arXiv subjects

Charles Wang

Publications and source records attributed to Charles Wang.

At least 19 recordsLinked to original sources

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 Nano Omni: Efficient and Open Multimodal Intelligence

We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, images, and video. Nemotron 3 Nano Omni delivers consistent accuracy improvements over its predecessor, Nemotron Nano V2 VL, across all modalities, enabled by advances in architecture, training data and recipes. In particular, Nemotron 3 delivers leading results in real-world document understanding, long audio-video comprehension, and agentic computer use. Built on the highly efficient Nemotron 3 Nano 30B-A3B backbone, Nemotron 3 Nano Omni further incorporates innovative multimodal token-reduction techniques to deliver substantially lower inference latency and higher throughput than other models of similar size. We are releasing model checkpoints in BF16, FP8, and FP4 formats, along with portions of the training data and codebase to facilitate further research and development.

cs.LG↗

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

HiL-Bench (Human-in-Loop Benchmark): Do Agents Know When to Ask for Help?

Frontier coding agents solve complex tasks when given complete context but collapse when specifications are incomplete or ambiguous. The bottleneck is not raw capability, but judgment: knowing when to act autonomously and when to ask for help. Current benchmarks are blind to this failure mode. They supply unambiguous detailed instructions and solely reward execution correctness, so an agent that makes a lucky guess for a missing requirement will score identically to one that would have asked to be certain. We present HiL-Bench (Human-in-the-Loop Benchmark) to measure this selective escalation skill. Each task contains human-validated blockers (missing information, ambiguous requests, contradictory information) that surface only through progressive exploration, not upfront inspection. Our core metric, Ask-F1, the harmonic mean of question precision and blocker recall, captures the tension between over-asking and silent guessing; its structure architecturally prevents gaming through question spam. Evaluation across SWE and text-to-SQL domains reveals a large universal judgment gap: no frontier model recovers more than a fraction of its full-information performance when deciding whether to ask. Failure analysis identifies three key help-seeking patterns: overconfident wrong beliefs with no gap detection; high uncertainty detection yet persistent errors; broad, imprecise escalation without self-correction. These consistent patterns confirm poor help-seeking is a model-level flaw, not task-specific. RL training on shaped Ask-F1 reward shows judgment is trainable: a 32B model improves both help-seeking quality and task pass rate, with gains that transfer across domains. The model does not learn domain-specific heuristics for when to ask; it learns to detect unresolvable uncertainty and act on it.

cs.AI↗

arXiv2026

PivotRL: High Accuracy Agentic Post-Training at Low Compute Cost

Post-training for long-horizon agentic tasks has a tension between compute efficiency and generalization. While supervised fine-tuning (SFT) is compute efficient, it often suffers from out-of-domain (OOD) degradation. Conversely, end-to-end reinforcement learning (E2E RL) preserves OOD capabilities, but incurs high compute costs due to many turns of on-policy rollout. We introduce PivotRL, a novel framework that operates on existing SFT trajectories to combine the compute efficiency of SFT with the OOD accuracy of E2E RL. PivotRL relies on two key mechanisms: first, it executes local, on-policy rollouts and filters for pivots: informative intermediate turns where sampled actions exhibit high variance in outcomes; second, it utilizes rewards for functional-equivalent actions rather than demanding strict string matching with the SFT data demonstration. We theoretically show that these mechanisms incentivize strong learning signals with high natural gradient norm, while maximally preserving policy probability ordering on actions unrelated to training tasks. In comparison to standard SFT on identical data, we demonstrate that PivotRL achieves +4.17% higher in-domain accuracy on average across four agentic domains, and +10.04% higher OOD accuracy in non-agentic tasks. Notably, on agentic coding tasks, PivotRL achieves competitive accuracy with E2E RL with 4x fewer rollout turns. PivotRL is adopted by NVIDIA's Nemotron-3-Super-120B-A12B, acting as the workhorse in production-scale agentic post-training.

cs.AI↗

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

A Comparative Study of Retrieval Methods in Azure AI Search

Increasingly, attorneys are interested in moving beyond keyword and semantic search to improve the efficiency of how they find key information during a document review task. Large language models (LLMs) are now seen as tools that attorneys can use to ask natural language questions of their data during document review to receive accurate and concise answers. This study evaluates retrieval strategies within Microsoft Azure's Retrieval-Augmented Generation (RAG) framework to identify effective approaches for Early Case Assessment (ECA) in eDiscovery. During ECA, legal teams analyze data at the outset of a matter to gain a general understanding of the data and attempt to determine key facts and risks before beginning full-scale review. In this paper, we compare the performance of Azure AI Search's keyword, semantic, vector, hybrid, and hybrid-semantic retrieval methods. We then present the accuracy, relevance, and consistency of each method's AI-generated responses. Legal practitioners can use the results of this study to enhance how they select RAG configurations in the future.

