arXiv subjects

Dong Ahn

Publications and source records attributed to Dong Ahn.

8 recordsLinked to original sources

arXiv2025

NVIDIA Nemotron 3: Efficient and Open Intelligence

We introduce the Nemotron 3 family of models - Nano, Super, and Ultra. These models deliver strong agentic, reasoning, and conversational capabilities. The Nemotron 3 family uses a Mixture-of-Experts hybrid Mamba-Transformer architecture to provide best-in-class throughput and context lengths of up to 1M tokens. Super and Ultra models are trained with NVFP4 and incorporate LatentMoE, a novel approach that improves model quality. The two larger models also include MTP layers for faster text generation. All Nemotron 3 models are post-trained using multi-environment reinforcement learning enabling reasoning, multi-step tool use, and support granular reasoning budget control. Nano, the smallest model, outperforms comparable models in accuracy while remaining extremely cost-efficient for inference. Super is optimized for collaborative agents and high-volume workloads such as IT ticket automation. Ultra, the largest model, provides state-of-the-art accuracy and reasoning performance. Nano is released together with its technical report and this white paper, while Super and Ultra will follow in the coming months. We will openly release the model weights, pre- and post-training software, recipes, and all data for which we hold redistribution rights.

cs.CL↗

arXiv2025

Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

We present Nemotron 3 Nano 30B-A3B, a Mixture-of-Experts hybrid Mamba-Transformer language model. Nemotron 3 Nano was pretrained on 25 trillion text tokens, including more than 3 trillion new unique tokens over Nemotron 2, followed by supervised fine tuning and large-scale RL on diverse environments. Nemotron 3 Nano achieves better accuracy than our previous generation Nemotron 2 Nano while activating less than half of the parameters per forward pass. It achieves up to 3.3x higher inference throughput than similarly-sized open models like GPT-OSS-20B and Qwen3-30B-A3B-Thinking-2507, while also being more accurate on popular benchmarks. Nemotron 3 Nano demonstrates enhanced agentic, reasoning, and chat abilities and supports context lengths up to 1M tokens. We release both our pretrained Nemotron 3 Nano 30B-A3B Base and post-trained Nemotron 3 Nano 30B-A3B checkpoints on Hugging Face.

cs.CL↗

arXiv2025

Pretraining Large Language Models with NVFP4

Large Language Models (LLMs) today are powerful problem solvers across many domains, and they continue to get stronger as they scale in model size, training set size, and training set quality, as shown by extensive research and experimentation across the industry. Training a frontier model today requires on the order of tens to hundreds of yottaflops, which is a massive investment of time, compute, and energy. Improving pretraining efficiency is therefore essential to enable the next generation of even more capable LLMs. While 8-bit floating point (FP8) training is now widely adopted, transitioning to even narrower precision, such as 4-bit floating point (FP4), could unlock additional improvements in computational speed and resource utilization. However, quantization at this level poses challenges to training stability, convergence, and implementation, notably for large-scale models trained on long token horizons. In this study, we introduce a novel approach for stable and accurate training of large language models (LLMs) using the NVFP4 format. Our method integrates Random Hadamard transforms (RHT) to bound block-level outliers, employs a two-dimensional quantization scheme for consistent representations across both the forward and backward passes, utilizes stochastic rounding for unbiased gradient estimation, and incorporates selective high-precision layers. We validate our approach by training a 12-billion-parameter model on 10 trillion tokens -- the longest publicly documented training run in 4-bit precision to date. Our results show that the model trained with our NVFP4-based pretraining technique achieves training loss and downstream task accuracies comparable to an FP8 baseline. These findings highlight that NVFP4, when combined with our training approach, represents a major step forward in narrow-precision LLM training algorithms.

cs.CL↗

arXiv2025

NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model

We introduce Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. Nemotron-Nano-9B-v2 builds on the Nemotron-H architecture, in which the majority of the self-attention layers in the common Transformer architecture are replaced with Mamba-2 layers, to achieve improved inference speed when generating the long thinking traces needed for reasoning. We create Nemotron-Nano-9B-v2 by first pre-training a 12-billion-parameter model (Nemotron-Nano-12B-v2-Base) on 20 trillion tokens using an FP8 training recipe. After aligning Nemotron-Nano-12B-v2-Base, we employ the Minitron strategy to compress and distill the model with the goal of enabling inference on up to 128k tokens on a single NVIDIA A10G GPU (22GiB of memory, bfloat16 precision). Compared to existing similarly-sized models (e.g., Qwen3-8B), we show that Nemotron-Nano-9B-v2 achieves on-par or better accuracy on reasoning benchmarks while achieving up to 6x higher inference throughput in reasoning settings like 8k input and 16k output tokens. We are releasing Nemotron-Nano-9B-v2, Nemotron-Nano12B-v2-Base, and Nemotron-Nano-9B-v2-Base checkpoints along with the majority of our pre- and post-training datasets on Hugging Face.

