13 papers
Hyperparameter Scaling Laws Across MoE Sparsity
Changxin Tian, Kunlong Chen, Jia Liu +3
Mixture-of-Experts (MoE) models expand model capacity without a proportional increase in training compute, but increasing sparsity makes reliable hyperparameter transfer challengin…
Effective Learning Rate Governs Loss Dynamics in Language Model Pretraining
Zihan Liu, Ruiheng Zheng, Shaobo Zhang +4
We uncover ELR collapse in language model pretraining: learning rate (LR) and parameter norm govern loss dynamics primarily through their ratio, the effective learning rate (ELR).…
Rethinking Shrinkage Bias in LLM FP4 Pretraining: Geometric Origin, Systemic Impact, and UFP4 Recipe
Qian Zhao, Kunlong Chen, Changxin Tian +9
FP4 training promises substantial reductions in memory and computation cost for LLM pretraining, yet current FP4 hardware paths and recipes, including NVIDIA Blackwell/Rubin-class…
Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale
Ang Li, Ben Liu, Bin Han +215
Efficient and scalable agentic intelligence requires models that can deliver both low-latency responses and strong reasoning capabilities while remaining practical to train, serve,…
PowLU: An Activation Function for Stable Pre-Training of LLMs
Peijie Jiang, Yuqi Feng, Cunyin Peng +5
In contemporary large language models (LLMs), the swish-gated linear unit (SwiGLU) activation function is widely adopted to regulate the information flow and introduce non-linearit…
SuperValid: Capability-Aligned OOD Validation for Generalizable Downstream Scaling
Quanen Sun, Changxin Tian, Ke Shi +6
Scaling laws guide large language model training by relating compute to cross-entropy loss, and recent work further extends them to predict downstream benchmark performance. Howeve…
MergeMix: Optimizing Mid-Training Data Mixtures via Learnable Model Merging
Jiapeng Wang, Changxin Tian, Kunlong Chen +5
Optimizing data mixtures is essential for unlocking the full potential of large language models (LLMs), yet identifying the optimal composition remains computationally prohibitive…
Every Activation Boosted: Scaling General Reasoner to 1 Trillion Open Language Foundation
Ling Team, Ang Li, Ben Liu +138
We introduce Ling 2.0, a series reasoning-oriented language foundation built upon the principle that every activation boosts reasoning capability. Designed to scale from tens of bi…
MaP: A Unified Framework for Reliable Evaluation of Pre-training Dynamics
Jiapeng Wang, Changxin Tian, Kunlong Chen +5
Reliable evaluation is fundamental to the progress of Large Language Models (LLMs), yet the evaluation process during pre-training is plagued by significant instability that obscur…
Arrows of Math Reasoning Data Synthesis for Large Language Models: Diversity, Complexity and Correctness
Sirui Chen, Changxin Tian, Binbin Hu +4
Enhancing the mathematical reasoning of large language models (LLMs) demands high-quality training data, yet conventional methods face critical challenges in scalability, cost, and…
WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training
Changxin Tian, Jiapeng Wang, Qian Zhao +7
Recent advances in learning rate (LR) scheduling have demonstrated the effectiveness of decay-free approaches that eliminate the traditional decay phase while maintaining competiti…
Towards Greater Leverage: Scaling Laws for Efficient Mixture-of-Experts Language Models
Changxin Tian, Kunlong Chen, Jia Liu +3
Mixture-of-Experts (MoE) has become a dominant architecture for scaling Large Language Models (LLMs) efficiently by decoupling total parameters from computational cost. However, th…
BOSE: A Systematic Evaluation Method Optimized for Base Models
Hongzhi Luan, Changxin Tian, Zhaoxin Huan +4
This paper poses two critical issues in evaluating base models (without post-training): (1) Unstable evaluation during training: in the early stages of pre-training, the models lac…