arXiv subjects

Tao Ji

Publications and source records attributed to Tao Ji.

At least 19 recordsLinked to original sources

arXiv2026

ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents

Large language models are deployed as long-horizon agents that compose diverse tools across MCP services into executable workflows. Existing MCP environment-scaling methods are data-centric: they first synthesize a database and scenario, then derive tools and tasks from them. As a result, tool interfaces may not match real ones, tool compositions are not grounded in execution, and tasks rarely span more than one environment. We propose ToolSpace, a tool-centric framework that reverses this order. ToolSpace mines data, state, and retry dependencies from adjacent calls in execution traces of real MCP services and open agent corpora, using deterministic rules rather than LLM judgment, and clusters the resulting weighted tool-composition graph into environments. Databases are synthesized only for state-mutating environments, while query and stateless tools replay recorded returns. ToolSpace then performs state-aware random walks with rollback and decompiles each chain into a natural-language task with assertion-based checks, keeping only samples that replay successfully from the initial state. Applied to AutomationBench and tau3-Bench Banking, the pipeline produces harder Pro versions on which 7 frontier models drop by up to 38.2 points. Across seven long-horizon tool-use benchmarks, ToolSpace raises Qwen3.6-35B-A3B from 23.0 to 41.0 average score (43.2 with RL) and GLM-5.2 from 44.6 to 56.6, on par with Claude Opus 5. Ablations show that execution-witnessed edges clearly outperform LLM-judged ones, and performance keeps improving with data and environment scale.

cs.CL↗

arXiv2026

Ordered Chaos: Towards Adaptive Multipath Load Balancing with Precise Fast Loss Detection

Load balancing in best-effort datacenter fabrics faces a tradeoff. Order-preserving schemes such as Equal-Cost Multipath (ECMP) retain loss semantics but use path diversity coarsely, while packet spraying balances load well but removes the ordering signals needed for fast, precise loss detection. We present Ordered Chaos, an end-host design that preserves enough structure for reliable loss detection without giving up fine-grained multipath routing. Reactive Subflow Spraying (RSS) divides each flow into ordered, dynamically routed subflows. Precise Fast Loss Detection (PFLD) combines this ordering with probes to distinguish reordering from congestion, corruption, tail, and retransmission losses within one RTT. Across simulations, AI and HPC workloads, and a DPU prototype, RSS+PFLD improves tail flow completion time by up to $3\times$, eliminates up to 99% of timeout-based recoveries, and accelerates collective and training communication by up to 35%.

cs.NI↗

arXiv2026

When the Weakest Language Keeps Moving: ARMOR-DP for Multilingual BabyLM

Multilingual BabyLM systems usually fix language sampling proportions even though the weakest language can change during training. We present ARMOR-DP, a modular framework that diagnoses imbalance arising from representation, exposure, capacity, and objective pressure, and uses per-language validation loss to update sampling. Controlled experiments show that equal character budgets create a severe representation artifact for Chinese; byte-level encoding removes much of it, while static Chinese protection can subsequently make English the weakest language. The dynamic controller tracks these transitions and achieves a better score--balance tradeoff than fixed protection. On official evaluation, curriculum-dynamic checkpoints outperform a matched static Chinese-protective baseline while using fewer Chinese epochs; scaling capacity yields the strongest internal ARMOR-DP system. These results support a simple conclusion: multilingual weakness is a training-time state rather than a fixed property of a language. ARMOR-DP therefore separates diagnosis from intervention, allowing protection to follow the current bottleneck instead of a predetermined language label.

cs.CL↗

arXiv2026

A Few More Tokens Ahead: Reinforcing Parallelism in Diffusion LLMs

Parallel generation is widely regarded as an inherent advantage of diffusion large language models. Yet our fixed-quota evaluation of four representative models reveals severe degradation across mathematics and code, with all scores below 15% at 16 tokens per step. Token-level probes identify local predictive potential near the observed prefix, especially in mathematical text. We propose RePara, which reinforces short consecutive token blocks through group-relative policy optimization and reduces rollout and log-probability recomputation costs. On LLaDA and SDAR, \ours improves GSM8K and MATH-500 accuracy over JustGRPO under reduced denoising budgets, with a 1.96x training-step speedup in our LLaDA/MATH setup.

