6 papers · 1 filter
When Does Generating More Help? Disentangling Fixed-Source Synthesis from Source Expansion in Synthetic Data Scaling
Xu Guo, Jian Tong, Zhihui Lu +1
Synthetic data can be scaled along two routes: Source Expansion (SE), which enlarges the source by adding seed materials or generators, and Fixed-Source Synthesis (FSS), which hold…
Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data
Xu Guo, Runyu Peng, Jian Tong +4
Large language models (LLMs) rely on web-scale corpora for pre-training. The noise inherent in these datasets tends to obscure meaningful patterns and ultimately degrade model perf…
Rethinking Multiple-Choice Questions for RLVR: Unlocking Potential via Distractor Design
Xu Guo, Qiming Ge, Jian Tong +8
Reinforcement Learning with Verifiable Rewards (RLVR) significantly enhances the reasoning capabilities of Large Language Models. When applied to RLVR, Multiple-Choice Questions (M…
Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding
Shijing Hu, Jingyang Li, Zhihui Lu +1
Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet e…
GRIFFIN: Effective Token Alignment for Faster Speculative Decoding
Shijing Hu, Jingyang Li, Xingyu Xie +3
Speculative decoding accelerates inference in large language models (LLMs) by generating multiple draft tokens simultaneously. However, existing methods often struggle with token m…
IFDECORATOR: Wrapping Instruction Following Reinforcement Learning with Verifiable Rewards
Xu Guo, Tianyi Liang, Tong Jian +6
Reinforcement Learning with Verifiable Rewards (RLVR) improves instruction following capabilities of large language models (LLMs), but suffers from training inefficiency due to ina…