6 papers
Multi-Head Attention as a Source of Catastrophic Forgetting in MoE Transformers
Anrui Chen, Ruijun Huang, Xin Zhang +15
Mixture-of-Experts (MoE) architectures are often considered a natural fit for continual learning because sparse routing should localize updates and reduce interference, yet MoE Tra…
SD-MoE: Spectral Decomposition for Effective Expert Specialization
Ruijun Huang, Fang Dong, Xin Zhang +16
Mixture-of-Experts (MoE) architectures scale Large Language Models via expert specialization induced by conditional computation. In practice, however, expert specialization often f…
Dispelling the Curse of Singularities in Neural Network Optimizations
Hengjie Cao, Mengyi Chen, Yifeng Yang +11
This work investigates the optimization instability of deep neural networks from a less-explored yet insightful perspective: the emergence and amplification of singularities in the…
Spectra: Rethinking Optimizers for LLMs Under Spectral Anisotropy
Zhendong Huang, Hengjie Cao, Fang Dong +14
Gradient signals in LLM training are highly anisotropic: recurrent linguistic structure concentrates energy into a small set of dominant spectral directions, while context specific…
MixServe: An Automatic Distributed Serving System for MoE Models with Hybrid Parallelism Based on Fused Communication Algorithm
Bowen Zhou, Jinrui Jia, Wenhao He +2
The Mixture of Experts (MoE) models are emerging as the latest paradigm for Large Language Models (LLMs). However, due to memory constraints, MoE models with billions or even trill…
Metis: Training LLMs with FP4 Quantization
Hengjie Cao, Mengyi Chen, Yifeng Yang +13
This work identifies anisotropy in the singular value spectra of parameters, activations, and gradients as the fundamental barrier to low-bit training of large language models (LLM…