Searcharxiv⌕ Search

arXiv subjects

Tao Ge

Publications and source records attributed to Tao Ge.

At least 19 recordsLinked to original sources

Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet

Kilometer-scale convection shapes precipitation extremes, tropical organization, and cloud feedbacks, but most global atmospheric models approximate these processes at 25-100 km resolution. Global storm-resolving physics models resolve convective systems explicitly, but at a cost (roughly one MWh per simulated day on exascale supercomputers) that limits long-duration simulation. We introduce STRATA (Storm-resolving Tile-based autoRegressive Atmosphere Transformer Architecture), the first autoregressive AI emulator for global storm-resolving atmospheric dynamics. STRATA is trained on the highest-resolution atmospheric dataset yet used for global AI emulation: 17 days of SCREAM physics-model output at 4.9-km resolution (~25 million grid cells) sampled every 10 minutes. Our central premise is that on 10-minute timescales atmospheric dynamics are predominantly local, so training on small spatial tiles trades scarce global temporal samples for abundant local spatial samples and enables global rollout via overlapping-tile blending. STRATA combines 3D patch embedding and local 3D neighborhood attention, a novel Stereographic Rotary Position Embedding (StereoRoPE) for grid-invariant encoding, and a pixel-space de-aliasing decoder that suppresses patch-scale rollout artifacts. An iso-FLOP scaling study reveals that km-scale emulation requires ~10x more FLOPs per grid point than coarse-resolution AI weather models, consistent with the higher information density of convective-scale dynamics. Trained on only 17 days of data, STRATA produces stable 24-hour global rollouts with realistic km-scale dynamics across diverse regimes, though large-scale biases develop with lead time. It achieves 48 simulated days per megawatt-hour (more than 50 times better energy efficiency than the SCREAM physics model) and 741 simulated days per wall-clock day at 512 H100 GPUs.

physics.ao-ph↗

WordInferBench: Can Agents Infer How to Edit from Word Document Context?

Document editing is among the most common knowledge-work tasks, yet existing benchmarks for evaluating autonomous agents on Word document editing provide fully specified attribute-modification instructions. They do not evaluate agents' ability to infer the appropriate attributes from document context, a capability essential in real-world editing where users expect consistency without spelling out every attribute. We introduce WordInferBench, a benchmark comprising 280 base document-editing instances, each paired with three instruction conditions: descriptive (the instruction explicitly lists every attribute change), referential (the instruction points to another object as an exemplar), and inferential (the agent must examine document context to infer what attributes to apply). To evaluate contextual attribute inference, each base instance is constructed so that the target formatting is recoverable from an in-document exemplar. The instances are derived from real Word documents collected from Common Crawl and cover 13 object types and 81 attributes. We evaluate six recent coding-agent systems and find that the best-performing evaluated system, Claude Code with Opus 4.6, achieves a 62.9% task-level pass rate on descriptive variants but only 40.7% on inferential variants, indicating that current systems complete tasks less reliably when the required formatting changes must be determined from document context.

cs.AI↗

WordEval: Evaluating Word-Native Operation Fidelity in Document Editing

Microsoft Word is a central environment for office automation, and the faithful execution of its editing operations is an important capability for document agents. Unlike plain-text editing, editing in Word modifies a structured object model, where the requested effect is realized in object properties. Existing benchmarks, however, evaluate office workflows or desktop automation, treating word-processor documents as one task domain among many and verifying outcomes with keyword matching, per-task property checks, or screenshot comparison. We introduce WordEval, an operation-fidelity benchmark that instead centers evaluation on Word editing operations, testing whether each requested edit is realized as intended changes to Word objects and properties. It contains 642 cases across 15 Word feature areas, ranging from text and lists to charts, shape geometry, and SmartArt. 546 atomic cases cover 231 commands, 96 compositional cases combine multiple edits in a single document, and a Word-native oracle checks task-specific object properties and preservation conditions. Cases are constructed from editing-unit specifications, yielding controlled input documents, reference documents, and aligned oracle checks. Each compositional case is linked to corresponding atomic cases, supporting fine-grained diagnosis of editing failures. We evaluate eight agent systems and find that many struggle on the Atomic track, with performance varying across feature areas. Moreover, every system scores lower on Compositional than on Atomic, with gaps that vary substantially by system. Exact atomic-intent links and recorded failure diagnoses provide complementary views of this variation.

