arXiv subjects

Jianling Li

Publications and source records attributed to Jianling Li.

10 recordsLinked to original sources

arXiv2026

SkillSight: Calibrating Generic Content Bias for Skill Retrieval

As large language model agents gain access to increasingly large skill libraries, retrieving the right skill becomes critical to reliable capability selection and execution. Existing retrievers often treat skill contents as ordinary documents, overlooking their highly regular structure: shared descriptive patterns recur across many skills while providing little evidence for distinguishing the required capability. We show that this shared descriptive background is reflected in dense relevance scores, induces a pronounced energy gap between queries and skill documents, and obscures discriminative signals, especially for structurally similar hard negatives. Based on this observation, we propose SkillSight, a training-free retrieval framework that calibrates shared background in both semantic and lexical spaces. Semantic Background Calibration estimates a background subspace from generic tokens identified by IDF, reducing similarity induced by shared descriptive patterns, while Lexical Evidence Calibration downweights shared background tokens to recover discriminative token-level evidence. Experiments on SRA-Bench and SkillBench-Supp demonstrate consistent improvements across retrieval metrics, with SkillSight improving Recall@10 by up to 20.21 percentage points over the original dense retriever. It is up to 1,248 times faster than the Dense + Reranker baseline. In end-to-end evaluation, SkillSight achieves the best overall performance across three agent models and outperforms LLM Selection by up to 4.97 percentage points. These results identify shared descriptive background as a source of ranking interference in skill retrieval and demonstrate that calibrating it enables accurate and efficient skill selection without additional training. Our code can be found at https://github.com/xiaojinying/SkillSight.

cs.AI↗

arXiv2025

AutoTriton: Automatic Triton Programming with Reinforcement Learning in LLMs

Kernel development in deep learning requires optimizing computational units across hardware while balancing memory management, parallelism, and hardware-specific optimizations through extensive empirical tuning. Although domain-specific languages like Triton simplify GPU programming by abstracting low-level details, developers must still manually tune critical parameters such as tile sizes and memory access patterns through iterative experimentation, creating substantial barriers to optimal performance and wider adoption. In this work, we introduce AutoTriton, the first model dedicated to Triton programming powered by reinforcement learning (RL). AutoTriton performs supervised fine-tuning (SFT) to be equipped with essential Triton programming expertise using a high-quality data gathering pipeline, and conducts RL with Group Relative Policy Optimization (GRPO) algorithm, combining a rule-based reward and an execution-based reward to further improve Triton programming ability, sequentially. Experiments across five evaluation channels of TritonBench and KernelBench illustrate that our 8B model AutoTriton achieves performance comparable to mainstream large models, including Claude-4-Sonnet and DeepSeek-R1-0528. Further experimental analysis demonstrates the crucial role of each module within AutoTriton, including the SFT stage, the RL stage, and the reward design strategy. These findings underscore the promise of RL for automatically generating high-performance kernels, and since high-performance kernels are core components of AI systems, this breakthrough establishes an important foundation for building more efficient AI systems. The model and code will be available at https://github.com/AI9Stars/AutoTriton.

cs.LG↗

arXiv2025

Contrastive Learning on LLM Back Generation Treebank for Cross-domain Constituency Parsing

Cross-domain constituency parsing is still an unsolved challenge in computational linguistics since the available multi-domain constituency treebank is limited. We investigate automatic treebank generation by large language models (LLMs) in this paper. The performance of LLMs on constituency parsing is poor, therefore we propose a novel treebank generation method, LLM back generation, which is similar to the reverse process of constituency parsing. LLM back generation takes the incomplete cross-domain constituency tree with only domain keyword leaf nodes as input and fills the missing words to generate the cross-domain constituency treebank. Besides, we also introduce a span-level contrastive learning pre-training strategy to make full use of the LLM back generation treebank for cross-domain constituency parsing. We verify the effectiveness of our LLM back generation treebank coupled with contrastive learning pre-training on five target domains of MCTB. Experimental results show that our approach achieves state-of-the-art performance on average results compared with various baselines.

cs.CL↗

arXiv2025

ChartEdit: How Far Are MLLMs From Automating Chart Analysis? Evaluating MLLMs' Capability via Chart Editing

