domain‑aware router to select the appropriate expert while freezing other parameters. Experiments demonstrate superior average performance across domains and high routing accuracy (90‑98%) with only 256 examples per domain.
Instruction-tuned models are deployed into environments where domains are heterogeneous and evolve, yet adding new domains or data typically requires costly retraining. We present AutoAdapt, a modular...
AutoAdapt is a modular framework that automatically discovers latent domains and trains per-domain Low‑Rank Adaptation (LoRA) adapters independently in parallel, without modifying other adapters. It uses parameter‑free routing to integrate new domains, achieving performance comparable to a LoRA adapter trained on all domains while avoiding full‑model retraining. The approach prevents domain interference by construction, enabling taxonomy‑free domain specialization across 14 benchmarks and GPT‑4o evaluations.
The paper investigates whether the sparsity of Mixture-of-Experts (MoE) models leads to intrinsic semantic organization across modalities and domains. It shows that experts naturally specialize semantically even without explicit modular training. The authors propose ExpertLens, a data‑free method that decodes router weights to identify domain‑specialized experts, enabling selective fine‑tuning that matches or exceeds full fine‑tuning while updating only 21.7–47.0% of parameters and achieving a 4.0× speedup, outperforming LoRA in both performance and efficiency.
arXiv:2606. 06098v1 Announce Type: cross Abstract: Foundational Large Language Models (LLMs) demonstrate proficiency on a wide range of general tasks, and achieve remarkable results on various specialized tasks via domain-expert LLMs.
arXiv:2606. 31413v1 Announce Type: new Abstract: 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.
arXiv:2610.34756v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) has emerged as the dominant paradigm for parameter-efficient fine-tuning of large language models. Standard all-layer LoRA...
arXiv:2608. 16384v1 Announce Type: cross Abstract: Universal visual representations require adaptation mechanisms that adapt across heterogeneous domains without fragmenting knowledge into domain-specific modules.
arXiv:2606. 08970v1 Announce Type: new Abstract: Vision-language models (VLMs) with varying performance and resource requirements are widely deployed, making it difficult for users to select the most appropriate one among numerous VLM candidates.
arXiv:2602. 05988v2 Announce Type: replace Abstract: Pre-training Large Language Models (LLMs) on web-scale datasets becomes fundamental for advancing general-purpose AI.
arXiv:2410. 13077v2 Announce Type: replace-cross Abstract: Transformer-based Large Language Models (LLMs) traditionally rely on final-layer loss for finetuning and final-layer representations for predictions, potentially overlooking the predictive power embedded in late layers.
DivMoE is a new framework that enables fine‑grained mixture‑of‑experts (MoE) upcycling from pre‑trained dense language models while avoiding the structural collapse that plagues naive approaches. It achieves this by initializing experts from domain‑adapted continual pre‑training and enforcing diversity‑constrained routing that forces each token to activate experts from distinct domain groups. Across two base models and 15 benchmarks, DivMoE consistently outperforms six upcycling baselines and improves over the dense base model on every benchmark after Stage 2 continual pre‑training, ultimately matching a larger 16B‑parameter MoE model with only 12B parameters.
arXiv:2606. 16456v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) models enable efficient scaling, but training them from scratch remains prohibitively expensive.