Zijie Yan

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.

Aditya Vavre, Ethan He, Dennis Liu et al. · nvidia

Scaling large language models (LLMs) significantly improves performance but comes with prohibitive computational costs. Mixture-of-Experts (MoE) models offer an efficient alternative, increasing capacity without a proportional rise in compute requirements. However, training MoE models from scratch poses challenges like overfitting and routing instability. We present an efficient training recipe leveraging pre-trained dense checkpoints, training an 8-Expert Top-2 MoE model from Llama 3-8B with less than $1\%$ of typical pre-training compute. Our approach enhances downstream performance on academic benchmarks, achieving a $\textbf{2%}$ improvement in 0-shot accuracy on MMLU, while reaching a Model FLOPs Utilization (MFU) of $\textbf{46.8%}$ during training using our framework. We also integrate online upcycling in NeMo for seamless use of pre-trained weights, enabling cost-effective development of high-capacity MoE models.

Zijie Yan

Modern large scale machine learning applications require stochastic optimization algorithms to be implemented on distributed computational architectures. A key bottleneck is the communication overhead for exchanging information, such as stochastic gradients, among different nodes. Recently, gradient sparsification techniques have been proposed to reduce communications cost and thus alleviate the network overhead. However, most of gradient sparsification techniques consider only synchronous parallelism and cannot be applied in asynchronous scenarios, such as asynchronous distributed training for federated learning at mobile devices. In this paper, we present a dual-way gradient sparsification approach (DGS) that is suitable for asynchronous distributed training. We let workers download model difference, instead of the global model, from the server, and the model difference information is also sparsified so that the information exchanged overhead is reduced by sparsifying the dual-way communication between the server and workers. To preserve accuracy under dual-way sparsification, we design a sparsification aware momentum (SAMomentum) to turn sparsification into adaptive batch size between each parameter. We conduct experiments at a cluster of 32 workers, and the results show that, with the same compression ratio but much lower communication cost, our approach can achieve better scalability and generalization ability.