Qiming Zhu

Qiming Zhu, Jialun Cao, Xuanang Chen et al.

Current research on large language models (LLMs) with retrieval-augmented code generation (RACG) has largely focused on single-language settings, leaving their cross-lingual effectiveness underexplored. Multilingual RACG systems are increasingly important for migrating and reusing code across programming languages (PLs), a common yet challenging task in modern software development. To systematically study cross-lingual code knowledge transfer in RACG, we construct a dataset covering 13 PLs with nearly 14K instances. Our experiments reveal three key insights: (1) Knowledge transfer in RACG across PLs is non-trivial even using direct injection. (2) RACG exhibits unequal cross-lingual knowledge transfer, and its efficacy depends on linguistic affinity of PL pair and diversity of LLM pretraining corpus. (3) RACG shows limited reliance on natural language information embedded in code when equipped with a code-specific retriever. These findings provide practical guidance for designing effective multilingual RACG systems. https://github.com/icip-cas/Cross-Lingual-RACG

Qiming Zhu, Jialun Cao, Yaojie Lu et al.

Code benchmarks such as HumanEval are widely adopted to evaluate the capabilities of Large Language Models (LLMs), providing insights into their strengths and weaknesses. However, current benchmarks primarily exercise LLMs' capability on common coding tasks (e.g., bubble sort, greatest common divisor), leaving domain-specific coding tasks (e.g., computation, system, cryptography) unexplored. To fill this gap, we propose a multi-domain code benchmark, DOMAINEVAL, designed to evaluate LLMs' coding capabilities thoroughly. Our pipeline works in a fully automated manner, enabling a push-bottom construction from code repositories into formatted subjects under study. Interesting findings are observed by evaluating 12 representative LLMs against DOMAINEVAL. We notice that LLMs are generally good at computation tasks while falling short on cryptography and system coding tasks. The performance gap can be as much as 68.94% (80.94% - 12.0%) in some LLMs. We also observe that generating more samples can increase the overall performance of LLMs, while the domain bias may even increase. The contributions of this study include a code generation benchmark dataset DOMAINEVAL, encompassing six popular domains, a fully automated pipeline for constructing code benchmarks, and an identification of the limitations of LLMs in code generation tasks based on their performance on DOMAINEVAL, providing directions for future research improvements. The leaderboard is available at https://domaineval.github.io/.

Qiming Zhu, Zhirui Fang, Tianming Zhang et al.

Embodied intelligence fundamentally requires a capability to determine where to act in 3D space. We formalize this requirement as embodied localization -- the problem of predicting executable 3D points conditioned on visual observations and language instructions. We instantiate embodied localization with two complementary target types: touchable points, surface-grounded 3D points enabling direct physical interaction, and air points, free-space 3D points specifying placement and navigation goals, directional constraints, or geometric relations. Embodied localization is inherently a problem of embodied 3D spatial reasoning -- yet most existing vision-language systems rely predominantly on RGB inputs, necessitating implicit geometric reconstruction that limits cross-scene generalization, despite the widespread adoption of RGB-D sensors in robotics. To address this gap, we propose SpatialPoint, a spatial-aware vision-language framework with careful design that integrates structured depth into a vision-language model (VLM) and generates camera-frame 3D coordinates. We construct a 2.6M-sample RGB-D dataset covering both touchable and air points QA pairs for training and evaluation. Extensive experiments demonstrate that incorporating depth into VLMs significantly improves embodied localization performance. We further validate SpatialPoint through real-robot deployment across three representative tasks: language-guided robotic arm grasping at specified locations, object placement to target destinations, and mobile robot navigation to goal positions.

Jiasheng Zheng, Xin Zheng, Boxi Cao et al.

Code sandboxes have emerged as a critical infrastructure for advancing the coding capabilities of large language models, providing verifiable feedback for both RL training and evaluation. However, existing systems fail to provide accurate verification and efficiency under high-concurrency workloads. We present ScaleBox, a high-fidelity and scalable system designed to address these limitations in large-scale code training. ScaleBox introduces automated special-judge generation and management, fine-grained parallel execution across test cases with seamless multi-node coordination, and a configuration-driven evaluation suite for reproducible benchmarking. A series of experiments demonstrates that ScaleBox significantly enhances code verification accuracy and efficiency. Our further RLVR experiments show that ScaleBox substantially improves both performance on LiveCodeBench and training stability, significantly outperforming heuristic-matching baselines. By providing a reliable and high-throughput infrastructure, ScaleBox facilitates more effective research and development in large-scale code training.

Yuhao Du, Qianwei Huang, Guo Zhu et al.

