arXiv subjects

Yifan Yao

Publications and source records attributed to Yifan Yao.

At least 19 recordsLinked to original sources

arXiv2026

GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions

Coding agents can modify and test code across large software projects. Game development is a domain where agents must implement gameplay rules. A game can end in a valid state even after violating its rules during the run. Current game-development benchmarks replay fixed examples, score videos, or ask another model to judge the result. However, no existing benchmark checks game rules throughout execution across varied evaluator-selected scenarios while ensuring exactly reproducible verdicts. We introduce GameLogicBench, a benchmark of 72 gameplay-logic tasks in Godot projects. An automated evaluator checks each game's rules at every simulation tick. Across 403 hand-designed scenarios, seeded parameter variations produce 1,451 test cases. To ensure that the evaluator measures behavior rather than implementation choice, it must accept different correct implementations for each task while rejecting mutants, implementations with one required capability removed. The tasks span isolated mechanics, multi-system interactions, and repository-scale features. Across 20 combinations of language models and scaffolds, the best observed run solves 52.78% of tasks. Under Claude Code, all twelve models solve fewer tasks as task scope expands from isolated mechanics, through interacting systems, to repository-scale features. Agents inspect code more often and make more tool calls on repository-scale tasks than on isolated-mechanic tasks. Most unsuccessful submissions are runnable, but implement some required game behavior incorrectly. We compared versions of our benchmark evaluator built with and without validation using mutants. Without this validation, incorrect agent submissions passed. A separate analysis finds agents copying code from public repositories when network access is open. Reliable evaluation thus depends both on what the tests reject and on what external code agents can access.

cs.SE↗

arXiv2026

HyperParallel-FSDP: Topology-Aware Fully Sharded Training with Layout-Driven Muon on Ascend SuperPods

Declarative SPMD programming uses tensor sharding descriptions to drive distributed execution, separating parallelization from model code. However, the evaluated PyTorch DTensor stack dispatches every operator below autograd, incurring repeated dispatch and metadata costs, while lacking an inexpensive end-to-end validation path. Existing FSDP and distributed Muon implementations also mismatch two-tier supernode topologies: FSDP relies on explicit parameter packing and unpacking, and Muon's whole-matrix orthogonalization conflicts with parameter sharding. We observe that distributed tensors need only express sharding semantics at the tensor API boundary above autograd, allowing differentiation and kernels to operate on plain tensors. Based on this insight, we present HyperParallel-FSDP, featuring: (1) dual-mode DTensor execution, using one sharding plan for both a production mode with one-time layout resolution and no steady-state dispatch overhead, and a validation mode with end-to-end metadata propagation, fail-fast checks, and gradient-equivalence testing; (2) topology-aware FSDP, with zero-copy intra-supernode collectives, fused inter-supernode reduction, and a cross-layer backward pipeline that avoids waits on slow links; and (3) layout-driven distributed Muon, with sharding-derived communication groups, deduplicated orthogonalization, and shape-fused Newton-Schulz iterations. On Atlas 900 A3 SuperPoD, HyperParallel-FSDP scales from 16 dies to 384 cards (768 ranks), sustaining 421k tokens/s for a 505B-parameter MoE while FSDP communication uses 2.9% of step time. It reduces mean step time by 29.7% versus PyTorch FSDP2 and 25.5% versus Megatron DDP, with Pearson correlation above 0.999997 over 1,000 steps. Distributed Muon improves profiler step time by 5.4-16.0% over competing systems. Source code is available at https://atomgit.com/mindspore/hyper-parallel.

cs.DC↗

arXiv2026

CLI-Universe: Towards Verifiable Task Synthesis Engine for Terminal Agents

While recent LLM-based terminal agents have demonstrated promising capabilities, the scarcity of high-quality, executable training data remains a critical bottleneck. Existing synthesis pipelines typically scale by retrofitting surface-level artifacts into tasks, frequently yielding ambiguous instructions, shallow execution paths, and brittle tests that provide weak learning signals. To overcome this, we introduce CLI-Universe, a principled synthesis engine that constructs terminal-agent tasks. CLI-Universe generates candidate tasks by sampling combinations across a multi-dimensional capability taxonomy (domain, skill type, capability, and engineering pillar), then grounds each candidate through evidence-guided deep research over real-world technical materials. To ensure rigorous supervision, validated blueprints are instantiated into Dockerized environments and subjected to a multi-stage executable verification pipeline featuring rubric-gated test construction, hint-conditional filtering, and strict fail-to-pass checking. Across the full pipeline, from candidate generation to verification, approximately two-thirds of candidates are discarded, retaining only those that are genuine, verifiable, and non-trivially challenging. To validate our framework, we instantiate a highly distilled dataset of 6,000 trajectories called CLI-Universe-6K. Remarkably, fine-tuning Qwen3-32B on CLI-Universe-6K achieves 33.4% on Terminal-Bench 2.0. This sets a new state-of-the-art for models trained on open-source data at or below 32B parameters, and outperforms several models an order of magnitude larger, demonstrating the profound data efficiency of structured, high-fidelity synthesis.

cs.AI↗

arXiv2026

Knowledge Index of Noah's Ark

Knowledge benchmarks for LLMs face three issues: scaling-driven designs that do not operationalize disciplinary representativeness; flat-payment annotation that permits lazy consensus; and unaudited ranking instability under bounded test budgets. We introduce KINA, an 899-item benchmark across 261 fine-grained disciplines, with two formal results. First, we cast representativeness as a coverage-style objective over expert-elicited anchors and operationalize disciplinary representativeness through a proxy, yielding a (1-1/e) greedy approximation (Proposition 1); the guarantee applies to the proxy, not to population representativeness. Second, we prove a bonus-on-bar tournament weakly FOSD-dominates flat payment in released-review quality, with incentive-compatibility threshold B > Delta C / Delta p_min (Theorem 1). Evaluating 42 models from 13 labs, the top model, Gemini-3.1-Pro-Preview, reaches 53.17%, followed by Claude-Opus-4.6 at 49.92% and GPT-5.4 at 48.55%, leaving substantial headroom below saturation. The full leaderboard shows a tiered structure rather than a smooth total order: a small frontier tier lies above 48%, a dense strong-model tier spans roughly 38-45%, and low-performing models remain only modestly above the 10% chance baseline. Tool augmentation adds up to 5.17 points across the five tool-use evaluations, with gains varying substantially across models. We report bootstrap ranking-stability statistics to make bounded-budget variance explicit and to discourage over-interpretation of adjacent ranks.

cs.AI↗

arXiv2026

Solvita: Enhancing Large Language Models for Competitive Programming via Agentic Evolution

Large language models (LLMs) still struggle with the rigorous reasoning demands of hard competitive programming. While recent multi-agent frameworks attempt to bridge this reliability gap, they remain fundamentally stateless: they rely on static retrieval and discard the valuable problem-solving and debugging experience gained from previous tasks. To address this, we present Solvita, an agentic evolution framework that enables continuous learning without requiring weight updates to the underlying LLM. Solvita reorganizes problem-solving into a closed-loop system of strategy selection, program synthesis, certified supervision, and targeted hacking, executed by four specialized agents: Planner, Solver, Oracle, and Hacker. Crucially, each agent is paired with a trainable, graph-structured knowledge network. As the system operates, outcome signals, such as pass/fail verdicts, test certification quality, and adversarial vulnerabilities discovered by the Hacker, are recast as reinforcement learning updates to these network weights. This allows the agents to dynamically route future queries based on past successes and failures, effectively accumulating transferable reasoning experience over time. Evaluated across CodeContests, APPS, AetherCode, and live Codeforces rounds, Solvita establishes a new state-of-the-art among code-generation agents, outperforming existing multi-agent pipelines and nearly doubling the accuracy of single-pass baselines.

cs.AI↗

arXiv2026

Ultrafast electron dynamics of electron-irradiated graphene

Electron irradiation is essential for materials characterization and modification, though the fundamental interactions between incident electrons and host materials remain under investigation. Here, we employ first-principles simulations to study electron dynamics under external electron irradiation. We quantify differences in key observables, including kinetic energy loss, secondary electron emission, and backscattered electrons, between classical and quantum mechanical descriptions of the incident electron. Around 400 eV incident energy, we identify significant differences in backscattered electron yields between classical point-charge and quantum wave-packet descriptions, whereas the quantum-mechanical effects diminish at incident energies above 600 eV. These differences highlight the critical importance of quantum effects in electron irradiation phenomena that occur in a specific energy range of the incident electron. Our results provide clear guidance for selecting appropriate incident, electron descriptions based on kinetic-energy regimes, identify a targeted experimental window for isolating quantum-only backscattering, and enable the rational design of 2D materials for nanofabrication and high-resolution electron-beam technologies.

cond-mat.mtrl-sci↗

arXiv2026

WebCompass: Towards Multimodal Web Coding Evaluation for Code Language Models

Large language models are rapidly evolving into interactive coding agents capable of end-to-end web coding, yet existing benchmarks evaluate only narrow slices of this capability, typically text-conditioned generation with static-correctness metrics, leaving visual fidelity, interaction quality, and codebase-level reasoning largely unmeasured. We introduce WebCompass, a multimodal benchmark that provides unified lifecycle evaluation of web engineering capability. Recognizing that real-world web coding is an iterative cycle of generation, editing, and repair, WebCompass spans three input modalities (text, image, video) and three task types (generation, editing, repair), yielding seven task categories that mirror professional workflows. Through a multi-stage, human-in-the-loop pipeline, we curate instances covering 15 generation domains, 16 editing operation types, and 11 repair defect types, each annotated at Easy/Medium/Hard levels. For evaluation, we adopt a checklist-guided LLM-as-a-Judge protocol for editing and repair, and propose a novel Agent-as-a-Judge paradigm for generation that autonomously executes generated websites in a real browser, explores interactive behaviors via the Model Context Protocol (MCP), and iteratively synthesizes targeted test cases, closely approximating human acceptance testing. We evaluate representative closed-source and open-source models and observe that: (1) closed-source models remain substantially stronger and more balanced; (2) editing and repair exhibit distinct difficulty profiles, with repair preserving interactivity better but remaining execution-challenging; (3) aesthetics is the most persistent bottleneck, especially for open-source models; and (4) framework choice materially affects outcomes, with Vue consistently challenging while React and Vanilla/HTML perform more strongly depending on task type.

cs.SE↗

arXiv2026

CodeTracer: Towards Traceable Agent States

Code agents are advancing rapidly, but debugging them is becoming increasingly difficult. As frameworks orchestrate parallel tool calls and multi-stage workflows over complex tasks, making the agent's state transitions and error propagation hard to observe. In these runs, an early misstep can trap the agent in unproductive loops or even cascade into fundamental errors, forming hidden error chains that make it hard to tell when the agent goes off track and why. Existing agent tracing analyses either focus on simple interaction or rely on small-scale manual inspection, which limits their scalability and usefulness for real coding workflows. We present CodeTracer, a tracing architecture that parses heterogeneous run artifacts through evolving extractors, reconstructs the full state transition history as a hierarchical trace tree with persistent memory, and performs failure onset localization to pinpoint the failure origin and its downstream chain. To enable systematic evaluation, we construct CodeTraceBench from a large collection of executed trajectories generated by four widely used code agent frameworks on diverse code tasks (e.g., bug fixing, refactoring, and terminal interaction), with supervision at both the stage and step levels for failure localization. Experiments show that CodeTracer substantially outperforms direct prompting and lightweight baselines, and that replaying its diagnostic signals consistently recovers originally failed runs under matched budgets. Our code and data are publicly available.

cs.SE↗

arXiv2025

The Imitation Game: Using Large Language Models as Chatbots to Combat Chat-Based Cybercrimes

Chat-based cybercrime has emerged as a pervasive threat, with attackers leveraging real-time messaging platforms to conduct scams that rely on trust-building, deception, and psychological manipulation. Traditional defense mechanisms, which operate on static rules or shallow content filters, struggle to identify these conversational threats, especially when attackers use multimedia obfuscation and context-aware dialogue. In this work, we ask a provocative question inspired by the classic Imitation Game: Can machines convincingly pose as human victims to turn deception against cybercriminals? We present LURE (LLM-based User Response Engagement), the first system to deploy Large Language Models (LLMs) as active agents, not as passive classifiers, embedded within adversarial chat environments. LURE combines automated discovery, adversarial interaction, and OCR-based analysis of image-embedded payment data. Applied to the setting of illicit video chat scams on Telegram, our system engaged 53 actors across 98 groups. In over 56 percent of interactions, the LLM maintained multi-round conversations without being noticed as a bot, effectively "winning" the imitation game. Our findings reveal key behavioral patterns in scam operations, such as payment flows, upselling strategies, and platform migration tactics.

cs.CR↗

arXiv2025

SWE-Compass: Towards Unified Evaluation of Agentic Coding Abilities for Large Language Models

Evaluating large language models (LLMs) for software engineering has been limited by narrow task coverage, language bias, and insufficient alignment with real-world developer workflows. Existing benchmarks often focus on algorithmic problems or Python-centric bug fixing, leaving critical dimensions of software engineering underexplored. To address these gaps, we introduce SWE-Compass1, a comprehensive benchmark that unifies heterogeneous code-related evaluations into a structured and production-aligned framework. SWE-Compass spans 8 task types, 8 programming scenarios, and 10 programming languages, with 2000 high-quality instances curated from authentic GitHub pull requests and refined through systematic filtering and validation. We benchmark ten state-of-the-art LLMs under two agentic frameworks, SWE-Agent and Claude Code, revealing a clear hierarchy of difficulty across task types, languages, and scenarios. Moreover, by aligning evaluation with real-world developer practices, SWE-Compass provides a rigorous and reproducible foundation for diagnosing and advancing agentic coding capabilities in large language models.

cs.SE↗

arXiv2025

KAT-Coder Technical Report

Recent advances in large language models (LLMs) have enabled progress in agentic coding, where models autonomously reason, plan, and act within interactive software development workflows. However, bridging the gap between static text-based training and dynamic real-world agentic execution remains a core challenge. In this technical report, we present KAT-Coder, a large-scale agentic code model trained through a multi-stage curriculum encompassing Mid-Term Training, Supervised Fine-Tuning (SFT), Reinforcement Fine-Tuning (RFT), and Reinforcement-to-Deployment Adaptation. The Mid-Term stage enhances reasoning, planning, and reflection capabilities through a corpus of real software engineering data and synthetic agentic interactions. The SFT stage constructs a million-sample dataset balancing twenty programming languages, ten development contexts, and ten task archetypes. The RFT stage introduces a novel multi-ground-truth reward formulation for stable and sample-efficient policy optimization. Finally, the Reinforcement-to-Deployment phase adapts the model to production-grade IDE environments using Error-Masked SFT and Tree-Structured Trajectory Training. In summary, these stages enable KAT-Coder to achieve robust tool-use reliability, instruction alignment, and long-context reasoning, forming a deployable foundation for real-world intelligent coding agents. Our KAT series 32B model, KAT-Dev, has been open-sourced on https://huggingface.co/Kwaipilot/KAT-Dev.

cs.CL↗

arXiv2025

3rd Place Solution to Large-scale Fine-grained Food Recognition

Food analysis is becoming a hot topic in health area, in which fine-grained food recognition task plays an important role. In this paper, we describe the details of our solution to the LargeFineFoodAI-ICCV Workshop-Recognition challenge held on Kaggle. We find a proper combination of Arcface loss[1] and Circle loss[9] can bring improvement to the performance. With Arcface and the combined loss, model was trained with carefully tuned configurations and ensembled to get the final results. Our solution won the 3rd place in the competition.

cs.CV↗

arXiv2025

Falcon: A Comprehensive Chinese Text-to-SQL Benchmark for Enterprise-Grade Evaluation

We introduce Falcon, a cross-domain Chinese text-to-SQL benchmark grounded in an enterprise-compatible dialect (MaxCompute/Hive). It contains 600 Chinese questions over 28 databases; 77% require multi-table reasoning and over half touch more than four tables. Each example is annotated along SQL-computation features and Chinese semantics. For evaluation, we release a robust execution comparator and an automated evaluation pipeline, under which all current state-of-the-art large-scale models (including Deepseek) achieve accuracies of at most 50%. Major errors originate from two sources: (1) schema linking in large enterprise landscapes - hundreds of tables, denormalized fields, ambiguous column names, implicit foreign-key relations and domain-specific synonyms that make correct join/column selection difficult; and (2) mapping concise, colloquial Chinese into the exact operators and predicates required for analytics - e.g., choosing the correct aggregation and group-by keys, expressing time windows and granularities, applying unit conversions, handling NULLs and data-quality rules, and formulating nested or windowed subqueries. Falcon therefore targets Chinese-specific semantics and enterprise dialects (abbreviations, business jargon, fuzzy entity references) and provides a reproducible middle ground before full production deployment by using realistic enterprise schemas, query templates, an execution comparator, and an automated evaluation pipeline for end-to-end validation.

cs.CL↗

arXiv2025

HiPO: Hybrid Policy Optimization for Dynamic Reasoning in LLMs

Large Language Models (LLMs) increasingly rely on Chain-of-Thought (CoT) reasoning to improve accuracy on complex tasks. However, always generating lengthy reasoning traces is inefficient, leading to excessive token usage and higher inference costs. This paper introduces the Hybrid Policy Optimization (i.e., HiPO), a framework for adaptive reasoning control that enables LLMs to selectively decide when to engage in detailed reasoning (Think-on) and when to respond directly (Think-off). Specifically, HiPO combines a hybrid data pipelineproviding paired Think-on and Think-off responseswith a hybrid reinforcement learning reward system that balances accuracy and efficiency while avoiding over-reliance on detailed reasoning. Experiments across mathematics and coding benchmarks demonstrate that HiPO can substantially reduce token length while maintaining or improving accuracy. Finally, we hope HiPO a can be a principled approach for efficient adaptive reasoning, advancing the deployment of reasoning-oriented LLMs in real-world, resource-sensitive settings.

cs.CL↗

arXiv2025

IF-VidCap: Can Video Caption Models Follow Instructions?

Although Multimodal Large Language Models (MLLMs) have demonstrated proficiency in video captioning, practical applications require captions that follow specific user instructions rather than generating exhaustive, unconstrained descriptions. Current benchmarks, however, primarily assess descriptive comprehensiveness while largely overlooking instruction-following capabilities. To address this gap, we introduce IF-VidCap, a new benchmark for evaluating controllable video captioning, which contains 1,400 high-quality samples. Distinct from existing video captioning or general instruction-following benchmarks, IF-VidCap incorporates a systematic framework that assesses captions on two dimensions: format correctness and content correctness. Our comprehensive evaluation of over 20 prominent models reveals a nuanced landscape: despite the continued dominance of proprietary models, the performance gap is closing, with top-tier open-source solutions now achieving near-parity. Furthermore, we find that models specialized for dense captioning underperform general-purpose MLLMs on complex instructions, indicating that future work should simultaneously advance both descriptive richness and instruction-following fidelity.

cs.CV↗

arXiv2025

Power System Voltage Stability Boundary: Computational Results and Applications

The objective of this paper is to report some computational results for the theory of DAE stability boundary, with the aim of advancing applications in power system voltage stability studies. Firstly, a new regularization transformation for standard differential-algebraic equations (DAEs) is proposed. Then the existence of anchor points on voltage stability boundary is examined, and an optimization method for computing the controlling pseudo-saddle is suggested. Subsequently, a local representation of the stable manifold of the pseudo-saddle on the stability boundary is presented, and a voltage stability margin expression is obtained. Finally, the proposed results are verified using several examples, demonstrating the accuracy and effectiveness of the suggested methods.

eess.SY↗

arXiv2025

CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization

Translating natural language mathematical statements into formal, executable code is a fundamental challenge in automated theorem proving. While prior work has focused on generation and compilation success, little attention has been paid to the critic phase-the evaluation of whether generated formalizations truly capture the semantic intent of the original problem. In this paper, we introduce CriticLean, a novel critic-guided reinforcement learning framework that elevates the role of the critic from a passive validator to an active learning component. Specifically, first, we propose the CriticLeanGPT, trained via supervised fine-tuning and reinforcement learning, to rigorously assess the semantic fidelity of Lean 4 formalizations. Then, we introduce CriticLeanBench, a benchmark designed to measure models' ability to distinguish semantically correct from incorrect formalizations, and demonstrate that our trained CriticLeanGPT models can significantly outperform strong open- and closed-source baselines. Building on the CriticLean framework, we construct FineLeanCorpus, a dataset comprising over 285K problems that exhibits rich domain diversity, broad difficulty coverage, and high correctness based on human evaluation. Overall, our findings highlight that optimizing the critic phase is essential for producing reliable formalizations, and we hope our CriticLean will provide valuable insights for future advances in formal mathematical reasoning.

cs.CL↗

arXiv2025

KORGym: A Dynamic Game Platform for LLM Reasoning Evaluation

Recent advancements in large language models (LLMs) underscore the need for more comprehensive evaluation methods to accurately assess their reasoning capabilities. Existing benchmarks are often domain-specific and thus cannot fully capture an LLM's general reasoning potential. To address this limitation, we introduce the Knowledge Orthogonal Reasoning Gymnasium (KORGym), a dynamic evaluation platform inspired by KOR-Bench and Gymnasium. KORGym offers over fifty games in either textual or visual formats and supports interactive, multi-turn assessments with reinforcement learning scenarios. Using KORGym, we conduct extensive experiments on 19 LLMs and 8 VLMs, revealing consistent reasoning patterns within model families and demonstrating the superior performance of closed-source models. Further analysis examines the effects of modality, reasoning strategies, reinforcement learning techniques, and response length on model performance. We expect KORGym to become a valuable resource for advancing LLM reasoning research and developing evaluation methodologies suited to complex, interactive environments.

cs.CL↗