4 papers
Predicting Rewards Alongside Tokens: Non-disruptive Parameter Insertion for Efficient Inference Intervention in Large Language Model
Chenhan Yuan, Fei Huang, Ru Peng +4
Transformer-based large language models (LLMs) exhibit limitations such as generating unsafe responses, unreliable reasoning, etc. Existing inference intervention approaches attemp…
MuggleMath: Assessing the Impact of Query and Response Augmentation on Math Reasoning
Chengpeng Li, Zheng Yuan, Hongyi Yuan +6
In math reasoning with large language models (LLMs), fine-tuning data augmentation by query evolution and diverse reasoning paths is empirically verified effective, profoundly narr…
How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition
Guanting Dong, Hongyi Yuan, Keming Lu +7
Large language models (LLMs) with enormous pre-training tokens and parameters emerge diverse abilities, including math reasoning, code generation, and instruction following. These…
Online Merging Optimizers for Boosting Rewards and Mitigating Tax in Alignment
Keming Lu, Bowen Yu, Fei Huang +3
Effectively aligning Large Language Models (LLMs) with human-centric values while preventing the degradation of abilities acquired through Pre-training and Supervised Fine-tuning (…