이진욱님의 블로그

https://arxiv.org/pdf/2407.06057

https://arxiv.org/abs/2401.11206 InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model GuidanceWith the rapid development of large language models (LLMs), they are not only used as general-purpose AI assistants but are also customized through further fine-tuning to meet the requirements of different applications. A pivotal factor in the success of carxiv.orgAbstract With th..

https://arxiv.org/pdf/2410.20290The safe and effective deployment of Large Language Models (LLMs) involves a critical step called alignment, which ensures that the model’s responses are in accordance with human preferences. Prevalent alignment techniques, such as DPO, PPO and their variants, align LLMs by changing the pre-trained model weights during a phase called post-training. While predomina..

https://arxiv.org/pdf/2406.12845https://github.com/RLHFlow/RLHF-Reward-Modeling Reinforcement learning from human feedback (RLHF) has emerged as the primary method for aligning large language models (LLMs) with human preferences. The RLHF process typically starts by training a reward model (RM) using human preference data. Conventional RMs are trained on pairwise responses to the same user reque..

https://arxiv.org/pdf/2407.08739 Multi-modal Large Language Models (MLLMs) have recently emerged as a significant focus in academia and industry. Despite their proficiency in general multi-modal scenarios, the mathematical problem-solving capabilities in visual contexts remain insufficiently explored. We identify three key areas within MLLMs that need to be improved: visual encoding of math diag..

https://arxiv.org/pdf/2410.16198https://github.com/RifleZhang/LLaVA-Reasoner-DPO GitHub - RifleZhang/LLaVA-Reasoner-DPOContribute to RifleZhang/LLaVA-Reasoner-DPO development by creating an account on GitHub.github.comChain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes lack robust CoT r..

https://arxiv.org/abs/2406.12624

https://arxiv.org/abs/2409.15268

2.3 RLAIF AND LLM-AS-A-JUDGE Reinforcement Learning from AI Feedback (RLAIF) presents an alternative approach to the standard RLHF pipeline. Bai et al. (2022b) demonstrate the efficacy of RLAIF in training helpful and harmless models without relying on human feedback labels for harmlessness assessment. Their work shows that as language model capabilities improve, AI identification of harms incre..

https://arxiv.org/abs/2410.17050 UnStar: Unlearning with Self-Taught Anti-Sample Reasoning for LLMsThe key components of machine learning are data samples for training, model for learning patterns, and loss function for optimizing accuracy. Analogously, unlearning can potentially be achieved through anti-data samples (or anti-samples), unlearning methodarxiv.orgThe key components of machine lear..