2025년 7월 9일
Skywork-R1V3 Technical Report
(Wei Shen, Jiangbo Pei, Yi Peng, Xuchen Song, Yang Liu, Jian Peng, Haofeng Sun, Yunzhuo Hao, Peiyu Wang, Yahui Zhou)
We introduce Skywork-R1V3, an advanced, open-source vision-language model (VLM) that pioneers a new approach to visual reasoning. Its key innovation lies in effectively transferring reasoning skills from text-only Large Language Models (LLMs) to visual tasks. The strong performance of Skywork-R1V3 primarily stems from our elaborate post-training RL framework, which effectively activates and enhances the model's reasoning ability, without the need for additional continue pre-training. Through this framework, we further uncover the fundamental role of the connector module in achieving robust cross-modal alignment for multimodal reasoning models. In addition, we introduce a unique indicator of reasoning capability, the entropy of critical reasoning tokens, which has proven highly effective for checkpoint selection during RL training. Skywork-R1V3 achieves state-of-the-art results on MMMU, significantly improving from 64.3% to 76.0%. This performance matches entry-level human capabilities. Remarkably, our RL-powered post-training approach enables even the 38B parameter model to rival top closed-source VLMs. The implementation successfully transfers mathematical reasoning to other subject-related reasoning tasks. We also include an analysis of curriculum learning and reinforcement finetuning strategies, along with a broader discussion on multimodal reasoning. Skywork-R1V3 represents a significant leap in multimodal reasoning, showcasing RL as a powerful engine for advancing open-source VLM capabilities.
Skywork가 멀티모달 모델을 또 냈군요. 여기에서도 결정적인 토큰의 엔트로피가 중요하다는 분석을 합니다 (https://arxiv.org/abs/2506.01939).
Skywork has released another multimodal model. This paper also emphasizes the importance of the entropy of critical tokens in their analysis (https://arxiv.org/abs/2506.01939).
#multimodal #rl #reasoning
BlueLM-2.5-3B Technical Report
(Baojiao Xiong, Boheng Chen, Chengzhi Wang, Daxiong Luo, Dongsheng Xu, Dongyang Liu, Fan Yang, Fangyuan Li, Fei Teng, Feng Wang, Fukang Qin, Fuquan Peng, Guanxin Tan, Guozhi Wang, Haibo Yu, Haohao Gao, Heng Liu, Hongbo Yang, Hongjian Zou, Houzheng Shen, Hu Meng, Huan Li, Hui Tan, Jiali Chen, Jianzhao Chen, Jinliang Zhu, Kai Wang, Lei Wu, Liangbing Liu, Liuyang Bian, Liyan He, Long Liu, Peiwen Li, Penggang Shi, Qi Ding, Rui Hu, Shuai Cao, Shuai Ren, Shuang Peng, Teng Xie, Weiji Chen, Weilin Xiang, Weixin Wu, Xi Yin, Xiaoxin Chen, Xu Chen, Yafei Wen, Yan Hu, Yanzhou Yang, Yina Xie, Yinghao Chen, Yixuan Liao, Yu Geng, Yuanjiang Ouyang, Yuanzhuo Yang, Yuehua He, Yushuai Peng, Zhaoxiong Wang, Zheng Wang, Zhibo Zhou, Ziyang Wu)
We present BlueLM-2.5-3B, a compact and unified dense Multimodal Large Language Model (MLLM) designed for efficient edge-device deployment, offering strong general-purpose and reasoning capabilities. To the best of our knowledge, this is the first 3B-scale MLLM to support both thinking and non-thinking modes, while also enabling explicit control over thinking token budget. BlueLM-2.5-3B is developed through diversified data curation, key data resampling, hybrid heterogeneous reinforcement learning, and a high-performance training infrastructure. Our model achieves superior multimodal capacity while preserving competitive pure-text performance with only 2.9 billion parameters. We conduct comprehensive evaluations across a broad range of multimodal and text-only benchmarks. In thinking mode, BlueLM-2.5-3B achieves comparable performance to Qwen3-4B on text-only benchmarks, and trails the larger Kimi-VL-A3B-16B by only about 5% on average across multimodal evaluations. In non-thinking mode, it outperforms Qwen2.5-VL-3B on the majority of multimodal benchmarks. Additionally, BlueLM-2.5-3B exhibits exceptional data efficiency. All of the aforementioned performance is achieved with substantially less total training data than Qwen2.5-VL-3B and Qwen3-4B. We hope our work contributes to the advancement of high-performance, on-device MLLMs and provides meaningful insights to the research community.
비보의 멀티모달 추론 모델.
Vivo's multimodal reasoning model.