The rapid advancement of speech-to-speech (S2S) large language models (LLMs) has significantly improved real-time spoken interaction. However, current evaluation frameworks remain inadequate for assessing performance in complex, multi-turn dialogues. To address this, we introduce MTalk-Bench, a multi-turn S2S benchmark covering three core dimensions: Semantic Information, Paralinguistic Information, and Ambient Sound. Each dimension includes nine realistic scenarios, along with targeted tasks to assess specific capabilities such as reasoning. Our dual-method evaluation framework combines Arena-style evaluation (pairwise comparison) and Rubrics-based evaluation (absolute scoring) for relative and absolute assessment. The benchmark includes both model and human outputs, evaluated by human evaluators and LLMs. Experimental results reveal two sets of findings. Overall performance of S2S LLMs: (1) models excel at semantic information processing yet underperform on paralinguistic information and ambient sounds perception; (2) models typically regain coherence by increasing response length, sacrificing efficiency in multi-turn dialogues; (3) modality-aware, task-specific designs outperform brute scaling. Evaluation framework and reliability: (1) Arena and Rubrics yield consistent, complementary rankings, but reliable distinctions emerge only when performance gaps are large; (2) LLM-as-a-judge aligns with humans when gaps are clear or criteria explicit, but exhibits position and length biases and is reliable on nonverbal evaluation only with text annotations. These results highlight current limitations in S2S evaluation and the need for more robust, speech-aware assessment frameworks.

Xin Zheng, Qiming Zhu, Hongyu Lin et al.

Executing computer programs described in natural language has long been a pursuit of computer science. With the advent of enhanced natural language understanding capabilities exhibited by large language models (LLMs), the path toward this goal has been illuminated. In this paper, we seek to examine the capacity of present-day LLMs to comprehend and execute algorithms outlined in natural language. We established an algorithm test set sourced from Introduction to Algorithm, a well-known textbook that contains many representative widely-used algorithms. To systematically assess LLMs' code execution abilities, we selected 30 algorithms, generated 300 random-sampled instances in total, and evaluated whether popular LLMs can understand and execute these algorithms. Our findings reveal that LLMs, notably GPT-4, can effectively execute programs described in natural language, as long as no heavy numeric computation is involved. We believe our findings contribute to evaluating LLMs' code execution abilities and would encourage further investigation and application for the computation power of LLMs.

Qiming Zhu, Shunian Chen, Rui Yu et al.

Long-horizon agents often compress interaction histories into write-time summaries. This creates a fundamental write-before-query barrier: compression decisions are made before the system knows what a future query will hinge on. As a result, summaries can cause unverifiable omissions -- decisive constraints (e.g., allergies) may be dropped, leaving the agent unable to justify an answer with traceable evidence. Retaining raw logs restores an authoritative source of truth, but grounding on raw logs by default is expensive: many queries are answerable from summaries, yet raw grounding still requires processing far longer contexts, inflating token consumption and latency. We propose TierMem, a provenance-linked framework that casts retrieval as an inference-time evidence allocation problem. TierMem uses a two-tier memory hierarchy to answer with the cheapest sufficient evidence: it queries a fast summary index by default, and a runtime sufficiency router Escalates to an immutable raw-log store only when summary evidence is insufficient. TierMem then writes back verified findings as new summary units linked to their raw sources. On LoCoMo, TierMem achieves 0.851 accuracy (vs.0.873 raw-only) while reducing input tokens by 54.1\% and latency by 60.7%.

Qiming Zhu, Zeliang Liu, Jinhui Yan

The recent explosion of machine learning (ML) and artificial intelligence (AI) shows great potential in the breakthrough of metal additive manufacturing (AM) process modeling. However, the success of conventional machine learning tools in data science is primarily attributed to the unprecedented large amount of labeled data-sets (big data), which can be either obtained by experiments or first-principle simulations. Unfortunately, these labeled data-sets are expensive to obtain in AM due to the high expense of the AM experiments and prohibitive computational cost of high-fidelity simulations. We propose a physics-informed neural network (PINN) framework that fuses both data and first physical principles, including conservation laws of momentum, mass, and energy, into the neural network to inform the learning processes. To the best knowledge of the authors, this is the first application of PINN to three dimensional AM processes modeling. Besides, we propose a hard-type approach for Dirichlet boundary conditions (BCs) based on a Heaviside function, which can not only enforce the BCs but also accelerate the learning process. The PINN framework is applied to two representative metal manufacturing problems, including the 2018 NIST AM-Benchmark test series. We carefully assess the performance of the PINN model by comparing the predictions with available experimental data and high-fidelity simulation results. The investigations show that the PINN, owed to the additional physical knowledge, can accurately predict the temperature and melt pool dynamics during metal AM processes with only a moderate amount of labeled data-sets. The foray of PINN to metal AM shows the great potential of physics-informed deep learning for broader applications to advanced manufacturing.