arXiv subjects

Abhinav Khattar

Publications and source records attributed to Abhinav Khattar.

16 recordsLinked to original sources

arXiv2026

Nemotron-Labs-3-Puzzle-75B-A9B: Compressing Hybrid MoE LLMs

We present Nemotron-Labs-3-Puzzle-75B-A9B, a compressed variant of Nemotron-3-Super optimized for interactive deployment. We designed the model to maximize server throughput under high user throughput constraints. In interactive serving workloads on a single 8xB200 node, Puzzle-75B-A9B achieves approximately 2x higher server throughput than Nemotron-3-Super at matched user throughput constraints. In ultra-long-context deployment on a single H100 GPU, the compressed model increases 1M-token concurrency from 1 request to 8 requests. Puzzle-75B-A9B is constructed using a multi-stage pipeline that combines the Iterative Puzzle compression framework with knowledge distillation, reinforcement learning, quantization, and a Multi-Token Prediction head. The compression process jointly optimizes heterogeneous MoE pruning, active parameter budget, and Mamba pruning to improve inference efficiency while preserving model quality. We evaluate Puzzle-75B-A9B on a broad suite of reasoning, coding, multilingual, long-context, and agentic benchmarks. Despite substantial compression, the model retains strong downstream accuracy relative to the parent model across a wide range of tasks. These results demonstrate that large hybrid MoE models can be substantially optimized for deployment efficiency while maintaining strong downstream capability. Our model is publicly available on Hugging Face.

cs.AI↗

arXiv2026

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

We introduce Nemotron 3 Ultra, a 550 billion total and 55 billion active parameter Mixture-of-Experts Hybrid Mamba-Attention language model. We pre-trained Nemotron 3 Ultra on 20 trillion text tokens, then extended the context length to 1M tokens, and post-trained using Supervised Fine Tuning (SFT), Reinforcement Learning (RL), and Multi-teacher On-Policy Distillation (MOPD). Nemotron 3 Ultra is our most capable model yet, employing multiple key technologies - LatentMoE, Multi Token Prediction (MTP), NVFP4 pre-training, multi-environment RLVR, MOPD, and reasoning budget control. Nemotron 3 Ultra achieves up to ~6x higher inference throughput as compared to state-of-the-art publicly available LLMs while attaining on-par accuracy. The state-of-the-art accuracy, high inference throughput, and 1M token context length make Nemotron 3 Ultra ideal for long-running autonomous agentic tasks. We open-source the base, post-trained, and quantized checkpoints, along with the training data and recipe on HuggingFace.

cs.CL↗

arXiv2026

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

We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.

cs.LG↗

arXiv2026

Scalable Training of Mixture-of-Experts Models with Megatron Core

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.

cs.DC↗

arXiv2026

LatentMoE: Toward Optimal Accuracy per FLOP and Parameter in Mixture of Experts

Mixture of Experts (MoEs) have become a central component of many state-of-the-art open-source and proprietary large language models. Despite their widespread adoption, it remains unclear how close existing MoE architectures are to optimal with respect to inference cost, as measured by accuracy per floating-point operation and per parameter. In this work, we revisit MoE design from a hardware-software co-design perspective, grounded in empirical and theoretical considerations. We characterize key performance bottlenecks across diverse deployment regimes, spanning offline high-throughput execution and online, latency-critical inference. Guided by these insights, we introduce LatentMoE, a new model architecture resulting from systematic design exploration and optimized for maximal accuracy per unit of compute. Empirical design space exploration at scales of up to 95B parameters and over a 1T-token training horizon, together with supporting theoretical analysis, shows that LatentMoE consistently outperforms standard MoE architectures in terms of accuracy per FLOP and per parameter. Given its strong performance, the LatentMoE architecture has been adopted by the flagship Nemotron-3 Super and Ultra models and scaled to substantially larger regimes, including longer token horizons and larger model sizes, as reported in Nvidia et al. (arXiv:2512.20856).

cs.LG↗

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

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.

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↗

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

Upcycling Large Language Models into Mixture of Experts

Upcycling pre-trained dense language models into sparse mixture-of-experts (MoE) models is an efficient approach to increase the model capacity of already trained models. However, optimal techniques for upcycling at scale remain unclear. In this work, we conduct an extensive study of upcycling methods and hyperparameters for billion-parameter scale language models. We propose a novel "virtual group" initialization scheme and weight scaling approach to enable upcycling into fine-grained MoE architectures. Through ablations, we find that upcycling outperforms continued dense model training. In addition, we show that softmax-then-topK expert routing improves over topK-then-softmax approach and higher granularity MoEs can help improve accuracy. Finally, we upcycled Nemotron-4 15B on 1T tokens and compared it to a continuously trained version of the same model on the same 1T tokens: the continuous trained model achieved 65.3% MMLU, whereas the upcycled model achieved 67.6%. Our results offer insights and best practices to effectively leverage upcycling for building MoE language models. Code is available.

cs.CL↗

arXiv2021

Analysis on Image Set Visual Question Answering

We tackle the challenge of Visual Question Answering in multi-image setting for the ISVQA dataset. Traditional VQA tasks have focused on a single-image setting where the target answer is generated from a single image. Image set VQA, however, comprises of a set of images and requires finding connection between images, relate the objects across images based on these connections and generate a unified answer. In this report, we work with 4 approaches in a bid to improve the performance on the task. We analyse and compare our results with three baseline models - LXMERT, HME-VideoQA and VisualBERT - and show that our approaches can provide a slight improvement over the baselines. In specific, we try to improve on the spatial awareness of the model and help the model identify color using enhanced pre-training, reduce language dependence using adversarial regularization, and improve counting using regression loss and graph based deduplication. We further delve into an in-depth analysis on the language bias in the ISVQA dataset and show how models trained on ISVQA implicitly learn to associate language more strongly with the final answer.

