Tag
This paper investigates the effectiveness of top-1 collapse rate as a stability monitor for short-horizon LoRA fine-tuning of discrete diffusion language models, finding it has zero precision, and proposes max gradient norm as a more reliable alternative with higher precision and F1 score on LLaDA-family models.
Proposes ARIADNE, a training-free, adapter-agnostic routing framework that selects the optimal PEFT adapter at inference time by measuring input proximity to adapter-specific centroids in embedding space, recovering 97.44% of upper-bound performance on 23 tasks.
Explores whether LoRA is the best parameter-efficient fine-tuning technique and introduces the PEFT library's tools to compare methods.
This paper benchmarks sub-1B models on mathematical reasoning tasks, revealing that full fine-tuning actively harms performance in models under 300M parameters, while parameter-efficient fine-tuning (PEFT) like LoRA and DoRA provides stability. The authors recommend defaulting to PEFT for all aligned sub-1B models and caution against full FT for architectures smaller than 500M to prevent catastrophic forgetting.
A desktop app that lets users correct model responses in chat and train LoRA adapters locally, closing the feedback loop without manual notebook work.
KappaTune, a fine-tuning method designed to mitigate catastrophic forgetting, has been integrated into Hugging Face's PEFT library.
Hugging Face's PEFT library enables parameter-efficient fine-tuning of large models on a single GPU, reducing compute and storage costs while maintaining performance.
The article introduces Echo-LoRA, a new parameter-efficient fine-tuning method that injects cross-layer representations from deeper source layers into shallow LoRA modules to improve performance without adding inference-time overhead.
The paper introduces CERSA, a novel parameter-efficient fine-tuning method that uses singular value decomposition to retain principal components, significantly reducing memory usage while outperforming existing methods like LoRA.
ShadowPEFT introduces a centralized parameter-efficient fine-tuning method that uses a depth-shared shadow module to refine transformer layer representations, matching or outperforming LoRA/DoRA with comparable trainable parameters.