cs.IR↗

arXiv2025

Leveraging Machine Learning and Large Language Models for Automated Image Clustering and Description in Legal Discovery

The rapid increase in digital image creation and retention presents substantial challenges during legal discovery, digital archive, and content management. Corporations and legal teams must organize, analyze, and extract meaningful insights from large image collections under strict time pressures, making manual review impractical and costly. These demands have intensified interest in automated methods that can efficiently organize and describe large-scale image datasets. This paper presents a systematic investigation of automated cluster description generation through the integration of image clustering, image captioning, and large language models (LLMs). We apply K-means clustering to group images into 20 visually coherent clusters and generate base captions using the Azure AI Vision API. We then evaluate three critical dimensions of the cluster description process: (1) image sampling strategies, comparing random, centroid-based, stratified, hybrid, and density-based sampling against using all cluster images; (2) prompting techniques, contrasting standard prompting with chain-of-thought prompting; and (3) description generation methods, comparing LLM-based generation with traditional TF-IDF and template-based approaches. We assess description quality using semantic similarity and coverage metrics. Results show that strategic sampling with 20 images per cluster performs comparably to exhaustive inclusion while significantly reducing computational cost, with only stratified sampling showing modest degradation. LLM-based methods consistently outperform TF-IDF baselines, and standard prompts outperform chain-of-thought prompts for this task. These findings provide practical guidance for deploying scalable, accurate cluster description systems that support high-volume workflows in legal discovery and other domains requiring automated organization of large image collections.

cs.IR↗

arXiv2025

From Bits to Rounds: Parallel Decoding with Exploration for Diffusion Language Models

Diffusion Language Models (DLMs) have recently emerged as a strong alternative to autoregressive language models (LMs). DLMs offer comparable accuracy with faster inference speed via parallel decoding. However, standard DLM decoding strategies relying on high-confidence tokens encounter an inherent information-theoretic bottleneck that restricts decoding progress and ultimately slows generation. We demonstrate both theoretically and empirically that prioritizing high-confidence tokens is inherently inefficient. High-probability tokens carry negligible information and strictly relying on them limits the effective progress made in each decoding round. We prove that the number of decoding rounds must grow linearly with the sample's total information (negative log-likelihood) and inversely with the per-round information budget, establishing a bits-to-rounds principle. We also propose Explore-Then-Exploit (ETE), a training-free decoding strategy that maximizes information throughput and decoding efficiency. ETE combines cross-block decoding with targeted exploration of high-uncertainty tokens to reshape the conditional distribution and trigger cascades of confident predictions. Experiments verify our theoretical bounds and demonstrate that ETE consistently reduces the required number of decoding rounds compared to confidence-only baselines without compromising generation quality.

cs.LG↗

arXiv2025

NVIDIA Nemotron Nano V2 VL

We introduce Nemotron Nano V2 VL, the latest model of the Nemotron vision-language series designed for strong real-world document understanding, long video comprehension, and reasoning tasks. Nemotron Nano V2 VL delivers significant improvements over our previous model, Llama-3.1-Nemotron-Nano-VL-8B, across all vision and text domains through major enhancements in model architecture, datasets, and training recipes. Nemotron Nano V2 VL builds on Nemotron Nano V2, a hybrid Mamba-Transformer LLM, and innovative token reduction techniques to achieve higher inference throughput in long document and video scenarios. We are releasing model checkpoints in BF16, FP8, and FP4 formats and sharing large parts of our datasets, recipes and training code.

cs.LG↗

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

Zebra-CoT: A Dataset for Interleaved Vision Language Reasoning

Humans often use visual aids, for example diagrams or sketches, when solving complex problems. Training multimodal models to do the same, known as Visual Chain of Thought (Visual CoT), is challenging due to: (1) poor off-the-shelf visual CoT performance, which hinders reinforcement learning, and (2) the lack of high-quality visual CoT training data. We introduce $\textbf{Zebra-CoT}$, a diverse large-scale dataset with 182,384 samples, containing logically coherent interleaved text-image reasoning traces. We focus on four categories of tasks where sketching or visual reasoning is especially natural, spanning scientific questions such as geometry, physics, and algorithms; 2D visual reasoning tasks like visual search and jigsaw puzzles; 3D reasoning tasks including 3D multi-hop inference, embodied and robot planning; visual logic problems and strategic games like chess. Fine-tuning the Anole-7B model on the Zebra-CoT training corpus results in an improvement of +12% in our test-set accuracy and yields up to +13% performance gain on standard VLM benchmark evaluations. Fine-tuning Bagel-7B yields a model that generates high-quality interleaved visual reasoning chains, underscoring Zebra-CoT's effectiveness for developing multimodal reasoning abilities. We open-source our dataset and models to support development and evaluation of visual CoT.