cs.CV↗

Skillful Thunderstorm Forecasts Learned Directly from Observations

Nowcasting the evolution of severe convective storms over the first few hours after observations are collected is critical for issuing timely, accurate warnings. Forecast accuracy is hindered because the variables that matter most, such as rain, hail, and lightning, are imperfectly represented by numerical model physics and are difficult to initialize correctly. Here we introduce StormScope, a computationally efficient generative forecasting system that learns storm-scale evolution directly from satellite, radar, and lightning observations over the continental United States with no numerical model used for training or initialization. StormScope produces coherent, physically plausible storm trajectories with well-calibrated uncertainty. Across more than 900 forecasts, it greatly exceeds the National Oceanic and Atmospheric Administration's state-of-the-art forthcoming operational numerical modeling system in both deterministic and probabilistic skill through the 0-2-h nowcasting window. StormScope's skill for strong radar echoes exceeds the operational state of the art by approximately 75% and 45% at lead times of 30 min and 2 h, a relative improvement several times larger than machine learning has delivered for global medium-range forecasting. Crucially, it also anticipates storms that develop after forecast initialization, establishing its utility beyond extrapolation of existing radar echoes. It likewise surpasses a leading AI nowcasting model for intense precipitation. Together, these results establish learning directly from observations as a route toward skillful, well-calibrated storm-scale forecasts that is unconstrained by the initialization and microphysics bottlenecks limiting numerical models and their emulators.

physics.ao-ph↗

reGRAF: a global MPAS reforecast data set with convection-allowing refinements over the US and Europe

NVIDIA and The Weather Company (TWC) have generated a data set of reforecasts from TWC's GRAF (Global high-Resolution Atmospheric Forecasting) model, a version of the National Center for Atmospheric Research (NCAR) Model for Predictions Across Scales (\href{https://ncar.ucar.edu/what-we-offer/models/model-prediction-across-scales-mpas}{\underline{MPAS}}). GRAF is global, but the configuration for this reforecast had a mesh refinement to \textasciitilde4 km over the US, Caribbean Basin, and Europe, and 15 km elsewhere. This model was designed to run much of the computation on graphical processing units (GPUs), with this development assisted by NVIDIA. The 1836 reforecast cases (\textasciitilde5 years) were generated from ECMWF reanalyses (ERA5) for selected initial condition dates spanning more than 20 years, 2004--2024. These dates of the chosen initial conditions were mostly selected based on high-impact weather in the contiguous US (CONUS) and Caribbean. Sampling in this way, the reforecast spanned a wider range of interesting, high-impact weather scenarios than had we performed five contiguous years of once-daily reforecasts. The reforecast still provides many samples in non-precipitating regions with more ordinary weather. GRAF reforecasts were mostly run to $+$27 h lead time, assuming a 3-h spin-up followed by a full diurnal cycle. Data were saved in zarr format on the native model vertical coordinate. Most fields were archived at 15-min intervals, though several precipitation variables were saved at 5-min cadence. Data are made publicly available to all through Amazon Web Services' Open-Data Initiative at https://registry.opendata.aws/graf-reforecast.

physics.ao-ph↗

ExecCritic: Learn to Test, Test to Improve for Coding Agents

Execution feedback can guide coding agents toward correct repository repairs, but only when the tests capture the behavior requested by the issue. Agent-generated tests can encode incomplete or incorrect behavioral targets; when the same trajectory writes both the patch and the test, their errors can agree and create false confidence. We introduce ExecCritic, combining a test--verify--revise scaffold with a role-specific reinforcement learning recipe for training agents within it. The scaffold separates test construction from source-code repair: a Test agent independently generates repository-native tests, a fail-closed harness qualifies and freezes them, and a Repair agent revises source code from their execution feedback without changing the tests. Both roles use Qwen-3.5-35B-A3B as the backbone and are trained separately. In Learn to Test, the Test agent learns to produce behaviorally valid tests that distinguish correct from incorrect patches. In Test to Improve, the Repair agent learns both direct task resolution and feedback-guided revision. On SWE-bench Verified, test quality determines whether feedback helps: holding the base Repair agent fixed, tests from the base Test agent reduce resolved rate from a no-test baseline of 61.2% to 57.3%, whereas tests from GPT-5.6-sol raise it to 65.3%. Role-specific post-training raises the Qwen Test agent's Base-to-Gold success from 22.2% to 62.2%; composing the two post-trained Qwen agents reaches 72.6%, an 11.4-point gain over the original no-test baseline without stronger-model or Oracle feedback at evaluation time. Code is publicly available at https://github.com/MSR-Orchard/execcritic.