Although multimodal large language models (MLLMs) show promise in generating chart rendering code, editing charts via code presents a greater challenge. This task demands MLLMs to integrate chart understanding and reasoning capacities, which are labor-intensive. While many MLLMs claim such editing capabilities, current evaluations rely on limited case studies, highlighting the urgent need for a comprehensive evaluation framework. In this work, we propose \textsc{ChartEdit}, a novel benchmark designed for chart editing tasks, featuring $1405$ diverse editing instructions applied to $233$ real-world charts, each manually annotated and validated for accuracy. Utilizing \textsc{ChartEdit}, we evaluate the performance of 10 mainstream MLLMs across two types of experiments at both the code and chart levels. The results suggest that large-scale models can generate code to produce images that partially match the reference images. However, their ability to generate accurate edits according to the instructions remains limited. The state-of-the-art (SOTA) model achieves a score of only $59.96$, highlighting significant challenges in precise modification. In contrast, small-scale models, including chart-domain models, struggle both with following editing instructions and generating overall chart images, underscoring the need for further development in this area. Code is available at https://github.com/xxlllz/ChartEdit.

cs.CL↗

arXiv2025

TritonBench: Benchmarking Large Language Model Capabilities for Generating Triton Operators

Triton, a high-level Python-like language designed for building efficient GPU kernels, is widely adopted in deep learning frameworks due to its portability, flexibility, and accessibility. However, programming and parallel optimization still require considerable trial and error from Triton developers. Despite advances in large language models (LLMs) for conventional code generation, these models struggle to generate accurate, performance-optimized Triton code, as they lack awareness of its specifications and the complexities of GPU programming. More critically, there is an urgent need for systematic evaluations tailored to Triton. In this work, we introduce TritonBench, the first comprehensive benchmark for Triton operator generation. TritonBench features two evaluation channels: a curated set of 184 real-world operators from GitHub and a collection of operators aligned with PyTorch interfaces. Unlike conventional code benchmarks prioritizing functional correctness, TritonBench also profiles efficiency performance on widely deployed GPUs aligned with industry applications. Our study reveals that current state-of-the-art code LLMs struggle to generate efficient Triton operators, highlighting a significant gap in high-performance code generation. TritonBench will be available at https://github.com/thunlp/TritonBench.

cs.CL↗

arXiv2023

LLM-enhanced Self-training for Cross-domain Constituency Parsing

Self-training has proven to be an effective approach for cross-domain tasks, and in this study, we explore its application to cross-domain constituency parsing. Traditional self-training methods rely on limited and potentially low-quality raw corpora. To overcome this limitation, we propose enhancing self-training with the large language model (LLM) to generate domain-specific raw corpora iteratively. For the constituency parsing, we introduce grammar rules that guide the LLM in generating raw corpora and establish criteria for selecting pseudo instances. Our experimental results demonstrate that self-training for constituency parsing, equipped with an LLM, outperforms traditional methods regardless of the LLM's performance. Moreover, the combination of grammar rules and confidence criteria for pseudo-data selection yields the highest performance in the cross-domain constituency parsing.

cs.CL↗

arXiv2021

Word Graph Guided Summarization for Radiology Findings

Radiology reports play a critical role in communicating medical findings to physicians. In each report, the impression section summarizes essential radiology findings. In clinical practice, writing impression is highly demanded yet time-consuming and prone to errors for radiologists. Therefore, automatic impression generation has emerged as an attractive research direction to facilitate such clinical practice. Existing studies mainly focused on introducing salient word information to the general text summarization framework to guide the selection of the key content in radiology findings. However, for this task, a model needs not only capture the important words in findings but also accurately describe their relations so as to generate high-quality impressions. In this paper, we propose a novel method for automatic impression generation, where a word graph is constructed from the findings to record the critical words and their relations, then a Word Graph guided Summarization model (WGSum) is designed to generate impressions with the help of the word graph. Experimental results on two datasets, OpenI and MIMIC-CXR, confirm the validity and effectiveness of our proposed approach, where the state-of-the-art results are achieved on both datasets. Further experiments are also conducted to analyze the impact of different graph designs to the performance of our method.

cs.CL↗

arXiv2014

Geometric Scaling for a Detonation Wave Governed by a Pressure-Dependent Reaction Rate and Yielding Confinement