cs.CV↗

arXiv2025

Diffeomorphism in Closed String Field Theory

We explain how the spacetime diffeomorphism in the classical bosonic closed string field theory (SFT) is represented as $L_\infty$ gauge transformations in weakly curved backgrounds. In particular, we demonstrate the explicit map between covariant spacetime fields and the string fields in the flat-vertex frame to the leading few orders in perturbation theory. We further consider a Batalin-Vilkovisky system of extended string fields that manifests diffeomorphism invariance and allows for extending certain perturbative SFT solutions to the large field regime.

hep-th↗

arXiv2024

Chevalley Polytopes and Newton-Okounkov Bodies

We construct a family of polytopes, which we call Chevalley polytopes, associated to homogeneous spaces $X=G/P$ in their projective embeddings $X\hookrightarrow \mathbb{P}(V_{\varpi})$ together with a choice of reduced expression for the minimal coset representative $w^P$ of $w_0$ in $W/W_P$. When $X$ is minuscule in its minimal embedding, we describe our construction in terms of order polytopes of minuscule posets and use the associated combinatorics to show that minuscule Chevalley polytopes are Newton-Okounkov bodies for $X$ and that the Pl\"ucker coordinates on $X$ form a Khovanskii basis for $\mathbb{C}[X]$. We conjecture similar properties for general $X$ and general embeddings $X\hookrightarrow\mathbb{P}(V_\varpi)$, along with a remarkable decomposition property which we consider as a polytopal shadow of the Littlewood-Richardson rule. We highlight a connection between Chevalley polytopes and string polytopes and give examples where Chevalley polytopes possess better combinatorial properties than string polytopes. We conclude with several examples further illustrating and supporting our conjectures.

math.AG↗

arXiv2024

Canonical Landau-Ginzburg models for cominuscule homogeneous spaces

We present a type-independent Landau-Ginzburg (LG) model $(X_\mathrm{can}, \mathcal{W}_\mathrm{can})$ for any cominuscule homogeneous space $X=G/P$. We give a fully combinatorial construction for our superpotential $\mathcal{W}_\mathrm{can}$ as a sum of $n+1$ rational functions in the (generalized) Pl\"ucker coordinates on the "Langlands dual" minuscule homogeneous space $\mathbb{X}=P^\vee\backslash G^\vee$. Explicitly, we define the denominators $\mathcal{D}_{i_*}$ of these rational functions using the combinatorics of order ideals of the corresponding minuscule poset, which can be interpreted as (generalized) Young diagrams, by a process that can be described by "moving boxes" and hence is easily implemented. To construct the corresponding numerators, we define derivations $\delta_{i_1}$ on $\mathbb{C}[\mathbb{X}]$ that act by "adding an appropriate box if possible" and then we apply each $\delta_{i_1}$ to the corresponding $\mathcal{D}_{i_*}$. By studying certain Weyl orbits in the fundamental representations of $\widetilde{G}^\vee$ and exploiting the existence of a certain dense algebraic torus in $\mathbb{X}$, we show that the polynomials $\mathcal{D}_{i_*}$ coincide with the generalized minors $\phi_{i_*}$ appearing in the cluster structures for homogeneous spaces studied by Gei\ss-Leclerc-Schr\"oer in arXiv:math/0609138. We then define the mirror variety $X_\mathrm{can}=\mathbb{X}\setminus D_\mathrm{ac}$ to be the complement of the anticanonical divisor $D_\mathrm{ac} = \sum_{i_*}\{\mathcal{D}_{i_*}=0\}$ formed by the $\mathcal{D}_{i_*}$. Moreover, we show that the LG models $(X_\mathrm{can},\mathcal{W}_\mathrm{can})$ are isomorphic to the Lie-theoretic LG-models $(X_\mathrm{Lie},\mathcal{W}_\mathrm{Lie})$ constructed by Rietsch in arXiv:math/0511124 and our models naturally generalize the type-dependent Pl\"ucker coordinate LG-models previously studied by various authors.

math.AG↗

arXiv2024

Conformal Perturbation Theory and Tachyon-Dilaton Eschatology via String Fields

We analyze deformations of two-dimensional conformal field theory (CFT) from the perspective of classical bosonic closed string field theory (SFT). The latter can be viewed as a version of Wilsonian renormalization group (RG) improved conformal perturbation theory, where the renormalization scheme is defined through the choice of string vertices in the construction of SFT. Furthermore, the CFT data at the RG fixed point can be recovered from the spectrum and amplitudes of string field fluctuations. As applications, we construct the Horowitz-Polchinski "string star" solution in SFT, and a solution of tachyon-dilaton condensation that deforms the noncompact boson to minimal models by creating a pair of "Runkel-Watts walls".

hep-th↗

arXiv2023

Canonical mirror models for maximal orthogonal Grassmannians

In this report we use our methods in arXiv:2204.03548 and its upcoming generalization to complete the construction of canonical mirror models for all cominuscule homogeneous spaces, by considering the maximal orthogonal Grassmannians $\mathrm{OG}(n + 1, 2n + 2)$.

math.AG↗