#multimodal #rl #reasoning #pretraining
RefineX: Learning to Refine Pre-training Data at Scale from Expert-Guided Programs
(Baolong Bi, Shenghua Liu, Xingzhang Ren, Dayiheng Liu, Junyang Lin, Yiwei Wang, Lingrui Mei, Junfeng Fang, Jiafeng Guo, Xueqi Cheng)
The foundational capabilities of large language models (LLMs) are deeply influenced by the quality of their pre-training corpora. However, enhancing data quality at scale remains a significant challenge, primarily due to the trade-off between refinement effectiveness and processing efficiency. While rule-based filtering remains the dominant paradigm, it typically operates at the document level and lacks the granularity needed to refine specific content within documents. Inspired by emerging work such as ProX, we propose RefineX, a novel framework for large-scale, surgical refinement of pre-training data through programmatic editing tasks. RefineX enables efficient and fine-grained data refinement while reliably preserving the diversity and naturalness of raw text. The core strength of RefineX lies in distilling high-quality, expert-guided end-to-end refinement results into minimal edit-based deletion programs. This high-precision distillation pipeline is used to train an efficient and reliable refine model that can systematically improve every instance in the corpus at scale. We evaluate RefineX across from-scratch pre-training at multiple model scales and find that it consistently outperforms models trained on raw, filtered, or alternatively refined data across diverse downstream tasks. On the 750M model, RefineX yields 2.6%-7.2% average gains on lighteval tasks, and achieves comparable performance using significantly fewer training tokens. Further analysis shows that RefineX reliably enhances text quality with both high efficiency and precision, outperforming prior approaches such as end-to-end generation and Prox-C. These results position RefineX as a scalable, effective, and reliable solution for optimizing pre-training data in modern LLM pipelines.
프로그램 생성 기반 프리트레이닝 데이터 재작성에 대한 개선 (https://arxiv.org/abs/2409.17115). LLM으로 프로그램을 직접 생성시키는 것이 아니라 재작성 후 프로그램을 추론하는 형태로 학습 데이터를 구성. 대신 프로그램을 삭제로 제약하긴 했네요.
An improvement on the method of rewriting pretraining data based on program generation (https://arxiv.org/abs/2409.17115). Instead of having an LLM directly generate programs to create training data, they have the model rewrite the text and then infer a program from it. However, they restricted the programs to deletion operations only.
#synthetic-data #pretraining #corpus
CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization
(Zhongyuan Peng, Yifan Yao, Kaijing Ma, Shuyue Guo, Yizhe Li, Yichi Zhang, Chenchen Zhang, Yifan Zhang, Zhouliang Yu, Luming Li, Minghao Liu, Yihang Xia, Jiawei Shen, Yuchen Wu, Yixin Cao, Zhaoxiang Zhang, Wenhao Huang, Jiaheng Liu, Ge Zhang)
Translating natural language mathematical statements into formal, executable code is a fundamental challenge in automated theorem proving. While prior work has focused on generation and compilation success, little attention has been paid to the critic phase-the evaluation of whether generated formalizations truly capture the semantic intent of the original problem. In this paper, we introduce CriticLean, a novel critic-guided reinforcement learning framework that elevates the role of the critic from a passive validator to an active learning component. Specifically, first, we propose the CriticLeanGPT, trained via supervised fine-tuning and reinforcement learning, to rigorously assess the semantic fidelity of Lean 4 formalizations. Then, we introduce CriticLeanBench, a benchmark designed to measure models' ability to distinguish semantically correct from incorrect formalizations, and demonstrate that our trained CriticLeanGPT models can significantly outperform strong open- and closed-source baselines. Building on the CriticLean framework, we construct FineLeanCorpus, a dataset comprising over 285K problems that exhibits rich domain diversity, broad difficulty coverage, and high correctness based on human evaluation. Overall, our findings highlight that optimizing the critic phase is essential for producing reliable formalizations, and we hope our CriticLean will provide valuable insights for future advances in formal mathematical reasoning.
Autoformalization의 결과를 검증하는 모델과 벤치마크.
A model and benchmark for verifying the results of autoformalization.
#math #benchmark #rl
Decoupled Relative Learning Rate Schedules
(Jan Ludziejewski, Jan Małaśnicki, Maciej Pióro, Michał Krutul, Kamil Ciebiera, Maciej Stefaniak, Jakub Krajewski, Piotr Sankowski, Marek Cygan, Kamil Adamczewski, Sebastian Jaszczur)
In this work, we introduce a novel approach for optimizing LLM training by adjusting learning rates across weights of different components in Transformer models. Traditional methods often apply a uniform learning rate across all network layers, potentially overlooking the unique dynamics of each part. Remarkably, our introduced relative learning rates, RLRS, method accelerates the training process by up to 23%, particularly in complex models such as Mixture of Experts (MoE). Hyperparameters of RLRS can be efficiently tuned on smaller models and then effectively reused on models up to 27× larger. This simple and effective method results in a substantial reduction in training time and computational resources, offering a practical and scalable solution for optimizing large-scale neural networks.
모듈마다 다른 LR 스케줄을 부여하자는 아이디어. 작은 모델에서 스케줄을 찾은 다음 큰 모델에 적용하는 간단한 방식입니다.
The idea of assigning different learning rate schedules to each module. It's a simple method where schedules are determined using small models and then applied to larger models.
#optimization #moe