7 papers
Turning Off-Policy Tokens On-Policy: A Plug-in Approach for Improving LLM Alignment
Yu Li, Xiuyu Li, Mingyang Yi +4
Reinforcement learning (RL) post-training for large language models (LLMs) follows a efficient paradigm of "rollout then update", which inevitably results in off-policy training da…
Rethinking Calibration for High-Sparsity LLM Pruning: From Single Sources to Capability Coverage
Hu Xu, Zhaolong Xing, Congcong Liu +5
Post-training LLM pruning relies on a small calibration set, usually chosen heuristically and judged by WikiText perplexity or by mean accuracy on general-domain commonsense benchm…
FlowTrain: Flow-Based Decoupled Training for Industrial-Grade Vision-Language Models
Zhida Jiang, Zhaolong Xing, Yang Pei +14
Industrial-grade distributed training of vision-language models (VLMs) remains far less efficient than that of unimodal LLMs. Existing solutions either follow a monolithic design t…
BLADE: Scalable Bi-level Adaptive Data Selection for LLM Training
Jiaxing Wang, Deping Xiang, Jin Xu +9
As Large Language Model (LLM) datasets scale to trillions of tokens, data selection has emerged as a critical frontier to filter out uninformative noise and construct adaptive lear…
TANDEM: Bi-Level Data Mixture Optimization with Twin Networks
Jiaxing Wang, Deping Xiang, Jin Xu +9
The capabilities of large language models (LLMs) significantly depend on training data drawn from various domains. Optimizing domain-specific mixture ratios can be modeled as a bi-…
NestPipe: Large-Scale Recommendation Training on 1,500+ Accelerators via Nested Pipelining
Zhida Jiang, Zhaolong Xing, Huichao Chai +12
Modern recommendation models have increased to trillions of parameters. As cluster scales expand to O(1k), distributed training bottlenecks shift from computation and memory to dat…
The Primacy of Magnitude in Low-Rank Adaptation
Zicheng Zhang, Haoran Li, Yifeng Zhang +5
Low-Rank Adaptation (LoRA) offers a parameter-efficient paradigm for tuning large models. While recent spectral initialization methods improve convergence and performance over the…