Qi Zhang

Zijie Yan, Hongxiao Bai, Xin Yao et al.

Scaling Mixture-of-Experts (MoE) training introduces systems challenges absent in dense models. Because each token activates only a subset of experts, this sparsity allows total parameters to grow much faster than per-token computation, creating coupled constraints across memory, communication, and computation. Optimizing one dimension often shifts pressure to another, demanding co-design across the full system stack. We address these challenges for MoE training through integrated optimizations spanning memory (fine-grained recomputation, offloading, etc.), communication (optimized dispatchers, overlapping, etc.), and computation (Grouped GEMM, fusions, CUDA Graphs, etc.). The framework also provides Parallel Folding for flexible multi-dimensional parallelism, low-precision training support for FP8 and NVFP4, and efficient long-context training. On NVIDIA GB300 and GB200, it achieves 1,233/1,048 TFLOPS/GPU for DeepSeek-V3-685B and 974/919 TFLOPS/GPU for Qwen3-235B. As a performant, scalable, and production-ready open-source solution, it has been used across academia and industry for training MoE models ranging from billions to trillions of parameters on clusters scaling up to thousands of GPUs. This report explains how these techniques work, their trade-offs, and their interactions at the systems level, providing practical guidance for scaling MoE models with Megatron Core.

Jiazheng Zhang, Wenqing Jing, Zizhuo Zhang et al.

Reward models (RMs) play a pivotal role in aligning large language models (LLMs) with human values. However, noisy preferences in human feedback can lead to reward misgeneralization - a phenomenon where reward models learn spurious correlations or overfit to noisy preferences, which poses important challenges to the generalization of RMs. This paper systematically analyzes the characteristics of preference pairs and aims to identify how noisy preferences differ from human-aligned preferences in reward modeling. Our analysis reveals that noisy preferences are difficult for RMs to fit, as they cause sharp training fluctuations and irregular gradient updates. These distinctive dynamics suggest the feasibility of identifying and excluding such noisy preferences. Empirical studies demonstrate that policy LLM optimized with a reward model trained on the full preference dataset, which includes substantial noise, performs worse than the one trained on a subset of exclusively high quality preferences. To address this challenge, we propose an online Collaborative Reward Modeling (CRM) framework to achieve robust preference learning through peer review and curriculum learning. In particular, CRM maintains two RMs that collaboratively filter potential noisy preferences by peer-reviewing each other's data selections. Curriculum learning synchronizes the capabilities of two models, mitigating excessive disparities to promote the utility of peer review. Extensive experiments demonstrate that CRM significantly enhances RM generalization, with up to 9.94 points improvement on RewardBench under an extreme 40\% noise. Moreover, CRM can seamlessly extend to implicit-reward alignment methods, offering a robust and versatile alignment strategy.

Guoqing Yang, Fuyou Xue, Qi Zhang et al.

We present the UrbanBIS benchmark for large-scale 3D urban understanding, supporting practical urban-level semantic and building-level instance segmentation. UrbanBIS comprises six real urban scenes, with 2.5 billion points, covering a vast area of 10.78 square kilometers and 3,370 buildings, captured by 113,346 views of aerial photogrammetry. Particularly, UrbanBIS provides not only semantic-level annotations on a rich set of urban objects, including buildings, vehicles, vegetation, roads, and bridges, but also instance-level annotations on the buildings. Further, UrbanBIS is the first 3D dataset that introduces fine-grained building sub-categories, considering a wide variety of shapes for different building types. Besides, we propose B-Seg, a building instance segmentation method to establish UrbanBIS. B-Seg adopts an end-to-end framework with a simple yet effective strategy for handling large-scale point clouds. Compared with mainstream methods, B-Seg achieves better accuracy with faster inference speed on UrbanBIS. In addition to the carefully-annotated point clouds, UrbanBIS provides high-resolution aerial-acquisition photos and high-quality large-scale 3D reconstruction models, which shall facilitate a wide range of studies such as multi-view stereo, urban LOD generation, aerial path planning, autonomous navigation, road network extraction, and so on, thus serving as an important platform for many intelligent city applications.