14 papers
One Model, Multiple Compute Budgets: Multi-Ratio ConceptMoE for Flexible Compute Across Turns
Yiyang Fu, Zihao Huang, Dongchen Han +4
Large language models typically support different inference budgets through separate model sizes or fixed compute configurations. In multi-turn interactions, however, compute deman…
Improved Large Language Diffusion Models
Shen Nie, Qiyang Min, Shaoxuan Xu +7
Modern large language models are predominantly trained with autoregressive factorization and causal attention. We present \emph{iLLaDA}, an 8B masked diffusion language model train…
StableVLA: Towards Robust Vision-Language-Action Models without Extra Data
Yiyang Fu, Chubin Zhang, Shukai Gong +7
It is infeasible to encompass all possible disturbances within the training dataset. This raises a critical question regarding the robustness of Vision-Language-Action (VLA) models…
ConceptMoE: Adaptive Token-to-Concept Compression for Implicit Compute Allocation
Zihao Huang, Jundong Zhou, Xingwei Qu +2
Large language models allocate uniform computation across all tokens, ignoring that some sequences are trivially predictable while others require deep reasoning. We introduce Conce…
Dynamic Large Concept Models: Latent Reasoning in an Adaptive Semantic Space
Xingwei Qu, Shaowen Wang, Zihao Huang +16
Large Language Models (LLMs) apply uniform computation to all tokens, despite language exhibiting highly non-uniform information density. This token-uniform regime wastes capacity…
Virtual Width Networks
Seed, Baisheng Li, Banggu Wu +115
We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN d…
Scaling Latent Reasoning via Looped Language Models
Rui-Jie Zhu, Zixuan Wang, Kai Hua +30
Modern LLMs are trained to "think" primarily via explicit text generation, such as chain-of-thought (CoT), which defers reasoning to post-training and under-leverages pre-training…
SeeDNorm: Self-Rescaled Dynamic Normalization
Wenrui Cai, Defa Zhu, Qingjie Liu +1
Normalization layer constitutes an essential component in neural networks. In transformers, the predominantly used RMSNorm constrains vectors to a unit hypersphere, followed by dim…
UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning
Zihao Huang, Yu Bao, Qiyang Min +8
While Mixture of Experts (MoE) models achieve remarkable efficiency by activating only subsets of parameters, they suffer from high memory access costs during inference. Memory-lay…
Frac-Connections: Fractional Extension of Hyper-Connections
Defa Zhu, Hongzhi Huang, Jundong Zhou +5
Residual connections are central to modern deep learning architectures, enabling the training of very deep networks by mitigating gradient vanishing. Hyper-Connections recently gen…
Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts
Yike Yuan, Ziyu Wang, Zihao Huang +4
Diffusion models have emerged as mainstream framework in visual generation. Building upon this success, the integration of Mixture of Experts (MoE) methods has shown promise in enh…
Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling
Hongzhi Huang, Defa Zhu, Banggu Wu +4
Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-…
Ultra-Sparse Memory Network
Zihao Huang, Qiyang Min, Hongzhi Huang +4
It is widely acknowledged that the performance of Transformer models is logarithmically related to their number of parameters and computational complexity. While approaches like Mi…
Hyper-Connections
Defa Zhu, Hongzhi Huang, Zihao Huang +5
We present hyper-connections, a simple yet effective method that can serve as an alternative to residual connections. This approach specifically addresses common drawbacks observed…