cs.AI↗

Orchard: An Open-Source Agentic Modeling Framework

Agentic modeling aims to transform LLMs into autonomous agents capable of solving complex tasks through planning, reasoning, tool use, and multi-turn interaction with external environments. We present Orchard, an open-source framework for scalable agentic modeling. At its core is Orchard Env, a lightweight Kubernetes-native environment service that provides reusable primitives for sandbox lifecycle management across task domains, agent harnesses, and training stages. On top of Orchard Env, we build three agentic modeling recipes. Orchard-SWE targets software engineering agents. We introduce credit-assignment supervised fine-tuning and a progression of RL signals: Balanced Adaptive Rollout (BAR) for sparse-reward optimization, on-policy distillation (OPD) and rubric-based process reward (RPR) for dense supervision, and historical experience distillation, which compresses rollouts from prior experiments into a compact value model for inference-time reranking. Built on the Qwen3.5-35B-A3B backbone, Orchard-SWE reaches 69.7% with RPR-based RL and 73.0% with value-model reranking on SWE-bench Verified, setting a new state of the art among open-source methods while approaching frontier systems over 10x larger. Orchard-GUI trains a 4B vision-language computer-use agent using only 0.4K distilled trajectories and 2.2K open-ended tasks, achieving 68.4% average success across WebVoyager, Online-Mind2Web, and DeepShop, making it the strongest open-source model while remaining competitive with proprietary systems. Orchard-Claw targets personal assistant agents. Trained with only 0.2K synthetic tasks, it achieves 59.6% pass@3 on Claw-Eval and 73.9% when paired with the stronger ZeroClaw harness. Collectively, these results demonstrate that a lightweight, open, harness-agnostic environment layer enables reusable agentic data, training recipes, and evaluation protocols across domains.

cs.AI↗

Learn to Memorize: Scalable Continual Learning in Semiparametric Models with Mixture-of-Neighbors Induction Memory

Semiparametric language models (LMs) have shown promise in various Natural Language Processing (NLP) tasks. However, they utilize non-parametric memory as static storage, which lacks learning capability and remains disconnected from the internal information flow of the parametric models, limiting scalability and efficiency. Based on recent interpretability theories of LMs, we reconceptualize the non-parametric memory represented by $k$NN-LM as a learnable Mixture-of-Neighbors Induction Memory (MoNIM), which synergizes the induction capabilities of attention heads with the memorization strength of feed-forward networks (FFN). By integrating into the model's information flow, MoNIM functions as an FFN-like bypass layer within the Transformer architecture, enabling effective learning of new knowledge. Extensive experiments demonstrate that MoNIM is a retentive and scalable continual learner in both data- and model-wise, enhancing the scalability and continual learning performance of semiparametric LMs.

cs.CL↗

TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents

Multi-turn agents solve complex tasks through extended sequences of tool interactions before producing a final answer, making credit assignment a fundamental challenge during post-training. Outcome rewards provide reliable supervision for short-horizon reasoning, but become sparse and high-variance as trajectories grow to tens or hundreds of tool calls. They can also be misleading: a failed rollout may contain many useful actions that move the agent closer to the goal, yet outcome-only training assigns them the same negative advantage as the eventual mistake. We propose TRACE (Turn-level Reward Assignment via Credit Estimation), a dense credit-assignment method for agentic reinforcement learning. TRACE represents rollouts as state transitions at tool-call boundaries, obtains gold-answer log-probabilities from a frozen reference model, transforms them into log-ratio state values, and derives per-action rewards as Temporal-Difference changes in those values. This requires no additional critic or process-label training, and its one-step log-ratio TD component telescopes across redundant tool calls. On long-horizon complex search, TRACE substantially improves base-model tool-use ability using pure RL, without a cold-start supervised fine-tuning stage, an agentic mid-training stage, or training on live-web data. On the closed-web BrowseComp-Plus benchmark, it raises Qwen3-4B from $7.2$ to $35.6$ and Qwen3-30B-A3B from $8.4$ to $42.6$. The learned search behavior also transfers to open-web benchmarks, and the learning curves show earlier improvement and faster convergence during RL training.

cs.LG↗

LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval

We introduce LEDGER to tackle the novel context engineering challenge of agentic document editing, where localized edits to long, structured documents must be applied efficiently without breaking cross-references or semantic consistency. LEDGER constructs a lightweight dependency graph that explicitly models document structure, including hierarchical organization, explicit references, implicit dependencies, and semantic relationships. For each edit, graph-guided retrieval selects only the necessary context, avoiding full-document processing while preserving consistency. We evaluate LEDGER on a curated benchmark of 1.9k test cases with various document types and lengths, spanning six state-of-the-art models: LEDGER improves consistency from 56% to 76% across all six models and test scenarios while reducing token usage. Notably, LEDGER with low reasoning effort matches baseline performance at high reasoning effort using fewer tokens, showing that explicit dependency representations can partially substitute for expensive internal reasoning in agentic document editing.