cs.LG↗

arXiv2026

MLAfication: Efficiently Enabling High-Fidelity MLA for Modern LLMs

Multi-Head Latent Attention (MLA) compresses the KV cache into a latent representation, reducing inference memory and bandwidth. Existing MHA2MLA/TransMLA methods do not support QKNorm and require costly recovery on modern LLMs, especially MoE. We present MLAfication, a QKNorm-compatible, parameter-efficient extension of MHA2MLA. It combines calibrated KNorm, parameter-efficient distillation, contribution-aware Partial-RoPE, and activation-aware joint KV initialization. Across MHA and GQA models, including QKNorm- and MoE-equipped variants, MLAfication preserves performance under substantial KV-cache compression, surpasses the current SOTA on commonsense and long-context benchmarks, and extends MLA conversion to modern architectures. On Qwen3-30B-A3B, MLAfication delivers up to $5.35\times$ the serving throughput of BF16 GQA and, combined with 4-bit KV quantization, reduces the KV-cache footprint by $94.73\%$. Our code is available at https://github.com/JT-Ushio/MLAfication.

cs.LG↗

arXiv2026

MuseCritic: Learning Multi-Aspect Song Rewards through Natural-Language Aesthetic Critiques

Long-form song generation models continue to improve in duration, structural coherence, and acoustic complexity, increasing the need for reliable aesthetic rewards aligned with human preferences. However, reward models for complete songs remain limited, and existing evaluators typically predict scores in a single forward pass without readable explanations. To this end, we introduce MuseCritic, a semi-scalar reward model that generates a natural-language critique covering five aesthetic dimensions and uses it as an intermediate representation to predict continuous reward scores. MuseCritic follows a two-stage training pipeline: a teacher model first provides high-quality critiques for supervised fine-tuning, then the fine-tuned model generates its own critiques for reward learning, mitigating training-inference distribution shift. On an in-domain test set of 200 SongEval songs, MuseCritic reduces macro-averaged mean squared error from 0.2875 to 0.2316 and improves macro-averaged LCC, SRCC, and Kendall's tau to 0.9068, 0.8838, and 0.7178, respectively. On the out-of-domain Music Arena benchmark with 733 preference pairs, it achieves 71.35% accuracy and remains competitive with strong music-specific reward models. Using MuseCritic with GRPO also improves Muse-0.6B on all nine aesthetic metrics from SongEval and Audiobox Aesthetics. These results show that critique-conditioned reward modeling reduces scoring error and provides an effective optimization signal for song generation. The project repository is available at https://github.com/WuqnEl/MuseCritic.

cs.SD↗

arXiv2026

HelloWorld: Towards Practical Applications of Generative Driving World Models

Driving world models provide a promising route toward scalable counterfactual data generation and interactive simulation beyond recorded driving logs. Realizing this potential requires a system that can generalize across diverse scenes, respond faithfully to prescribed controls, generate coherent multi-sensor observations, and operate efficiently under repeated inference. We present \textbf{HelloWorld}, a 2B driving world model system designed around these requirements. HelloWorld progressively specializes broad visual and motion priors from heterogeneous video data into controllable driving generation using ego pose, HD maps, and 3D boxes. A block-causal generation interface, together with adaptation to self-generated context, aligns the model with sequential simulation. The system further supports synchronized seven-camera RGB generation and conditional LiDAR synthesis, and is distilled toward few-step inference for efficient deployment. Experiments evaluate visual quality, control fidelity, cross-view consistency, robustness under repeated generation, inference efficiency, and LiDAR synthesis. Together, HelloWorld provides a unified framework for scalable driving data generation and interactive simulation.

cs.CV↗

arXiv2026

Looped GPT-BERT: Trading Parameters for Computation in Small Language Modeling

When training data are limited, increasing parameter count is not the only way to improve language-model performance. A small parameter set, when repeatedly applied, can also deliver comparable performance. We study Looped GPT-BERT in the BabyLM 2026 Strict-small setting, combining GPT-BERT's masked next-token and causal language-modeling objectives with depth-wise parameter sharing. We train on a preprocessed 7.48M-word English corpus and compare objective ratios, non-looped and looped architectures, and loop counts. Our final $4\times12$ model uses four physical layers for twelve recurrent traversals and contains 12.18M parameters. The BabyLM 2026 leaderboard reports an Overall Average of 35.42 and an NLP Average of 48.48. Compared with public BabyLM 10M Strict-small GPT-2 and GPT-BERT baselines, it achieves comparable performance on selected linguistic and downstream metrics, including BLiMP and GLUE, with fewer parameters. The loop ablations show that additional recurrent computation can improve training and preserve strong performance on selected linguistic tasks, whereas poorer performance on other tasks may reveal an inherent limitation of the looped design: using only a few physical layers restricts the model's representational space.

cs.CL↗

arXiv2026

Strongly Magnetized Super-Eddington Accretion: How Spin and Accretion Rate Regulate Energy Output and Mass Loss

Strongly magnetized super-Eddington accretion flows power many important astrophysical systems, but how black hole parameters control their output is unclear. We present 32 general relativistic radiation magnetohydrodynamics simulations of super-Eddington magnetically arrested disks onto stellar-mass black holes, varying mass ($M_{\rm BH}= 5, 15, 30\,M_{\odot}$), spin ($a=0,0.9$), and accretion rate ($\dot{M}_{\rm acc} \approx 1-2000\,\dot{M}_{\rm Edd}$). We find that black hole spin and accretion rate jointly regulate wind loss rates and energy output efficiencies, while black hole mass has no effect over the mass range studied here. The BH accretes only $10-40\%$ of the mass supplied to the accretion flow, while the rest is expelled in winds. This accretion fraction decreases with mass supply rate and is lower for high-spin systems. Both spin states produce strong magnetically driven outflows. For $a = 0$, the wind kinetic, radiative, and electromagnetic efficiencies are modest and show little variation across the full simulated range of accretion rates. For $a = 0.9$, both wind power and jet power increase super-linearly with $\dot{M}_{\rm acc}$, with the jet power saturating beyond $\dot{M}_{\rm acc} \sim 100\,\dot{M}_{\rm Edd}$. Radiation is strongly beamed along the funnel, with inverse beaming factors exceeding $100$ for high-spin, high-$\dot{m}$ models viewed face-on. Our results establish that rapid BH spin boosts energy-extraction efficiency, while high accretion rate amplifies total power. We provide scaling relations for luminosities, jet power, accretion ratio, and beaming, offering a framework for interpreting observations of ULXs and other super-Eddington systems.

astro-ph.HE↗

arXiv2026

PDFBench: A Benchmark for De novo Protein Design from Function

Function-guided protein design is a crucial task with significant applications in drug discovery and enzyme engineering. However, the field lacks a unified and comprehensive evaluation framework. Current models are assessed using inconsistent and limited subsets of metrics, which prevents fair comparison and a clear understanding of the relationships between different evaluation criteria. To address this gap, we introduce PDFBench, the first comprehensive benchmark for function-guided denovo protein design. Our benchmark systematically evaluates eight state-of-the-art models on 16 metrics across two key settings: description-guided design, for which we repurpose the Mol-Instructions dataset, originally lacking quantitative benchmarking, and keyword-guided design, for which we introduce a new test set, SwissTest, created with a strict datetime cutoff to ensure data integrity. By benchmarking across a wide array of metrics and analyzing their correlations, PDFBench enables more reliable model comparisons and provides key insights to guide future research.

cs.LG↗

arXiv2026

CL-bench Life: Can Language Models Learn from Real-Life Context?

Today's AI assistants such as OpenClaw are designed to handle context effectively, making context learning an increasingly important capability for models. As these systems move beyond professional settings into everyday life, the nature of the contexts they must handle also shifts. Real-life contexts are often messy, fragmented, and deeply tied to personal and social experience, such as multi-party conversations, personal archives, and behavioral traces. Yet it remains unclear whether current frontier language models can reliably learn from such contexts and solve tasks grounded in them. To this end, we introduce CL-bench Life, a fully human-curated benchmark comprising 405 context-task pairs and 5,348 verification rubrics, covering common real-life scenarios. Solving tasks in CL-bench Life requires models to reason over complex, messy real-life contexts, calling for strong real-life context learning abilities that go far beyond those evaluated in existing benchmarks. We evaluate ten frontier LMs and find that real-life context learning remains highly challenging: even the best-performing model achieves only 19.3% task solving rate, while the average performance across models is only 13.8%. Models still struggle to reason over contexts such as messy group chat histories and fragmented behavioral records from everyday life. CL-bench Life provides a crucial testbed for advancing real-life context learning, and progress on it can enable more intelligent and reliable AI assistants in everyday life.

