Trends · 研究趋势
数据来自 Hugging Face 论文的 AI 提取关键词,按月统计研究方向的增长与热度。
Kaijie Zhu, Yuzhou Nie, Yijiang Li +10 authors
TermiGen introduces a pipeline for generating verifiable terminal environments and resilient trajectories to improve open-weight LLMs' ability to execute complex tasks and recover from runtime errors.
Jian Yang, Shawn Guo, Wei Zhang +16 authors
Parallel loop Transformers achieve better code generation performance with two loops due to refined representations, while additional loops cause diminishing returns and increased positional mismatch costs.
Kang Liao, Size Wu, Zhonghua Wu +6 authors
Puffin, a unified multimodal model, integrates language regression and diffusion-based generation to enhance camera-centric spatial understanding and generation by treating camera parameters as language.
Yujia Qin, Shihao Liang, Yining Ye +15 authors
ToolLLM is a framework that enhances open-source LLMs with tool-use capabilities through a comprehensive dataset, efficient decision tree, automatic evaluation, and neural API retrieval.
Violet Xiang, Charlie Snell, Kanishk Gandhi +11 authors
The Meta Chain-of-Thought framework extends traditional CoT by explicitly modeling reasoning, enabling more powerful and human-like reasoning in LLMs through process supervision, synthetic data generation, and reinforcement learning.
Zheyuan Yang, Lyuhao Chen, Arman Cohan +1 authors
Two post-training strategies, distillation and RLVR, enable inference-time scaling in table reasoning tasks, resulting in a model (Table-R1-Zero) that matches GPT-4.1's performance using fewer parameters and shows strong generalization.
Byung-Kwan Lee, Beomchan Park, Chae Won Kim +1 authors
MoAI, a new LLVM, integrates outputs from external CV models to enhance VL tasks without expanding model size or curating extra datasets.
Shengqiong Wu, Weicai Ye, Jiahao Wang +8 authors
Any2Caption framework improves video generation quality and controllability by decoupling condition interpretation and leveraging multimodal large language models.
Wenxuan Zhang, Hou Pong Chan, Yiran Zhao +9 authors
SeaLLMs 3 is a large language model optimized for Southeast Asian languages, offering state-of-the-art performance in various tasks while addressing safety and cultural considerations.
Jun Zhao, Zhihao Zhang, Qi Zhang +2 authors
A comprehensive study on transferring English-dominant LLMs to non-English languages using LLaMA demonstrates that comparable performance can be achieved with minimal pretraining data across various language capabilities and response quality metrics.
Arindam Mitra, Luciano Del Corro, Guoqing Zheng +11 authors
AgentInstruct, an agentic framework, generates high-quality synthetic data for post-training language models, leading to significant performance improvements across various benchmarks.
Zhaojian Yu, Xin Zhang, Ning Shang +5 authors
A Generator-Discriminator framework using LLMs generates high-quality instruction data, enhancing the fine-tuning of code-related tasks and model generalization.
Ting Jiang, Shaohan Huang, Shengyue Luo +8 authors
MoRA, a high-rank updating method using square matrices, enhances the ability of large language models to learn and memorize new knowledge, especially in memory-intensive tasks, compared to LoRA.
Yinghao Ma, Anders Øland, Anton Ragni +40 authors
A review of foundation models in music, including large language models and latent diffusion models, highlights their impact, architectural choices, and the need for ethical considerations in music applications.
Zhangchen Xu, Fengqing Jiang, Luyao Niu +2 authors
The Larger Models' Paradox reveals that larger models are not always better teachers for fine-tuning smaller models, and a new metric, Compatibility-Adjusted Reward (CAR), is introduced to measure and improve the effectiveness of response generators.
Houwen Peng, Kan Wu, Yixuan Wei +17 authors
A new FP8 automatic mixed-precision framework for training large language models reduces memory usage and increases speed compared to BF16 and Nvidia Transformer Engine.
Niklas Muennighoff, Qian Liu, Armel Zebaze +7 authors
Instruction tuning using Git commits improves performance on natural language and coding tasks compared to other benchmarks, with models achieving state-of-the-art results on expanded HumanEvalPack.
Wenhan Xiong, Jingyu Liu, Igor Molybog +18 authors
Long-context LLMs achieve significant advancements in handling extended contexts through continual pretraining and efficient instruction tuning, surpassing existing models on various benchmarks.
Chunting Zhou, Pengfei Liu, Puxin Xu +12 authors
A 65B parameter LLaMa language model trained with minimal instructional data matches or outperforms models with extensive human preference modeling in most cases, indicating that pretraining is predominantly responsible for knowledge acquisition.
Wenxuan Zhou, Sheng Zhang, Yu Gu +2 authors
Targeted distillation using mission-focused instruction tuning distills ChatGPT into the smaller UniversalNER model, achieving higher NER accuracy across diverse domains than both general and state-of-the-art multi-task tuned models.
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