arXiv subjects

Sisi Liu

Publications and source records attributed to Sisi Liu.

5 recordsLinked to original sources

arXiv2026

RouteRelay: Event-Triggered Cross-Layer Route Reuse for Efficient Dynamic Sparse Attention

Dynamic sparse attention reduces long-context prefill cost by routing each query chunk to a small set of key chunks at every Transformer layer. The sparse attention kernel avoids most token interactions, but the router still rebuilds a chunk--chunk score matrix layer after layer, even when the selected routes change little. We introduce RouteRelay, a router-agnostic method that reuses only route metadata across depth while continuing to compute attention with the current layer's queries, keys, and values. Anchor layers perform full routing. Intermediate layers rescore the previous top-$k$ route and a compact sentinel set of near-miss and randomly probed chunks. A query row is rerouted only when a sentinel challenges its weakest selected chunk. We give a top-$k$ stability condition, a probabilistic bound on missed challengers, and a row-selective GPU execution design. In a reproducible empirical evaluation, RouteRelay retains at least 99.99% route recall while rerouting 25.0%, 55.4%, and 78.2% of rows under low, moderate, and high cross-layer drift, respectively. Across routing scales, RouteRelay retains 100.0% recall while evaluating 38.4--51.6% of full-routing score pairs as the key-chunk count grows from 128 to 1024. Its unfused CPU execution remains slower than dense matrix multiplication, exposing row compaction and ledger updates as the main kernel-engineering targets.

cs.CL↗

arXiv2026

Reducing False Positives in Static Bug Detection with LLMs: An Empirical Study in Industry

Static analysis tools (SATs) are widely adopted in both academia and industry for improving software quality, yet their practical use is often hindered by high false positive rates, especially in large-scale enterprise systems. These false alarms demand substantial manual inspection, creating severe inefficiencies in industrial code review. While recent work has demonstrated the potential of large language models (LLMs) for false alarm reduction on open-source benchmarks, their effectiveness in real-world enterprise settings remains unclear. To bridge this gap, we conduct the first comprehensive empirical study of diverse LLM-based false alarm reduction techniques in an industrial context at Tencent, one of the largest IT companies in China. Using data from Tencent's enterprise-customized SAT on its large-scale Advertising and Marketing Services software, we construct a dataset of 433 alarms (328 false positives, 105 true positives) covering three common bug types. Through interviewing developers and analyzing the data, our results highlight the prevalence of false positives, which wastes substantial manual effort (e.g., 10-20 minutes of manual inspection per alarm). Meanwhile, our results show the huge potential of LLMs for reducing false alarms in industrial settings (e.g., hybrid techniques of LLM and static analysis eliminate 94-98% of false positives with high recall). Furthermore, LLM-based techniques are cost-effective, with per-alarm costs as low as 2.1-109.5 seconds and $0.0011-$0.12, representing orders-of-magnitude savings compared to manual review. Finally, our case analysis further identifies key limitations of LLM-based false alarm reduction in industrial settings.

cs.SE↗

arXiv2024

Ultrafast Excited-State Energy Transfer in Phenylene Ethynylene Dendrimer: Quantum Dynamics with Tensor Network Method

Photo-induced excited-state energy transfer (EET) processes play an important role in the solar energy conversions. The phenylene ethynylene (PE) dendrimers display great potential in improving the efficiency of solar cells, because of their excellent photo-harvesting and exciton-transport properties. In this work, we investigated the intramolecular EET dynamics in a dendrimer composed of two linear PE units (2-ring and 3-ring) using the full quantum dynamics based on the tensor network method. We first constructed a diabatic model Hamiltonian based on the electronic structure calculations. Using this diabatic vibronic coupling model, we tried to obtain the main features of the EET dynamics in terms of the several diabatic models with different numbers of vibrational modes (from 4 modes to 129 modes) and to explore the corresponding vibronic coupling interactions. The results show that the EET in the current PE dendrimer is an ultrafast process. Four modes with A' symmetry play dominant roles in the dynamics, other 86 modes with A' symmetry can damp the electronic coherence, and the modes of A" symmetry do not show the significant influence on the EET process. Overall, the first-order intrastate vibronic coupling terms show the dominant roles in the EET dynamics, while the second-order intrastate vibronic coupling terms give the visible impact here by damping the electronic coherence and slowing down the overall EET process. This work provides a valuable understanding of the physical insight in the EET dynamics of PE dendrimers.

physics.chem-ph↗

arXiv2022

TiBERT: Tibetan Pre-trained Language Model

The pre-trained language model is trained on large-scale unlabeled text and can achieve state-of-the-art results in many different downstream tasks. However, the current pre-trained language model is mainly concentrated in the Chinese and English fields. For low resource language such as Tibetan, there is lack of a monolingual pre-trained model. To promote the development of Tibetan natural language processing tasks, this paper collects the large-scale training data from Tibetan websites and constructs a vocabulary that can cover 99.95$\%$ of the words in the corpus by using Sentencepiece. Then, we train the Tibetan monolingual pre-trained language model named TiBERT on the data and vocabulary. Finally, we apply TiBERT to the downstream tasks of text classification and question generation, and compare it with classic models and multilingual pre-trained models, the experimental results show that TiBERT can achieve the best performance. Our model is published in http://tibert.cmli-nlp.com/

cs.CL↗

arXiv2015

Parallel Toolkit for Measuring the Quality of Network Community Structure

Many networks display community structure which identifies groups of nodes within which connections are denser than between them. Detecting and characterizing such community structure, which is known as community detection, is one of the fundamental issues in the study of network systems. It has received a considerable attention in the last years. Numerous techniques have been developed for both efficient and effective community detection. Among them, the most efficient algorithm is the label propagation algorithm whose computational complexity is O(|E|). Although it is linear in the number of edges, the running time is still too long for very large networks, creating the need for parallel community detection. Also, computing community quality metrics for community structure is computationally expensive both with and without ground truth. However, to date we are not aware of any effort to introduce parallelism for this problem. In this paper, we provide a parallel toolkit to calculate the values of such metrics. We evaluate the parallel algorithms on both distributed memory machine and shared memory machine. The experimental results show that they yield a significant performance gain over sequential execution in terms of total running time, speedup, and efficiency.

cs.SI↗