cs.CV↗

arXiv2019

Elites Tweet? Characterizing the Twitter Verified User Network

Social network and publishing platforms, such as Twitter, support the concept of verification. Verified accounts are deemed worthy of platform-wide public interest and are separately authenticated by the platform itself. There have been repeated assertions by these platforms about verification not being tantamount to endorsement. However, a significant body of prior work suggests that possessing a verified status symbolizes enhanced credibility in the eyes of the platform audience. As a result, such a status is highly coveted among public figures and influencers. Hence, we attempt to characterize the network of verified users on Twitter and compare the results to similar analysis performed for the entire Twitter network. We extracted the entire network of verified users on Twitter (as of July 2018) and obtained 231,246 user profiles and 79,213,811 connections. Subsequently in the network analysis, we found that the sub-graph of verified users mirrors the full Twitter users graph in some aspects such as possessing a short diameter. However, our findings contrast with earlier findings on multiple aspects, such as the possession of a power law out-degree distribution, slight dissortativity and a significantly higher reciprocity rate, as elucidated in the paper. Moreover, we attempt to gauge the presence of salient components within this sub-graph and detect the absence of homophily with respect to popularity, which again is in stark contrast to the full Twitter graph. Finally, we demonstrate stationarity in the time series of verified user activity levels. To the best of our knowledge, this work represents the first quantitative attempt at characterizing verified users on Twitter.

cs.SI↗

arXiv2019

What sets Verified Users apart? Insights, Analysis and Prediction of Verified Users on Twitter

Social network and publishing platforms, such as Twitter, support the concept of a secret proprietary verification process, for handles they deem worthy of platform-wide public interest. In line with significant prior work which suggests that possessing such a status symbolizes enhanced credibility in the eyes of the platform audience, a verified badge is clearly coveted among public figures and brands. What are less obvious are the inner workings of the verification process and what being verified represents. This lack of clarity, coupled with the flak that Twitter received by extending aforementioned status to political extremists in 2017, backed Twitter into publicly admitting that the process and what the status represented needed to be rethought. With this in mind, we seek to unravel the aspects of a user's profile which likely engender or preclude verification. The aim of the paper is two-fold: First, we test if discerning the verification status of a handle from profile metadata and content features is feasible. Second, we unravel the features which have the greatest bearing on a handle's verification status. We collected a dataset consisting of profile metadata of all 231,235 verified English-speaking users (as of July 2018), a control sample of 175,930 non-verified English-speaking users and all their 494 million tweets over a one year collection period. Our proposed models are able to reliably identify verification status (Area under curve AUC > 99%). We show that number of public list memberships, presence of neutral sentiment in tweets and an authoritative language style are the most pertinent predictors of verification status. To the best of our knowledge, this work represents the first attempt at discerning and classifying verification worthy users on Twitter.

cs.SI↗

arXiv2018

Collective Classification of Spam Campaigners on Twitter: A Hierarchical Meta-Path Based Approach

Cybercriminals have leveraged the popularity of a large user base available on Online Social Networks to spread spam campaigns by propagating phishing URLs, attaching malicious contents, etc. However, another kind of spam attacks using phone numbers has recently become prevalent on OSNs, where spammers advertise phone numbers to attract users' attention and convince them to make a call to these phone numbers. The dynamics of phone number based spam is different from URL-based spam due to an inherent trust associated with a phone number. While previous work has proposed strategies to mitigate URL-based spam attacks, phone number based spam attacks have received less attention. In this paper, we aim to detect spammers that use phone numbers to promote campaigns on Twitter. To this end, we collected information about 3,370 campaigns spread by 670,251 users. We model the Twitter dataset as a heterogeneous network by leveraging various interconnections between different types of nodes present in the dataset. In particular, we make the following contributions: (i) We propose a simple yet effective metric, called Hierarchical Meta-Path Score (HMPS) to measure the proximity of an unknown user to the other known pool of spammers. (ii) We design a feedback-based active learning strategy and show that it significantly outperforms three state-of-the-art baselines for the task of spam detection. Our method achieves 6.9% and 67.3% higher F1-score and AUC, respectively compared to the best baseline method. (iii) To overcome the problem of less training instances for supervised learning, we show that our proposed feedback strategy achieves 25.6% and 46% higher F1-score and AUC respectively than other oversampling strategies. Finally, we perform a case study to show how our method is capable of detecting those users as spammers who have not been suspended by Twitter (and other baselines) yet.

cs.SI↗

arXiv2018

White or Blue, the Whale gets its Vengeance: A Social Media Analysis of the Blue Whale Challenge

The Blue Whale Challenge is a series of self-harm causing tasks that are propagated via online social media under the disguise of a "game." The list of tasks must be completed in a duration of 50 days and they cause both physical and mental harm to the player. The final task is to commit suicide. The game is supposed to be administered by people called "curators" who incite others to cause self-mutilation and commit suicide. The curators and potential players are known to contact each other on social networking websites and the conversations between them are suspected to take place mainly via direct messages which are difficult to track. Though, in order to find curators, the players make public posts containing certain hashtags/keywords to catch their attention. Even though a lot of these social networks have moderated posts talking about the game, yet some posts manage to pass their filters. Our research focuses on (1) understanding the social media spread of the challenge, (2) spotting the behaviour of the people taking interest in Blue Whale challenge and, (3) analysing demographics of the users who may be involved in playing the game.

cs.SI↗