cs.CL↗

arXiv2025

Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models

As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.

cs.CL↗

arXiv2022

Workflows Community Summit: Tightening the Integration between Computing Facilities and Scientific Workflows

The importance of workflows is highlighted by the fact that they have underpinned some of the most significant discoveries of the past decades. Many of these workflows have significant computational, storage, and communication demands, and thus must execute on a range of large-scale computer systems, from local clusters to public clouds and upcoming exascale HPC platforms. Historically, infrastructures for workflow execution consisted of complex, integrated systems, developed in-house by workflow practitioners with strong dependencies on a range of legacy technologies. Due to the increasing need to support workflows, dedicated workflow systems were developed to provide abstractions for creating, executing, and adapting workflows conveniently and efficiently while ensuring portability. While these efforts are all worthwhile individually, there are now hundreds of independent workflow systems. The resulting workflow system technology landscape is fragmented, which may present significant barriers for future workflow users due to many seemingly comparable, yet usually mutually incompatible, systems that exist. In order to tackle some of these challenges, the DOE-funded ExaWorks and NSF-funded WorkflowsRI projects have organized in 2021 a series of events entitled the "Workflows Community Summit". The third edition of the ``Workflows Community Summit" explored workflows challenges and opportunities from the perspective of computing centers and facilities. The third summit brought together a small group of facilities representatives with the aim to understand how workflows are currently being used at each facility, how facilities would like to interact with workflow developers and users, how workflows fit with facility roadmaps, and what opportunities there are for tighter integration between facilities and workflows. More information at: https://workflowsri.org/summits/facilities/

cs.DC↗

arXiv2021

Workflows Community Summit: Advancing the State-of-the-art of Scientific Workflows Management Systems Research and Development

Scientific workflows are a cornerstone of modern scientific computing, and they have underpinned some of the most significant discoveries of the last decade. Many of these workflows have high computational, storage, and/or communication demands, and thus must execute on a wide range of large-scale platforms, from large clouds to upcoming exascale HPC platforms. Workflows will play a crucial role in the data-oriented and post-Moore's computing landscape as they democratize the application of cutting-edge research techniques, computationally intensive methods, and use of new computing platforms. As workflows continue to be adopted by scientific projects and user communities, they are becoming more complex. Workflows are increasingly composed of tasks that perform computations such as short machine learning inference, multi-node simulations, long-running machine learning model training, amongst others, and thus increasingly rely on heterogeneous architectures that include CPUs but also GPUs and accelerators. The workflow management system (WMS) technology landscape is currently segmented and presents significant barriers to entry due to the hundreds of seemingly comparable, yet incompatible, systems that exist. Another fundamental problem is that there are conflicting theoretical bases and abstractions for a WMS. Systems that use the same underlying abstractions can likely be translated between, which is not the case for systems that use different abstractions. More information: https://workflowsri.org/summits/technical

cs.DC↗

arXiv2021

Workflows Community Summit: Bringing the Scientific Workflows Community Together

Scientific workflows have been used almost universally across scientific domains, and have underpinned some of the most significant discoveries of the past several decades. Many of these workflows have high computational, storage, and/or communication demands, and thus must execute on a wide range of large-scale platforms, from large clouds to upcoming exascale high-performance computing (HPC) platforms. These executions must be managed using some software infrastructure. Due to the popularity of workflows, workflow management systems (WMSs) have been developed to provide abstractions for creating and executing workflows conveniently, efficiently, and portably. While these efforts are all worthwhile, there are now hundreds of independent WMSs, many of which are moribund. As a result, the WMS landscape is segmented and presents significant barriers to entry due to the hundreds of seemingly comparable, yet incompatible, systems that exist. As a result, many teams, small and large, still elect to build their own custom workflow solution rather than adopt, or build upon, existing WMSs. This current state of the WMS landscape negatively impacts workflow users, developers, and researchers. The "Workflows Community Summit" was held online on January 13, 2021. The overarching goal of the summit was to develop a view of the state of the art and identify crucial research challenges in the workflow community. Prior to the summit, a survey sent to stakeholders in the workflow community (including both developers of WMSs and users of workflows) helped to identify key challenges in this community that were translated into 6 broad themes for the summit, each of them being the object of a focused discussion led by a volunteer member of the community. This report documents and organizes the wealth of information provided by the participants before, during, and after the summit.

cs.DC↗