When Should LLMs Be Less Specific? Selective Abstraction for Reliable Long-Form Text Generation

LLMs are widely used, yet they remain prone to factual errors that erode user trust and limit adoption in high-risk settings. One approach to mitigate this risk is to equip models with uncertainty estimation mechanisms that abstain when confidence is low. However, this binary "all-or-nothing" approach is excessively restrictive in long-form settings, often discarding valuable information. We introduce Selective Abstraction (SA), a framework that enables LLMs to trade specificity for reliability by selectively reducing the detail of uncertain content. We first formalize SA through the lenses of selective risk and coverage. We then propose Atom-wise Selective Abstraction, a claim-level instantiation that decomposes responses into atomic claims (short, self-contained statements each expressing a single fact) and replaces uncertain atoms with higher confidence, less specific abstractions. To evaluate this framework, we develop a novel end-to-end pipeline for open-ended generation that instantiates risk as factual correctness and measures coverage using an information-theoretic measure of retained information. Across six open-source models on the FactScore and LongFact-Objects benchmarks, atom-wise SA consistently outperforms existing baselines, improving the area under the risk-coverage curve (AURC) by up to 27.73% over claim removal, demonstrating that reducing specificity can boost accuracy and reliability while preserving most of their original meaning.

Extending Puzzle for Mixture-of-Experts Reasoning Models with Application to GPT-OSS Acceleration

Reasoning-focused LLMs improve answer quality by generating longer reasoning traces, but the additional tokens dramatically increase serving cost, motivating inference optimization. We extend and apply Puzzle, a post-training neural architecture search (NAS) framework, to gpt-oss-120B to produce gpt-oss-puzzle-88B, a deployment-optimized derivative. Our approach combines heterogeneous MoE expert pruning, selective replacement of full-context attention with window attention, FP8 KV-cache quantization with calibrated scales, and post-training reinforcement learning to recover accuracy, while maintaining low generation length. In terms of per-token speeds, on an 8XH100 node we achieve 1.63X and 1.22X throughput speedups in long-context and short-context settings, respectively. gpt-oss-puzzle-88B also delivers throughput speedups of 2.82X on a single NVIDIA H100 GPU. However, because token counts can change with reasoning effort and model variants, per-token throughput (tok/s) and latency (ms/token) do not necessarily lead to end-to-end speedups: a 2X throughput gain is erased if traces grow 2X. Conversely, throughput gains can be spent on more reasoning tokens to improve accuracy; we therefore advocate request-level efficiency metrics that normalize throughput by tokens generated and trace an accuracy--speed frontier across reasoning efforts. We show that gpt-oss-puzzle-88B improves over gpt-oss-120B along the entire frontier, delivering up to 1.29X higher request-level efficiency. Across various benchmarks, gpt-oss-puzzle-88B matches or slightly exceeds the parent on suite-average accuracy across reasoning efforts, with retention ranging from 100.8% (high) to 108.2% (low), showing that post-training architecture search can substantially reduce inference costs without sacrificing quality.

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.

Llama-Nemotron: Efficient Reasoning Models

We introduce the Llama-Nemotron series of models, an open family of heterogeneous reasoning models that deliver exceptional reasoning capabilities, inference efficiency, and an open license for enterprise use. The family comes in three sizes -- Nano (8B), Super (49B), and Ultra (253B) -- and performs competitively with state-of-the-art reasoning models such as DeepSeek-R1 while offering superior inference throughput and memory efficiency. In this report, we discuss the training procedure for these models, which entails using neural architecture search from Llama 3 models for accelerated inference, knowledge distillation, and continued pretraining, followed by a reasoning-focused post-training stage consisting of two main parts: supervised fine-tuning and large scale reinforcement learning. Llama-Nemotron models are the first open-source models to support a dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference. To further support open research and facilitate model development, we provide the following resources: 1. We release the Llama-Nemotron reasoning models -- LN-Nano, LN-Super, and LN-Ultra -- under the commercially permissive NVIDIA Open Model License Agreement. 2. We release the complete post-training dataset: Llama-Nemotron-Post-Training-Dataset. 3. We also release our training codebases: NeMo, NeMo-Aligner, and Megatron-LM.

FFN Fusion: Rethinking Sequential Computation in Large Language Models

We introduce FFN Fusion, an architectural optimization technique that reduces sequential computation in large language models by identifying and exploiting natural opportunities for parallelization. Our key insight is that sequences of Feed-Forward Network (FFN) layers, particularly those remaining after the removal of specific attention layers, can often be parallelized with minimal accuracy impact. We develop a principled methodology for identifying and fusing such sequences, transforming them into parallel operations that significantly reduce inference latency while preserving model behavior. Applying these techniques to Llama-3.1-405B-Instruct, we create Llama-Nemotron-Ultra-253B-Base (Ultra-253B-Base), an efficient and soon-to-be publicly available model that achieves a 1.71X speedup in inference latency and 35X lower per-token cost while maintaining strong performance across benchmarks. Through extensive experiments on models from 49B to 253B parameters, we demonstrate that FFN Fusion becomes increasingly effective at larger scales and can complement existing optimization techniques like quantization and pruning. Most intriguingly, we find that even full transformer blocks containing both attention and FFN layers can sometimes be parallelized, suggesting new directions for neural architecture design.

