V
主页
Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirect
发布人
论文简述:在这篇名为Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model的论文中,作者提出了一种名为奖励增强解码(RAD)的方法,用于在文本生成过程中引导大型语言模型产生具有特定属性的输出。具体来说,RAD使用一个小型单向奖励模型来评估生成的文本,并根据奖励分数调整采样概率,以便优先选择具有高奖励值的标记。通过使用单向奖励模型,RAD可以缓存先前生成步骤的激活,从而降低计算开销。在实验中,作者展示了RAD在各种任务上优于仅改变生成过程的方法,并在很大程度上达到了重新训练语言模型的现有技术的性能。此外,论文还验证了RAD在大规模语言模型上的有效性,同时具有较小的计算开销。总之,这篇论文提出了一种有效且计算效率高的文本生成方法,可以在保持高性能的同时控制生成的文本属性。 论文链接: https://arxiv.org/pdf/2310.09520
打开封面
下载高清视频
观看高清视频
视频下载器
Visual In-Context Prompting
Aligning Text-to-Image Diffusion Models with Reward Backpropagation
Progressive3D: Progressively Local Editing for Text-to-3D Content Creation with
LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes
Parrot: Pareto-optimal Multi-Reward Reinforcement Learning Framework for Text-to
CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion
The Chosen One: Consistent Characters in Text-to-Image Diffusion Models
RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augment
The Consensus Game: Language Model Generation via Equilibrium Search
AutoStory: Generating Diverse Storytelling Images with Minimal Human Effort
ChatCoder: Chat-based Refine Requirement Improves LLMs' Code Generation
ReFT: Reasoning with Reinforced Fine-Tuning
Kosmos-G: Generating Images in Context with Multimodal Large Language Models
In-Context Learning Creates Task Vectors
Controlled Decoding from Language Models
Retrieval meets Long Context Large Language Models
PALP: Prompt Aligned Personalization of Text-to-Image Models
Branch-Solve-Merge Improves Large Language Model Evaluation and Generation
NLP学起来太难了吧!迪哥带你高效入门NLP自然语言处理,从原理到分类实战,3小时完全吃透!
GraphLLM: Boosting Graph Reasoning Ability of Large Language Model
Approximating Two-Layer Feedforward Networks for Efficient Transformers
Compressing Context to Enhance Inference Efficiency of Large Language Models
Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model
RLVF: Learning from Verbal Feedback without Overgeneralization
Enhancing High-Resolution 3D Generation through Pixel-wise Gradient Clipping
TofuEval: Evaluating Hallucinations of LLMs on Topic-Focused Dialogue Summarizat
Language Models can be Logical Solvers
Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning
Secrets of RLHF in Large Language Models Part II: Reward Modeling
KITAB: Evaluating LLMs on Constraint Satisfaction for Information Retrieval
Amortizing intractable inference in large language models
Memory Consolidation Enables Long-Context Video Understanding
LAMP: Learn A Motion Pattern for Few-Shot-Based Video Generation
GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection
Rethinking FID: Towards a Better Evaluation Metric for Image Generation
UNcommonsense Reasoning: Abductive Reasoning about Uncommon Situations
Asynchronous Local-SGD Training for Language Modeling
Efficient Tool Use with Chain-of-Abstraction Reasoning
Auto-Instruct: Automatic Instruction Generation and Ranking for Black-Box Langua
FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation