Searcharxiv⌕ Search

arXiv subjects

Daxin Jiang

Publications and source records attributed to Daxin Jiang.

At least 19 recordsLinked to original sources

Metal Sandbox: Isolated Execution and Command Scheduling for GPU Coding Agents on Apple Silicon

On-device inference requires kernels that make effective use of local accelerators. Efficient training and evaluation of models that generate Apple Metal kernels benefit from native macOS tools and concurrent agent sessions. These sessions alternate between model requests, development commands, and GPU execution, leaving GPU capacity idle for much of a rollout. We present Metal Sandbox, infrastructure that combines restricted per-instance file access, native command execution, and command-level scheduling across interconnected Macs. Each agent retains its workspace and private development environment. Parallel commands can run concurrently, while serial commands share one execution slot per node; eligible waiting commands can migrate to available Macs and return results to the original agent. Integration with Harbor supports task execution, automatic verification, and rollout collection through a common workflow. Controlled experiments on four Mac mini M4 nodes achieve $1.68\times$--$2.05\times$ gains in command throughput under concentrated demand. Sixteen static agent traces give an estimated $2.14\times$ sustained-throughput gain for four sessions per node versus one, assuming continuously replenished sessions and unchanged mean durations. Replaying these measured traces under dynamic scheduling reduces mean tool waiting by 18.8% and batch elapsed time by 22.3%, assuming unchanged phase durations. Metal Sandbox and its Harbor integration are open-source on GitHub.

cs.LG↗

LiveSearch: Questions as Executable Verifier Code for a Live Benchmark of Professional-Source Financial Search Agents

In finance, the answer that matters is usually the current one: prices, rates, and disclosures hold only as of a given day, so retrieving up-to-date facts is a core capability of a financial agent. Yet most financial benchmarks overlook timeliness. They fix each answer at annotation time, so the gold goes stale as markets move and, once leaked into training data, rewards recall rather than search and analysis. We introduce LiveSearch, a live benchmark that stores each question as executable verifier code: a typed program of calls to a professional financial data service, followed by a declarative extractor. The question is frozen, while its gold answer is determined by executing the program against professional financial data at evaluation time, and a daily job re-executes it with no human or language model in the loop. A memorized answer is thus likely outdated, and a model must reach the current one through tools and analysis. Construction is equally automatic: a language model only verbalizes programs enumerated from real tool responses and never chooses an answer. Under general web search, six contemporary tool-calling models reach 70.9% to 76.7% answer accuracy. Over three consecutive days, 70% of served questions change their gold value, and even under tolerance-aware matching, a literally reused earlier snapshot fails on a majority of questions within a single day. Per-question tracking uncovers dynamic shifts in correctness masked by aggregate accuracy.

cs.AI↗

AutoVend: Benchmarking LLM Agents on Long-Horizon Autonomous Vending Operations

Large language model (LLM) agents are extending agent-mediated commerce on the Web from shopping for consumers to operating businesses for sellers. In online-to-offline (O2O) commerce, decisions are issued through Web-based platforms but realized through costly physical execution. Evaluating such agents requires going beyond individual transactions to ask whether decisions made online turn into sustained economic returns. We introduce AutoVend, a benchmark in which one LLM agent manages 20 vending machines across five site types through a Web-based operating interface for 180 simulated days. Its data-grounded simulator combines a shared warehouse, procurement lead times, paid replenishment, inventory expiration, and recurring operating costs. We evaluate five LLM agents against a rule-based inventory-control baseline and diagnose their site coordination, execution, and performance over time. Four of five evaluated LLM agents fall short of this baseline, and returns per dollar of inference cost rank models differently from returns alone. Operational analyses show that gains in individual metrics do not consistently yield stronger economic growth, as inventory and execution costs erode these gains. Trajectory analysis further reveals difficulties in grounding feasible actions, managing inventory commitments, and converting feedback into effective policy revisions.

cs.AI↗

P$^2$-MOPD: Accelerating Multi-Teacher On-Policy Distillation via Parameter Priors

Multi-teacher on-policy distillation consolidates expert capabilities by supervising student-generated trajectories, whereas task arithmetic offers a training-free alternative through linear parameter combinations. We view the linear combination of student and expert parameters as a single differentiable model and propose P$^2$-MOPD, which jointly optimizes the mixing coefficients and the student weights under the same on-policy distillation objective. Unlike task arithmetic, which stays within the span of the frozen student and expert parameters for any choice of merging coefficients, P$^2$-MOPD keeps the student trainable, so the experts act as parameter priors without confining it to this span. Across four domains at 1.7B and 4B scales, P$^2$-MOPD achieves faster and more balanced recovery than MOPD from the same initialization, surpassing its final average recovery in half the training steps. At the end of training, our 1.7B model reaches 94.7% average recovery, versus 80.9% for MOPD.

cs.LG↗

CertClaim: Towards Certified Model-Claim Verification for Coding-Agent Services

Coding-agent services are priced by the model they claim, making the verification of such model claims crucial for their users. In this paper, we reveal that existing model auditing methods rely on a latent assumption that the deployment is static, i.e., the claimed model is always served in one reference configuration. This assumption does not hold for coding-agent services, whose providers may legitimately change the sampling preset, the reasoning mode or the system prompt without changing the model. As a result, existing methods either falsely accuse honest providers or miss cheap derivatives of the claimed model. To address this problem, we propose a certified model-claim verification method (dubbed CertClaim). Instead of testing the service against one reference deployment, we make the authorized deployments part of the claim and test whether some authorized deployment explains all the evidence of the service, by calibrating multiple evidence channels in every authorized deployment and fusing them within each deployment. We provide theoretical analyses of its false-accusation rate and detection condition. Extensive experiments on eight claims, two agent harnesses and three task suites verify the effectiveness of CertClaim, which is valid in every setting while separating substitutes far better than existing methods, as well as its resistance to potential adaptive attacks.

cs.CR↗

RubricAtlas: Mining Type-Level Rubrics from Expert Reports to Evaluate Deep Research Agents

Deep research agents conduct multi-step Web searches and synthesise retrieved information into analytical reports. Evaluating these outputs requires checking which analyses they deliver, beyond the presence of relevant facts. Query-specific binary rubrics make individual requirements explicit, but task-level checklists may leave the overall analytical structure underexamined. We introduce RubricAtlas, which derives type-level analytical requirements from expert reports on different objects within the same research type and organises detailed checks around recurring analytical questions. The requirements give related tasks a shared standard for what to analyse, without prescribing their answers. Our framework links binary checkpoints to report passages and evaluates completeness separately from argument validity. Built from 3,079 Chinese financial research reports, RubricAtlas covers 99 research types. Reference-report selection experiments show that individual reports cover only about half of their type's analytical questions, and that selecting requirements from different reports changes evaluation scores and agent rankings. Across ten agents, two judge pipelines achieve checkpoint agreement of $κ$ = 0.71 and agent-ranking agreement of Kendall $τ$ = 0.87. Controlled deletions reduce scores on the affected analyses, while padding produces no detectable increase. These findings support type-level rubrics for comparing the analytical completeness of Web deep research outputs, with individual decisions traceable to report evidence.

cs.AI↗

$Φ$-Bench: Can Large Language Models Engineer the Infrastructure That Powers Them?

Large language models (LLMs) now show strong reasoning and coding abilities, raising the prospect that they could assist in developing and optimizing the very infrastructure that powers them. Existing benchmarks, however, mainly focus on isolated kernels, predefined operators, or pre-specified optimization targets, and cannot evaluate open-ended, long-horizon LLM infrastructure engineering. To address this gap, we present $Φ$-Bench, a benchmark of 85 tasks grounded in real LLM training and inference codebases. The tasks cover nine categories of LLM infrastructure and come in three formats, ranging from kernel function completion to long-horizon implementation and repository-wide end-to-end optimization. They are built by a construction pipeline that turns real pull requests, issues, and repository code into executable tasks with validated tests, and every reward is gated by correctness and anti-hacking checks. Across eight frontier models, the strongest model, Claude Opus 5, scores only 36.53%, and every model performs worse on long-horizon, repository-level implementation than on kernel-level tasks. Trajectory analysis further suggests that stronger models treat optimization as controlled experimentation, validating ideas with cheap local runs and ruling out confounders before drawing conclusions. We will open-source $Φ$-Bench to facilitate future research on LLM agents' ability to develop and optimize the LLM infrastructure that powers themselves.

cs.CL↗

StepAudio 3 Realtime Technical Report

Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally. Crucially, we resolve the tension between deep deliberation and latency via Think-While-Speaking, executing private reasoning in parallel with spoken delivery. In reasoning mode, StepAudio 3 reaches a 73.0 macro average on StepAudioChat. With Think-While-Speaking, it achieves dialogue and reasoning performance comparable to dedicated reasoning models while speaking in real time. Furthermore, an integrated Voice Agent handles asynchronous tool execution without disrupting the dialogue flow. StepAudio 3 Realtime achieves top-tier performance across key dimensions: an exceptional 90.6 on the MMSU benchmark, 98.9 Overall on the Artificial Analysis Full-Duplex Bench, and a 56.0% macro task-success rate on $τ$-Voice.

cs.SD↗

JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting

Speculative decoding (SD) accelerates autoregressive Large Language Models (LLMs) by drafting multiple tokens and verifying them in parallel, but it faces a scaling limitation: increasing the draft budget improves speed only when acceptance remains high and drafting overhead stays low. This ceiling has been difficult to break because prior head-based SD methods face a causality-efficiency dilemma. Autoregressive drafters produce path-conditioned candidates that are effective for tree speculative decoding with higher acceptance length, but their drafting cost grows with tree depth. Bidirectional block-diffusion drafters generate all positions in one pass, but their branch-agnostic marginals can form individually plausible yet mutually inconsistent trees, wasting budget and reducing acceptance. We propose JetSpec, a head-based SD framework that combines one-forward drafting efficiency with branch-wise causal conditioning. JetSpec trains a causal parallel draft head over fused hidden states from the frozen target model, producing candidate trees whose scores align with the target model's autoregressive factorization. This enables JetSpec to convert larger draft budgets into longer accepted prefixes and higher end-to-end speedup. Across math, coding, and chat benchmarks on dense and MoE Qwen3 models, JetSpec consistently outperforms bidirectional-head and tree-based SD baselines. On H100 GPUs, JetSpec achieves up to 9.64x speedup on MATH-500 and 4.58x on open-ended conversational workloads, with further latency gains demonstrated through vLLM integration under realistic serving loads. Our code and models are available at https://github.com/hao-ai-lab/JetSpec.

cs.CL↗

From Search to Research: Exploring Search Scaling in Autonomous Quantitative Factor Mining

Inference scaling has been shown to improve large language model (LLM) performance, and this principle naturally extends to autonomous LLM agents through increased search budgets, which we refer to as *search scaling*. Although prior work has characterized the mechanisms, scaling behavior, and performance limits of LLM inference scaling, much less is known about these questions in autonomous research. Therefore, we investigate how search scaling affects research performance and what mechanisms drive these gains using 50 quantitative factor-mining tasks grounded in financial research reports. Each task requires an agent to carry out an end-to-end research loop, from interpreting a hypothesis and implementing it in code to evaluating and iteratively refining the resulting factor. Across nine models, we examine how model capability, search depth, and search organization shape factor quality by tracing performance across varying budgets, transferring intermediate research states between models, and comparing different search strategies. We find that (1) initial performance is more strongly associated with model capability, while deeper search can narrow cross-model gaps; (2) model grafting shows that the early research state materially shapes final performance; and (3) parallel search outperforms sequential search under the same iteration budget, consistent with benefits from broader coverage of the search space. Further trajectory analysis shows that higher-performing models more effectively diagnose failures, revise search directions, and preserve the intended economic hypothesis when selecting candidates. These findings suggest that future progress in autonomous research will require stronger models together with adaptive policies for deploying test-time computation throughout the research process.

cs.AI↗

Probe to Act: Elevating Browser-Use Agent via Active Visual Probing

Browser-use agents require seamless alignment between structured web metadata and visual information, while preserving relevant context across long interactions. Existing interfaces often rely on either screenshot-level action prediction or static Set-of-Marks overlays, leaving the model to resolve dense DOM-pixel alignment before every operation. We introduce Probe to Act (P2A), an active probing framework for the browser-agent loop that moves this alignment into decision time. P2A addresses an asymmetric bridge between symbolic DOM hypotheses and screenshot layout by rendering on-demand symbolic DOM structure back into pixels. Before committing a state-changing browser operation, the agent can issue lightweight probes to translate DOM handles into pixel evidence, map screen regions back to DOM candidates, register visual-only targets, and commit verified notes. These interleaved processes naturally produce evidence-based memory: only probed, acted-on, or explicitly committed observations are kept across steps, preserving only decision-critical evidence in long-horizon contexts. P2A can be used as a prompting strategy for proprietary models under the standard DOM+SoM interface, and can be distilled into open-weight models through cold-start synthesis and self-bootstrapped SFT. Across three browser-use benchmarks, P2A shows clear gains on task success rate for both proprietary and fine-tuned models; on VisualWebArena, for example, it improves Gemini-3-Pro from 54.1% to 61.2% and Qwen3-VL-8B from 24.6% to 32.9%, while matching the costly full-observation history ($\sim$3$\times$) at only $\sim$1.2$\times$ the peak retained input context of action-only history.

cs.AI↗

GUI-GenBench: Evaluating Image Generation Models as Interactive GUI Environments

An image generation model used as a graphical user interface (GUI) environment must produce not only a plausible screen but also the correct response to an action. Existing visual-generation benchmarks do not directly assess this combination of functional correctness and temporal consistency in GUIs. We introduce GUI-GenBench, a benchmark of 700 curated samples across five task suites: single-step transitions, multi-step planning, fictional-app generation, real-app trajectories, and coordinate-based grounding. We also introduce GUI-Score, a VLM-based metric that assesses Goal Achievement, Interaction Logic, Content Consistency, UI Plausibility, and Visual Quality. Evaluating 12 image generation models on Chinese and English subsets, we find that strong single-step performance does not consistently extend to coherent multi-step trajectories or accurate grounding. The highest aggregate GUI-Scores are 69.62 on the Chinese subset and 63.16 on the English subset. Qualitative analysis further identifies failures in text rendering, icon interpretation, and coordinate-conditioned transitions. Together, these results distinguish visual plausibility from functional reliability and identify limitations that must be addressed before generated GUIs can serve as dependable interaction environments.

cs.AI↗

One Analyst Is Not Ground Truth: Grading Agent-Built Financial Models Against Observed Professional Practice

Language model agents can now construct complete financial models, but it remains unclear whether they have learned the financial judgment that gives those models meaning. Existing benchmarks often anchor correctness to an expert-authored solution (e.g., numerical targets or detailed rubrics) which introduces a implicit assumption: \emph{one expert solution can serve as ground truth.} This is appropriate when finance provides a unique answer, but not when judgment is required. Evidence from professional practice challenges this assumption: When financial models built by different analysts for the same company are graded against one another, the median score is only 0.33 under standard tolerances, \emph{revealing that single-reference grading confounds professional disagreement with error.} We therefore introduce GAUGE, a benchmark that decomposes financial-model evaluation into deterministic checks, rubric judgments and numerical rules. Built from 1,001 professional valuation spanning 922 companies and all 25 GICS industry groups, GAUGE checks mechanical properties deterministically and evaluates judgment-bearing quantities against ranges in professional practice. We then validate GAUGE as a measurement instrument by testing expertise ordering, held-out professional values, and robustness to LM judgements. Our findings show that current LM agents are far better at constructing financial models than at deriving the company-specific assumptions that drive valuation, a gap that even persists after fine-tuning.

cs.LG↗

KITE: KV-Invariant Transformer Expansion for Efficient Agentic LLM Scaling

Scaling a language model is not only a question of final quality: the architectural choice determines how much computation is spent during training, prompt processing, and autoregressive decoding to achieve certain model quality. An ideal model architecture should lower all above computation costs to facilitate scaling to a larger model, while ensure the larger model indeed outperforms smaller baselines. We introduce KV-Invariant Transformer Expansion (KITE), a scaling paradigm that achieves this goal. It trains the model from a smaller size to a larger size (i.e., saving training costs via upcycling), while places newly added parameters in regions that do not affect attention KV. Consequently, during inference, prefilling KV only relies on the smaller part of the model, so the inference costs are saved. As a concrete instantiation, we present Step Scale Transformer (SST), a two-tower decoder in which one tower produces KV and the other reads them. At comparable cumulative training compute, SST, a 67B MoE model with 2.15B active body parameters per decode token, achieves lower training loss than 47B and 63B MoE Transformers with 1.48B and 2.02B active body parameters, respectively, while reducing estimated inference cost by 6.7% and 31.6%.

cs.LG↗

onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction

We present onPanda, an interactive tool for efficiently annotating LLM alignment data and agent trajectories. onPanda adopts token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate token and either picks a substitute from the model's candidate tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate-correct-continue loop until a satisfactory response is obtained. This mechanism lets annotators precisely steer model outputs at low cost: a small controlled study suggests that onPanda reduces median annotation time by 52% over manual post-editing. Since the vast majority of tokens in the final response are generated by the model itself, the resulting data largely preserves the model's sampling distribution and is well suited for constructing on-policy SFT and preference data. Furthermore, the token-level corrections recorded during annotation provide fine-grained supervision with precise positions and naturally paired positive--negative samples. onPanda also connects to external tools and harnesses, enabling interactive trajectory annotation in realistic environments. In addition, we release Panda-CVL, a dataset annotated with onPanda, together with a benchmark for token-level correction.

cs.CL↗

StepAudio 3 Gen Technical Report

We introduce StepAudio 3 Gen, a general-purpose audio generation model that supports zero-shot text-to-speech (TTS), voice design, vocal generation, sound effects, music, vibe speech, and mixtures of multiple audio types within a unified framework. At its core, StepAudio 3 Gen is a discrete autoregressive generator that models audio directly over residual vector quantization (RVQ) tokens, departing from the diffusion Transformer-based continuous generation paradigm prevalent in recent general audio models. Its StepAudio Tokenizer represents general audio at 12.5 Hz in a shared $16 \times 2048$ residual code space, jointly quantizing semantic and waveform-level acoustic features so that each code layer preserves both types of information. For generation, the backbone predicts the first codebook along the time axis using autoregressive modeling, while a lightweight causal Transformer completes the remaining fifteen codebooks along the codebook axis. Our study further identifies three key design principles: (1) interference-aware progressive pretraining for acquiring audio capabilities while preserving the textual abilities of the large language model, (2) RVQ Adaptor for effectively incorporating multi-codebook acoustic representations, and (3) discrete autoregressive modeling over a shared representation across general audio domains. With progressive pretraining, multi-task instruction training, and supervised fine-tuning, StepAudio 3 Gen achieves state-of-the-art performance on both TTS and voice design, while retaining strong generation capabilities across speech, vocals, sound effects, and music. Audio samples are available at https://stepaudiollm.github.io/step-audio-3-gen/.

cs.SD↗

Alpha-R1: Alpha Screening with LLM Reasoning via Reinforcement Learning

Signal decay and regime shifts pose recurring challenges for data-driven investment strategies in non-stationary markets, where conventional time-series and machine learning approaches often struggle to generalize beyond historical correlations. While large language models (LLMs) offer strong capabilities for processing unstructured information, their potential to support quantitative factor screening through explicit economic reasoning remains underexplored. Existing factor-based methods typically reduce alphas to numerical time series, overlooking the semantic rationale that determines when a factor is economically relevant. We present Alpha-R1, an RL-aligned LLM framework for context-aware alpha screening. Its core mechanism, semantic gating, evaluates each candidate factor's semantic profile against a dynamically constructed market state description, selecting a sparse subset of factors whose economic rationale aligns with current market conditions. The selection model is trained via group relative policy optimization (GRPO), using realized portfolio returns as the primary reward signal. Under a 12-month out-of-sample evaluation, Alpha-R1 achieves annualized returns of 47.87% on S&P 500 and 40.57% on CSI 300 with Sharpe ratios of 1.62 and 2.23. These results, obtained under a bounded candidate-pool evaluation protocol, provide evidence for second-stage semantic factor reranking in non-stationary markets. The full implementation and resources are available at https://github.com/FinStep-AI/Alpha-R1.

q-fin.TR↗

Scheduling Mixed RL Rollouts Beyond Prefix Locality

Modern reinforcement learning (RL) post-training pipelines for large language models (LLMs) increasingly combine rollout workloads across multiple domains and feedback paradigms. Prefix-aware routing improves inference efficiency through cache reuse and load balancing, but it does not control how heterogeneous rollout sessions compete for KV-cache capacity. When reinforcement learning with verifiable rewards (RLVR), reinforcement learning from human feedback (RLHF), and agentic rollouts share an asynchronous inference service, their distinct sequence structures, interaction patterns, and KV-residency times create substantially different serving demands. Rollout scheduling must account for this heterogeneity without distorting the workload mixture specified by the trainer. We present MISA-T, a routing-layer admission policy for mixed rollout serving. MISA-T combines adaptive session admission, workload-aware KV-capacity allocation, and residency-time-aware KV accounting. In rollout-only ablations on Step3.7 and Qwen3.6-35B-A3B, MISA-T improves rollout throughput over a sweep-tuned cache-aware vLLM Router by 53.3% and 43.6%, respectively, while maintaining high prefix-cache hit rates. In a matched 50-iteration Step3.7 experiment, it increases rollout throughput by 35.6% and reduces mean iteration time by 22.8%, while keeping the consumed workload mixture close to the trainer target and achieving comparable task scores.

cs.DC↗