cs.CL↗

arXiv2026

SpeechRole: A Large-Scale Dataset and Benchmark for Evaluating Speech Role-Playing Agents

Speech is essential for realistic role-playing, yet existing work on role-playing agents largely centers on text, leaving Speech Role-Playing Agents (SRPAs) underexplored and without systematic evaluation. We introduce SpeechRole, a unified framework for developing and assessing SRPAs. SpeechRole-Data contains 98 roles and 111k speech-to-speech conversations with rich timbre and prosodic variation, providing large-scale resources for training SRPAs. SpeechRole-Eval offers a multidimensional benchmark that directly evaluates generated speech, preserving paralinguistic cues and measuring interaction ability, speech expressiveness, and role-playing fidelity. Experiments show that end-to-end SRPAs such as GPT-4o Audio achieve strong fluency and naturalness, but remain limited in prosody consistency and emotion appropriateness. In contrast, current open-source end-to-end models exhibit substantial performance gaps across multiple evaluation dimensions. Cascaded and end-to-end systems achieve comparable results in interaction ability and role-playing fidelity, suggesting that these aspects are still largely influenced by the underlying text-based language models. We release all data, code, and evaluation tools at https://github.com/yuhui1038/SpeechRole.

cs.CL↗

arXiv2026

Why Reinforcement Fine-Tuning Enables MLLMs Preserve Prior Knowledge Better: A Data Perspective

Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt (multimodal) large language models to downstream tasks. While effective at task adaptation, their impact on retaining prior knowledge remains unclear. In this paper, we introduce jigsaw puzzles as a novel task absent from existing pretraining corpora and systematically study the behavior of SFT and RFT on the open-source Qwen2.5-VL series. Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly but better maintains prior knowledge. We study this phenomenon through learning dynamics by examining both the magnitude and direction of how training data influence prior knowledge. Our analysis shows that RFT mainly reinforces correct samples naturally aligned with the base model's probability landscape, leading to weaker interference with prior knowledge. Moreover, training on RFT-simulated rollouts, which exert a smaller magnitude of influence and are better aligned in direction to prior knowledge, allows SFT to preserve prior knowledge better while rapidly learning new tasks. We further validate our framework on Qwen2.5 post-training in math and scientific QA, observing consistent forgetting and learning-dynamics trends. These findings suggest that the distribution of post-training data, rather than algorithmic differences alone, plays a central role in forgetting, and highlight RFT as a promising ingredient for stable continual post-training.

cs.CL↗

arXiv2026

What Makes a Good Speech Tokenizer for LLM-Centric Speech Generation? A Systematic Study

Speech-language models (SLMs) offer a promising path toward unifying speech and text understanding and generation. However, challenges remain in achieving effective cross-modal alignment and high-quality speech generation. In this work, we systematically investigate the role of speech tokenizer designs in LLM-centric SLMs, augmented by speech heads and speaker modeling. We compare coupled, semi-decoupled, and fully decoupled speech tokenizers under a fair SLM framework and find that decoupled tokenization significantly improves alignment and synthesis quality. To address the information density mismatch between speech and text, we introduce multi-token prediction (MTP) into SLMs, enabling each hidden state to decode multiple speech tokens. This leads to up to 12$\times$ faster decoding and a substantial drop in word error rate (from 6.07 to 3.01). Furthermore, we propose a speaker-aware generation paradigm and introduce RoleTriviaQA, a large-scale role-playing knowledge QA benchmark with diverse speaker identities. Experiments demonstrate that our methods enhance both knowledge understanding and speaker consistency.

cs.CL↗

arXiv2026

MHA2MLA-VLM: Enabling DeepSeek's Economical Multi-Head Latent Attention across Vision-Language Models

As vision-language models (VLMs) tackle increasingly complex and multimodal tasks, the rapid growth of Key-Value (KV) cache imposes significant memory and computational bottlenecks during inference. While Multi-Head Latent Attention (MLA) offers an effective means to compress the KV cache and accelerate inference, adapting existing VLMs to the MLA architecture without costly pretraining remains largely unexplored. In this work, we present MHA2MLA-VLM, a parameter-efficient and multimodal-aware framework for converting off-the-shelf VLMs to MLA. Our approach features two core techniques: (1) a modality-adaptive partial-RoPE strategy that supports both traditional and multimodal settings by selectively masking nonessential dimensions, and (2) a modality-decoupled low-rank approximation method that independently compresses the visual and textual KV spaces. Furthermore, we introduce parameter-efficient fine-tuning to minimize adaptation cost and demonstrate that minimizing output activation error, rather than parameter distance, substantially reduces performance loss. Extensive experiments on three representative VLMs show that MHA2MLA-VLM restores original model performance with minimal supervised data, significantly reduces KV cache footprint, and integrates seamlessly with KV quantization.