The propagation of detonation waves in reactive media bounded by an inert, compressible layer is examined via computational simulations in two different geometries, axisymmetric cylinders and two dimensional, planar slabs. For simplicity, an ideal gas equation of state is used with a pressure-dependent reaction rate that results in a stable detonation wave structure. The detonation is initiated as an ideal Chapman-Jouguet (CJ) detonation with a one-dimensional structure, and then allowed to propagate into a finite diameter or thickness layer of explosive surrounded by an inert layer. The yielding confinement of the inert layer results in the detonation wave decaying to a sub-CJ steady state velocity or failing entirely. Simulations are performed with different values of the reaction rate pressure exponent (n = 2 and 3) and different impedance confinement (greater than, less than, and equal to the confinement of the explosive). The velocity decrement and critical dimension (critical diameter or thickness) are determined, and a 2:1 scaling between the cylinder diameter and slab thickness results is confirmed, in good agreement with curvature-based models of detonation propagation. The measured shock front curvature and detonation velocity relation agrees with the classic model of Wood and Kirkwood. The computational simulations are compared to a simple, analytic model that treats the interaction of the confinement with the detonation products via Newtonian theory and a model that assumes a continuous variation in shock front curvature with the shock angle at the interface with the confinement matching the angle determined by shock polar analysis. The Newtonian model works very well for the case of high impedance confinement, while the shock front curvature model agrees with the simulations for the case of low impedance confinement.

physics.flu-dyn↗

arXiv2014

Propagation Distance Required to Reach Steady-State Detonation Velocity in Finite-Sized Charges

The decay of a detonation wave from its initial CJ velocity to its final, steady state velocity upon encountering a finite thickness or diameter charge is investigated numerically and theoretically. The numerical simulations use an ideal gas equation of state and pressure dependent reaction rate in order to ensure a stable wave structure. The confinement is also treated as an ideal gas with variable impedance. The velocity decay along the centerline is extracted from the simulations and compared to predictions base on a front evolution equation that uses the steady state detonation velocity-front curvature relation ($D_n-κ$). This model fails to capture the finite signaling speed of the leading rarefaction resulting from the interaction with the yielding confinement. This signaling speed is verified to be the maximum signal velocity occurring in the ideal ZND wave structure of the initial CJ velocity. A simple heuristic model based on the rarefaction generated by a one-dimensional interaction between the post-shock state and the confinement is proposed to provide an approximate description of the rest of the relaxation to steady state velocity.

physics.flu-dyn↗

arXiv2014

Effect of Spatial Heterogeneity on Near-Limit Propagation of a Stable Detonation

The effect of introducing a spatial heterogeneity into an explosive medium is studied computationally by examining the detonation velocity near the limit to propagation in a thin explosive layer. The explosive system studied is an ideal gas with a single exothermic reaction governed by a pressure-dependent reaction rate ($p^n$) with a pressure exponent of $n = 3$. A pressure-dependent reaction rate, rather than the exponential dependence of reaction on temperature of Arrhenius kinetics, is used so that the detonation wave is stable in the homogeneous case and can be modelled with simple, analytical techniques, and thus the effect of introducing heterogeneity can be clearly identified. The two-dimensional slab of explosive is bounded by a layer of inert gas with the same thermodynamic properties as the explosive. The heterogeneity is introduced into the explosive via a large-amplitude, two-dimensional sinusoidal ripple in density in the initialization of the simulation, while maintaining a constant pressure. The computational simulations are initialized with a ZND solution for the ideal CJ detonation, and the detonation is allowed to propagate into the explosive layer. The simulations show that the detonation in the heterogeneous media exhibits a cellular-like structure of complex shock interactions. The detonation is able to propagate into a significantly thinner layer of explosive and can exhibit a greater velocity than the corresponding homogeneous case. A parametric study of varying the wavelength of the sinusoid shows the existence of an optimal size of heterogeneity at which the favorable effect is the greatest corresponding to a wavelength that is approximately 10 to 50 times the half reaction zone length of the ideal CJ detonation. As the wavelength is decreased to the size of the reaction zone length, the behavior reverts back to the homogeneous case.

physics.flu-dyn↗