cs.IR↗

DocReward: A Document Reward Model for Structuring and Stylizing

Recent agentic workflows automate professional document generation but focus narrowly on textual quality, overlooking structural and stylistic professionalism, which is equally critical for readability. This gap stems mainly from a lack of effective reward models capable of guiding agents toward producing documents with high structural and stylistic professionalism. We introduce DocReward, a document reward model that evaluates documents based on their structure and style. To achieve this, we propose a textual-quality-agnostic framework that ensures assessments are not confounded by content quality, and construct DocPair, a dataset of 117K paired documents covering 32 domains and 267 types. Each pair shares identical content but differs in structural and stylistic professionalism. DocReward is trained using the Bradley-Terry loss. On a manually annotated benchmark, DocReward outperforms GPT-5 by 14.6 percentage points in the same setting. Reinforcement learning experiments further show that DocReward effectively guides agents toward generating documents with consistently higher structural and stylistic professionalism, highlighting its practical utility.

cs.CV↗

Synthetic Computers at Scale for Long-Horizon Productivity Simulation

Realistic long-horizon productivity work is strongly conditioned on user-specific computer environments, where much of the work context is stored and organized through directory structures and content-rich artifacts. To scale synthetic data creation for such productivity scenarios, we introduce Synthetic Computers at Scale, a scalable methodology for creating such environments with realistic folder hierarchies and content-rich artifacts (e.g., documents, spreadsheets, and presentations). Conditioned on each synthetic computer, we run long-horizon simulations: one agent creates productivity objectives that are specific to the computer's user and require multiple professional deliverables and about a month of human work; another agent then acts as that user and keeps working across the computer -- for example, navigating the filesystem for grounding, coordinating with simulated collaborators, and producing professional artifacts -- until these objectives are completed. In preliminary experiments, we create 1,000 synthetic computers and run long-horizon simulations on them; each run requires over 8 hours of agent runtime and spans more than 2,000 turns on average. These simulations produce rich experiential learning signals, whose effectiveness is validated by significant improvements in agent performance on both in-domain and out-of-domain productivity evaluations. Given that personas are abundant at billion scale, this methodology can in principle scale to millions or even billions of synthetic user worlds with sufficient compute, enabling broader coverage of diverse professions, roles, contexts, environments, and productivity needs. We argue that scalable synthetic computer creation, together with at-scale simulations, is highly promising as a foundational substrate for agent self-improvement and agentic reinforcement learning in long-horizon productivity scenarios.

cs.AI↗

Router-Tuning: A Simple and Effective Approach for Enabling Dynamic-Depth in Transformers

Traditional transformer models often allocate a fixed amount of computational resources to every input token, leading to inefficient and unnecessary computation. To address this, the Mixture of Depths (MoD) was introduced to dynamically adjust the computational depth by skipping less important layers. Despite its promise, current MoD approaches remain under-explored and face two main challenges: (1) high training costs due to the need to train the entire model along with the routers that determine which layers to skip, and (2) the risk of performance degradation when important layers are bypassed. In response to the first issue, we propose Router-Tuning, a method that fine-tunes only the router on a small dataset, drastically reducing the computational overhead associated with full model training. For the second challenge, we propose MindSkip, which deploys Attention with Dynamic Depths. This method preserves the model's performance while significantly enhancing computational and memory efficiency. Extensive experiments demonstrate that our approach delivers competitive results while dramatically improving the computation efficiency, e.g., 21\% speedup and only a 0.2\% performance drop. The code is released at https://github.com/CASE-Lab-UMD/Router-Tuning.

cs.CL↗

Climate in a Bottle: Towards a Generative Foundation Model for the Kilometer-Scale Global Atmosphere

Climate modeling is reaching unprecedented resolution, producing petabytes of data. AI climate model emulators offer a path to computationally cheap analysis, enabling new scientific insight and scenario planning. Recent advances show promise in faithfully emulating climate data. However, prevailing auto-regressive paradigms are difficult to train on climate time horizons due to drifts, instabilities, and component-coupling challenges. They are hard to scale to high resolution and require sifting through troves of output to identify rare extremes of interest. We present Climate in a Bottle (cBottle), a generative diffusion-based framework emulating global 5 km climate simulations and reanalysis on the HEALPix grid. cBottle samples directly from the full distribution of atmospheric states, avoiding auto-regressive rollout, and is the first to reach this 12.5M-pixel global resolution. It consists of two stages: a coarse-resolution generator conditioned on sea surface temperatures and solar position, followed by a patch-based 16x super-resolution stage. cBottle passes a battery of tests, including diurnal-to-seasonal variability, large-scale modes of variability, tropical cyclone statistics, and trends of climate change and weather extremes. It is a step toward a foundation model: bridging data modalities (reanalysis and simulation), enabling zero-shot bias correction, downscaling, and data infilling. It also enables new interactivity via guided diffusion. For example, we train a tropical cyclone (TC) classifier alongside the generator, guide towards TC states, and obtain physically credible samples. This opens the door to guidance methods for a wide array of user queries and new ways of interacting with climate data.

