Effective Distillation to Hybrid xLSTM Architectures
There have been numerous attempts to distill quadratic attention-based large language models (LLMs) into sub-quadratic linearized architectures. However, despite extensive research, such distilled models often fail to match the performance of their teacher LLMs on various downstream tasks. We set out the goal of lossless distillation, which we define in terms of tolerance-corrected Win-and-Tie rates between student and teacher on sets of tasks. To this end, we introduce an effective distillation pipeline for xLSTM-based students. We propose an additional merging stage, where individually linearized experts are combined into a single model. We show the effectiveness of this pipeline by distilling base and instruction-tuned models from the Llama, Qwen, and Olmo families. In many settings, our xLSTM-based students recover most of the teacher's performance, and even exceed it on some downstream tasks. Our contributions are an important step towards more energy-efficient and cost-effective replacements for transformer-based LLMs.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
On Subquadratic Architectures: From Applications to Principles
Transformers dominate modern sequence modeling, but their quadratic attention incurs substantial computational cost. Subquadratic architectures offer a scalable alternative. However, it remains unclear which designs yiel…
Are Vision xLSTM Embedded UNet More Reliable in Medical 3D Image Segmentation?
The development of efficient segmentation strategies for medical images has evolved from its initial dependence on Convolutional Neural Networks (CNNs) to the current investigation of hybrid models that combine CNNs with…
Image SegmentationMedical Image SegmentationSegmentationSemantic SegmentationxLSTM-SENet: xLSTM for Single-Channel Speech Enhancement
While attention-based architectures, such as Conformers, excel in speech enhancement, they face challenges such as scalability with respect to input sequence length. In contrast, the recently proposed Extended Long Short…
MambaSpeech EnhancementShort window attention enables long-term memorization
Recent works show that hybrid architectures combining local sliding window attention layers and global attention layers outperform either of these architectures taken separately. However, the impact of the window length …
xLSTM-UNet can be an Effective 2D & 3D Medical Image Segmentation Backbone with Vision-LSTM (ViL) better than its Mamba Counterpart
Convolutional Neural Networks (CNNs) and Vision Transformers (ViT) have been pivotal in biomedical image segmentation, yet their ability to manage long-range dependencies remains constrained by inherent locality and comp…
3D Medical Imaging Segmentationimage-classificationImage ClassificationImage Segmentation+5