Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
42 篇论文 · 按点赞排序
Yilun Du, Mengjiao Yang, Pete Florence +10 authors
Video language planning leverages vision-language and text-to-video models to generate detailed long-horizon video plans, improving task success rates across multiple robotics domains.
Haikang Deng, Colin Raffel
Reward-Augmented Decoding (RAD) enhances text generation by using a unidirectional reward model to favor desirable text attributes, achieving performance comparable to re-training methods with minimal computational cost.
Wenhua Cheng, Yiyang Cai, Kaokao Lv +1 authors
TEQ is a trainable transformation for low-precision quantization in large language models that maintains model accuracy with minimal computational overhead during inference.
Ruiqi Wu, Liangyu Chen, Tong Yang +3 authors
LAMP framework enables text-to-video generation using few-shot tuning of text-to-image diffusion models with limited resources, focusing on motion learning while leveraging existing content generation techniques.
Noveen Sachdeva, Zexue He, Wang-Cheng Kang +3 authors
Farzi distills large event sequence datasets into smaller synthetic datasets, significantly reducing data size while maintaining high model performance for auto-regressive tasks.
Ju-Chieh Chou, Chung-Ming Chien, Wei-Ning Hsu +5 authors
A joint language model for speech and text improves performance on spoken language understanding tasks and demonstrates zero-shot cross-modal transferability.
Bita Darvish Rouhani, Ritchie Zhao, Ankit More +29 authors
Microscaling data formats reduce computational and storage costs while maintaining model accuracy and enabling sub-8-bit training for generative language models.
Yixin Liu, Avi Singh, C. Daniel Freeman +2 authors
Three fine-tuning strategies significantly improve PaLM 2 models' performance in solving math problems on the MATH dataset.
Zijie Pan, Jiachen Lu, Xiatian Zhu +1 authors
Pixel-wise Gradient Clipping (PGC) improves high-resolution 3D object generation by controlling unregulated gradients in Latent Diffusion Models (LDM).
Yangruibo Ding, Zijian Wang, Wasi Uddin Ahmad +8 authors
CrossCodeEval is a code completion benchmark that tests models' ability to understand and use cross-file context from real-world, multilingual repositories, demonstrating the need for improved context retrieval and model performance.
Enric Boix-Adsera, Omid Saremi, Emmanuel Abbe +3 authors
Transformers require vast amounts of data for regression tasks and exhibit decreased generalization with higher embedding dimensions in symbolic next-token-prediction tasks, but simple parameter modifications can improve their performance.
Saikat Chakraborty, Shuvendu K. Lahiri, Sarah Fakhoury +6 authors
A re-ranking approach using a contrastive ranker improves the accuracy of synthesized loop invariants from LLMs, reducing the need for program verifier calls.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号