cs.CV↗

arXiv2026

Muse: Towards Reproducible Long-Form Song Generation with Fine-Grained Style Control

Recent commercial systems such as Suno demonstrate strong capabilities in long-form song generation, while academic research remains largely non-reproducible due to the lack of publicly available training data, hindering fair comparison and progress. To this end, we release a fully open-source system for long-form song generation with fine-grained style conditioning, including a licensed synthetic dataset, training and evaluation pipelines, and Muse, an easy-to-deploy song generation model. The dataset consists of 116k fully licensed synthetic songs with automatically generated lyrics and style descriptions paired with audio synthesized by SunoV5. We train Muse via single-stage supervised finetuning of a Qwen-based language model extended with discrete audio tokens using MuCodec, without task-specific losses, auxiliary objectives, or additional architectural components. Our evaluations find that although Muse is trained with a modest data scale and model size, it achieves competitive performance on phoneme error rate, text--music style similarity, and audio aesthetic quality, while enabling controllable segment-level generation across different musical structures. All data, model weights, and training and evaluation pipelines will be publicly released, paving the way for continued progress in controllable long-form song generation research. The project repository is available at https://github.com/yuhui1038/Muse.

cs.SD↗

arXiv2026

CMDAR: A Chinese Multi-scene Dynamic Audio Reasoning Benchmark with Diverse Challenges

The ability to reason from audio, including speech, environmental sounds, and music, is essential for AI agents to interact effectively in real-world scenarios. Existing benchmarks mainly focus on static or single-scene settings and English audio data and do not fully capture scenarios where multiple speakers, unfolding events, and heterogeneous audio sources interact. To address these challenges, we introduce CMDAR, a Chinese benchmark for evaluating models on complex, multi-scene, and dynamically evolving audio reasoning tasks. CMDAR comprises 3,000 carefully curated question-answer pairs linked to diverse audio clips, covering five categories of complex reasoning and spanning three question types. We benchmark 26 state-of-the-art audio language models on CMDAR and observe that they exhibit limitations in complex reasoning tasks. In CMDAR-main, Qwen2.5-Omni achieves 76.67% accuracy, whereas GPT-4o Audio reaches 68.47%. However, GPT-4o Audio substantially outperforms Qwen2.5-Omni on the more challenging multiple-choice with multiple audios and open-ended tasks. And we provide detail analysis corresponding suggestions for the future development of large audio language models.

cs.SD↗

arXiv2025

The mass distribution of clumpy accretion onto the nearby young star TW Hya

The proliferation of high time-resolution and decades-long monitoring of classical T Tauri stars provides a vast opportunity to test the variability of the star-disk connections. However, most monitoring surveys use single broad-band filters, which makes the conversion of photometric variability into accretion rate difficult. In this study, we analyze accretion bursts onto the nearby young star TW Hya over short (hours, days) and long (months, years) timescales by calibrating TESS and ASAS-SN $g$-band photometry to accretion rates with simultaneous spectroscopy. The high cadence TESS light curve shows bursts of accretion in clumps with masses from a sensitivity limit of $\sim10^{-13}$~M$_\odot$ up to $3\times 10^{-11}$\,M$_\odot$. The average burst duration of 1.8 days is longer than a simple estimate of the thermal response timescale, supporting the interpretation that the photometric variability probes the instantaneous accretion rate. The reset timescale of 1.2--2 days derived from the structure function and previously reported quasi-periods of 3.5--4 days are consistent with bursts that may be related to the different rotation between the stellar magnetosphere and inner disk or with azimuthal asymmetries in the inner disk. The near-daily ASAS-SN light curve across 8 years reveals some seasonal changes in brightness with a standard deviation of $\sim 0.13$ mag, about half of the scatter seen on short timescales. This study demonstrates the importance of coordinating contemporaneous multi-epoch spectroscopy with time domain surveys to interpret light curves of young stars.

astro-ph.SR↗