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。
Yubo Wang, Xiang Yue, Wenhu Chen
Critique Fine-Tuning (CFT) improves language model performance by training them to critique noisy responses, showing better reasoning ability than standard Supervised Fine-Tuning (SFT).
Edward Yeo, Yuxuan Tong, Morry Niu +2 authors
Investigation into long chains-of-thought reasoning in large language models reveals the critical role of training compute, reward shaping, and verifiable reward signals in enabling and measuring this capability.
Guosheng Dong, Da Pan, Yiding Sun +17 authors
A data processing pipeline and open-sourced details for training a large language model achieve competitive performance with commercial models.
Tengchao Lv, Yupan Huang, Jingye Chen +11 authors
Kosmos-2.5, a unified multimodal model, generates spatially-aware and structured text from text-intensive images using a Transformer architecture and task-specific prompts.
DeepSeek-AI, Xiao Bi, Deli Chen +84 authors
DeepSeek LLM, an open-source language model project, develops a large dataset and employs SFT and DPO to achieve performance surpassing LLaMA-2 70B and GPT-3.5 in various benchmarks and open-ended evaluations.
Kejian Zhu, Zhuoran Jin, Shangqing Tu +5 authors
Reinforcement learning enables stable multi-task reasoning in large language models through sparse, near-orthogonal parameter updates, unlike supervised fine-tuning which suffers from task conflicts, motivating a decoupled Parallel-RL training approach.
Xuekai Zhu, Daixuan Cheng, Hengli Li +7 authors
The use of synthetic data in language model training leads to model collapse, which is mitigated by token-level editing of human-produced data to create semi-synthetic data.
NVIDIA, Alisson Azzolini, Hannah Brandon +42 authors
Cosmos-Reason1 models, using hierarchical and two-dimensional ontologies for physical common sense and embodied reasoning, generate embodied decisions through multimodal large language models trained in vision and Physical AI stages.
RadixArk, Tom Chen, Mao Cheng +10 authors
Miles is an open-source, production-ready system for large-scale reinforcement learning and post-training that supports diverse backends, weight synchronization, LoRA, distillation, and diffusion models.
Mingan Lin, Fan Yang, Yanjun Shen +22 authors
Baichuan Alignment provides comprehensive insights into alignment methodologies used in Baichuan models, detailing improvements through Prompt Augmentation System, Supervised Fine-Tuning, and Preference Alignment across various benchmarks.
Zhen Fang, Yu Zeng, Wenxuan Huang +17 authors
Video-DeepResearch extends multimodal agents to continuous video via decoupled perception-exploration and GRPO training, achieving state-of-the-art results on a new video reasoning benchmark.
Alex Havrilla, Yuqing Du, Sharath Chandra Raparthy +6 authors
Multiple RLHF algorithms, including Expert Iteration, PPO, and Return-Conditioned RL, similarly improve LLM reasoning with similar sample complexity, with Expert Iteration slightly outperforming others.
Junyu Luo, Xiao Luo, Xiusi Chen +3 authors
A semi-supervised fine-tuning framework named SemiEvol enhances LLM adaptation using both labeled and unlabeled data, showing improved performance through bi-level knowledge propagation and collaborative learning.
Sihang Li, Jin Huang, Jiaxi Zhuang +7 authors
A hybrid approach combining continual pre-training and supervised fine-tuning creates specialized LLMs for scientific literature, addressing challenges in constructing corpora and generating instructions.
Zilong Chen, Chaorui Deng, Kunchang Li +2 authors
Structured language in prompts drives diffusion loss scaling, enabling improved visual generation through annotated prompts and a trained prompter.
Haoran Xu, Amr Sharaf, Yunmo Chen +5 authors
Contrastive Preference Optimization (CPO) enhances moderate-sized LLMs in machine translation, enabling them to match or exceed state-of-the-art model performance.
DiffusionGemma Team, Adrien Ali Taïga, James Assiene +41 authors
DiffusionGemma is a fine-tuned mixture-of-experts language model that uses discrete diffusion to generate text blocks in parallel, achieving high speed while preserving capabilities like multimodal inputs and reasoning.
Huizhuo Yuan, Zixiang Chen, Kaixuan Ji +1 authors
Self-play fine-tuning improves diffusion models' performance and alignment with human preferences using competition between model versions, outperforming supervised and RL-based methods with less data.
Zheng Cai, Maosong Cao, Haojiong Chen +97 authors
InternLM2 is an open-source LLM that outperforms predecessors through innovative pre-training and optimization techniques, including Supervised Fine-Tuning and Conditional Online Reinforcement Learning from Human Feedback.
Xiaoliang Dai, Ji Hou, Chih-Yao Ma +23 authors
Quality-tuning enhances the aesthetic appeal of text-to-image models through supervised fine-tuning with a small set of high-quality images.
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