7 papers · 1 filter
Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO
Yunpeng Ba, Zhi Zheng, Yue Xie +7
Evolution Strategies (ES) have recently emerged as a memory-efficient post-training paradigm for LLM reasoning. However, the optimization behavior of ES remains understudied, makin…
ASymPO: Asymmetric-Scale Policy Optimization for Asynchronous LLM Post-Training Without Behavior Information
Zehua Liu, Yuxuan Yao, Xiaojin Fu +2
Asynchronous reinforcement learning can improve language-model post-training throughput by decoupling response generation from policy optimization, but stale responses introduce di…
Automated Optimization Modeling via a Localizable Error-Driven Perspective
Weiting Liu, Han Wu, Yufei Kuang +4
Automated optimization modeling via Large Language Models (LLMs) has emerged as a promising approach to assist complex human decision-making. While post-training has become a pivot…
RIFT: Repurposing Negative Samples via Reward-Informed Fine-Tuning
Zehua Liu, Shuqi Liu, Tao Zhong +1
While Supervised Fine-Tuning (SFT) and Rejection Sampling Fine-Tuning (RFT) are standard for LLM alignment, they either rely on costly expert data or discard valuable negative samp…
REG: A Regularization Optimizer for Robust Training Dynamics
Zehua Liu, Han Wu, Xiaojin Fu +4
Optimizers are crucial for the efficient training of Large Language Models (LLMs). While AdamW is the de facto standard, recent structure-aware optimizers like Muon have emerged, w…
Automatic Operator-level Parallelism Planning for Distributed Deep Learning -- A Mixed-Integer Programming Approach
Ruifeng She, Bowen Pang, Kai Li +2
As the artificial intelligence community advances into the era of large models with billions of parameters, distributed training and inference have become essential. While various…
MoLAE: Mixture of Latent Experts for Parameter-Efficient Language Models
Zehua Liu, Han Wu, Ruifeng She +4
Mixture of Experts (MoE) has become a key architectural paradigm for efficiently scaling Large Language Models (LLMs) by selectively activating a subset of parameters for each inpu…