7 papers
ISO: An RLVR-Native Optimization Stack
Hanqing Zhu, Wenyan Cong, Zhizhou Sha +8
Reinforcement learning with verifiable rewards (RLVR) advances the reasoning capabilities of language models, yet the optimization layer that converts reward feedback into weight-s…
AffectCodec: Emotion-Preserving Neural Speech Codec for Expressive Speech Modeling
Jiacheng Shi, Hongfei Du, Xinyuan Song +3
Neural speech codecs provide discrete representations for speech language models, but emotional cues are often degraded during quantization. Existing codecs mainly optimize acousti…
When Does Sparsity Mitigate the Curse of Depth in LLMs
Dilxat Muhtar, Xinyuan Song, Sebastian Pokutta +4
Recent work has demonstrated the curse of depth in large language models (LLMs), where later layers contribute less to learning and representation than earlier layers. Such under-u…
ActTail: Global Activation Sparsity in Large Language Models
Wenwen Hou, Xinyuan Song, Shiwei Liu +1
Activation sparsity is a promising approach for accelerating large language model (LLM) inference by reducing computation and memory movement. However, existing activation sparsity…
Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoning
Xinyuan Song, Keyu Wang, PengXiang Li +2
Recent studies suggest that the deeper layers of Large Language Models (LLMs) contribute little to representation learning and can often be removed without significant performance…
GPAS: Accelerating Convergence of LLM Pretraining via Gradient-Preserving Activation Scaling
Tianhao Chen, Xin Xu, Zijing Liu +12
Modern Large Language Models, such as the LLaMA, Qwen and DeepSeek series, predominantly adopt the Pre-LayerNorm (Pre-LN) Transformer architecture. While being stable during pretra…
The Curse of Depth in Large Language Models
Wenfang Sun, Xinyuan Song, Pengxiang Li +3
In this paper, we introduce the Curse of Depth, a concept that highlights, explains, and addresses the recent observation in modern Large Language Models (LLMs) where nearly half o…