No Data, No Optimization: A Lightweight Method To Disrupt Neural Networks With Sign-Flips

Deep Neural Networks (DNNs) can be catastrophically disrupted by flipping only a handful of sign bits in their parameters. We introduce Deep Neural Lesion (DNL), a data-free, lightweight method that locates these critical parameters and triggers massive accuracy drops. We validate its efficacy on a wide variety of computer vision models and datasets. The method requires no training data or optimization and can be carried out via common exploits software, firmware or hardware based attack vectors. An enhanced variant that uses a single forward and backward pass further amplifies the damage beyond DNL's zero-pass approach. Flipping just two sign bits in ResNet50 on ImageNet reduces accuracy by 99.8\%. We also show that selectively protecting a small fraction of vulnerable sign bits provides a practical defense against such attacks.

Padding Tone: A Mechanistic Analysis of Padding Tokens in T2I Models

Text-to-image (T2I) diffusion models rely on encoded prompts to guide the image generation process. Typically, these prompts are extended to a fixed length by adding padding tokens before text encoding. Despite being a default practice, the influence of padding tokens on the image generation process has not been investigated. In this work, we conduct the first in-depth analysis of the role padding tokens play in T2I models. We develop two causal techniques to analyze how information is encoded in the representation of tokens across different components of the T2I pipeline. Using these techniques, we investigate when and how padding tokens impact the image generation process. Our findings reveal three distinct scenarios: padding tokens may affect the model's output during text encoding, during the diffusion process, or be effectively ignored. Moreover, we identify key relationships between these scenarios and the model's architecture (cross or self-attention) and its training process (frozen or trained text encoder). These insights contribute to a deeper understanding of the mechanisms of padding tokens, potentially informing future model design and training practices in T2I systems.

Puzzle: Distillation-Based NAS for Inference-Optimized LLMs

Large language models (LLMs) have demonstrated remarkable capabilities, but their adoption is limited by high computational costs during inference. While increasing parameter counts enhances accuracy, it also widens the gap between state-of-the-art capabilities and practical deployability. We present Puzzle, a framework to accelerate LLM inference on specific hardware while preserving their capabilities. Through an innovative application of neural architecture search (NAS) at an unprecedented scale, Puzzle systematically optimizes models with tens of billions of parameters under hardware constraints. Our approach utilizes blockwise local knowledge distillation (BLD) for parallel architecture exploration and employs mixed-integer programming for precise constraint optimization. We demonstrate the real-world impact of our framework through Llama-3.1-Nemotron-51B-Instruct (Nemotron-51B), a publicly available model derived from Llama-3.1-70B-Instruct. Nemotron-51B achieves a 2.17x inference throughput speedup, fitting on a single NVIDIA H100 GPU while preserving 98.4% of the original model's capabilities. Nemotron-51B currently stands as the most accurate language model capable of inference on a single GPU with large batch sizes. Remarkably, this transformation required just 45B training tokens, compared to over 15T tokens used for the 70B model it was derived from. This establishes a new paradigm where powerful models can be optimized for efficient deployment with only negligible compromise of their capabilities, demonstrating that inference performance, not parameter count alone, should guide model selection. With the release of Nemotron-51B and the presentation of the Puzzle framework, we provide practitioners immediate access to state-of-the-art language modeling capabilities at significantly reduced computational costs.

Hierarchical Selective Classification

Deploying deep neural networks for risk-sensitive tasks necessitates an uncertainty estimation mechanism. This paper introduces hierarchical selective classification, extending selective classification to a hierarchical setting. Our approach leverages the inherent structure of class relationships, enabling models to reduce the specificity of their predictions when faced with uncertainty. In this paper, we first formalize hierarchical risk and coverage, and introduce hierarchical risk-coverage curves. Next, we develop algorithms for hierarchical selective classification (which we refer to as "inference rules"), and propose an efficient algorithm that guarantees a target accuracy constraint with high probability. Lastly, we conduct extensive empirical studies on over a thousand ImageNet classifiers, revealing that training regimes such as CLIP, pretraining on ImageNet21k and knowledge distillation boost hierarchical selective performance.

What Can We Learn From The Selective Prediction And Uncertainty Estimation Performance Of 523 Imagenet Classifiers

When deployed for risk-sensitive tasks, deep neural networks must include an uncertainty estimation mechanism. Here we examine the relationship between deep architectures and their respective training regimes, with their corresponding selective prediction and uncertainty estimation performance. We consider some of the most popular estimation performance metrics previously proposed including AUROC, ECE, AURC as well as coverage for selective accuracy constraint. We present a novel and comprehensive study of selective prediction and the uncertainty estimation performance of 523 existing pretrained deep ImageNet classifiers that are available in popular repositories. We identify numerous and previously unknown factors that affect uncertainty estimation and examine the relationships between the different metrics. We find that distillation-based training regimes consistently yield better uncertainty estimations than other training schemes such as vanilla training, pretraining on a larger dataset and adversarial training. Moreover, we find a subset of ViT models that outperform any other models in terms of uncertainty estimation performance. For example, we discovered an unprecedented 99% top-1 selective accuracy on ImageNet at 47% coverage (and 95% top-1 accuracy at 80%) for a ViT model, whereas a competing EfficientNet-V2-XL cannot obtain these accuracy constraints at any level of coverage. Our companion paper, also published in ICLR 2023 (A framework for benchmarking class-out-of-distribution detection and its application to ImageNet), examines the performance of these classifiers in a class-out-of-distribution setting.