physics.ao-ph↗

Scaling Synthetic Data Creation with 1,000,000,000 Personas

We propose a novel persona-driven data synthesis methodology that leverages various perspectives within a large language model (LLM) to create diverse synthetic data. To fully exploit this methodology at scale, we introduce Persona Hub -- a collection of 1 billion diverse personas automatically curated from web data. These 1 billion personas (~13% of the world's total population), acting as distributed carriers of world knowledge, can tap into almost every perspective encapsulated within the LLM, thereby facilitating the creation of diverse synthetic data at scale for various scenarios. By showcasing Persona Hub's use cases in synthesizing high-quality mathematical and logical reasoning problems, instructions (i.e., user prompts), knowledge-rich texts, game NPCs and tools (functions) at scale, we demonstrate persona-driven data synthesis is versatile, scalable, flexible, and easy to use, potentially driving a paradigm shift in synthetic data creation and applications in practice, which may have a profound impact on LLM research and development.

cs.CL↗

Improving LLM General Preference Alignment via Optimistic Online Mirror Descent

Reinforcement learning from human feedback (RLHF) has demonstrated remarkable effectiveness in aligning large language models (LLMs) with human preferences. Many existing alignment approaches rely on the Bradley-Terry (BT) model assumption, which assumes the existence of a ground-truth reward for each prompt-response pair. However, this assumption can be overly restrictive when modeling complex human preferences. In this paper, we drop the BT model assumption and study LLM alignment under general preferences, formulated as a two-player game. Drawing on theoretical insights from learning in games, we integrate optimistic online mirror descent into our alignment framework to approximate the Nash policy. Theoretically, we demonstrate that our approach achieves an $O(T^{-1})$ bound on the duality gap, improving upon the previous $O(T^{-1/2})$ result. More importantly, we implement our method and show through experiments that it outperforms state-of-the-art RLHF algorithms across multiple representative benchmarks.

cs.LG↗

OpenCharacter: Training Customizable Role-Playing LLMs with Large-Scale Synthetic Personas

Customizable role-playing in large language models (LLMs), also known as character generalization, is gaining increasing attention for its versatility and cost-efficiency in developing and deploying role-playing dialogue agents. This study explores a large-scale data synthesis approach to equip LLMs with character generalization capabilities. We begin by synthesizing large-scale character profiles using personas from Persona Hub and then explore two strategies: response rewriting and response generation, to create character-aligned instructional responses. To validate the effectiveness of our synthetic instruction tuning data for character generalization, we perform supervised fine-tuning (SFT) using the LLaMA-3 8B model. Our best-performing model strengthens the original LLaMA-3 8B Instruct model and achieves performance comparable to GPT-4o models on role-playing dialogue. We release our synthetic characters and instruction-tuning dialogues to support public research.

cs.CL↗

xRAG: Extreme Context Compression for Retrieval-augmented Generation with One Token

This paper introduces xRAG, an innovative context compression method tailored for retrieval-augmented generation. xRAG reinterprets document embeddings in dense retrieval--traditionally used solely for retrieval--as features from the retrieval modality. By employing a modality fusion methodology, xRAG seamlessly integrates these embeddings into the language model representation space, effectively eliminating the need for their textual counterparts and achieving an extreme compression rate. In xRAG, the only trainable component is the modality bridge, while both the retriever and the language model remain frozen. This design choice allows for the reuse of offline-constructed document embeddings and preserves the plug-and-play nature of retrieval augmentation. Experimental results demonstrate that xRAG achieves an average improvement of over 10% across six knowledge-intensive tasks, adaptable to various language model backbones, ranging from a dense 7B model to an 8x7B Mixture of Experts configuration. xRAG not only significantly outperforms previous context compression methods but also matches the performance of uncompressed models on several datasets, while reducing overall FLOPs by a factor of 3.53. Our work pioneers new directions in retrieval-augmented generation from the perspective of multimodality fusion, and we hope it lays the foundation for future efficient and scalable retrieval-augmented systems

cs.CL↗