5 papers
Beyond Few-Step Inference: Accelerating Video Diffusion Transformer Model Serving with Inter-Request Caching Reuse
Hao Liu, Ye Huang, Chenghuan Huang +5
Video Diffusion Transformer (DiT) models are a dominant approach for high-quality video generation but suffer from high inference cost due to iterative denoising. Existing caching…
Distill Not Only Data but Also Rewards: Can Smaller Language Models Surpass Larger Ones?
Yudi Zhang, Lu Wang, Meng Fang +8
Distilling large language models (LLMs) typically involves transferring the teacher model's responses through supervised fine-tuning (SFT). However, this approach neglects the pote…
Pretrain Value, Not Reward: Decoupled Value Policy Optimization
Chenghua Huang, Lu Wang, Fangkai Yang +6
In this paper, we explore how directly pretraining a value model simplifies and stabilizes reinforcement learning from human feedback (RLHF). In reinforcement learning, value estim…
Token-level Proximal Policy Optimization for Query Generation
Yichen Ouyang, Lu Wang, Fangkai Yang +13
Query generation is a critical task for web search engines (e.g. Google, Bing) and recommendation systems. Recently, state-of-the-art query generation methods leverage Large Langua…
Self-Evolved Reward Learning for LLMs
Chenghua Huang, Zhizhen Fan, Lu Wang +7
Reinforcement Learning from Human Feedback (RLHF) is a crucial technique for aligning language models with human preferences, playing a pivotal role in the success of conversationa…