arXiv subjects

Qingan Li

Publications and source records attributed to Qingan Li.

7 recordsLinked to original sources

arXiv2026

POP: Prefill-Only Pruning for Efficient Large Model Inference

Large Language Models (LLMs) and Vision-Language Models (VLMs) have demonstrated remarkable capabilities. However, their deployment is hindered by significant computational costs. Existing structured pruning methods, while hardware-efficient, often suffer from significant accuracy degradation. In this paper, we argue that this failure stems from a stage-agnostic pruning approach that overlooks the asymmetric roles between the prefill and decode stages. By introducing a virtual gate mechanism, our importance analysis reveals that deep layers are critical for next-token prediction (decode) but largely redundant for context encoding (prefill). Leveraging this insight, we propose Prefill-Only Pruning (POP), a stage-aware inference strategy that safely omits deep layers during the computationally intensive prefill stage while retaining the full model for the sensitive decode stage. To enable the transition between stages, we introduce independent Key-Value (KV) projections to maintain cache integrity, and a boundary handling strategy to ensure the accuracy of the first generated token. Extensive experiments on Llama-3.1, Qwen3-VL, and Gemma-3 across diverse modalities demonstrate that POP achieves up to 1.37$\times$ speedup in prefill latency with minimal performance loss, effectively overcoming the accuracy-efficiency trade-off limitations of existing structured pruning methods.

cs.CL↗

arXiv2025

Turing Pattern Engineering Enables Kinetically Ultrastable yet Ductile Metallic Glasses

Enhancing the kinetic stability of glasses often necessitates deepening thermodynamic stability, which typically compromises ductility due to increased structural rigidity. Decoupling these properties remains a critical challenge for functional applications. Here, we demonstrate that pattern engineering in metallic glasses (MGs) enables unprecedented kinetic ultrastability while retaining thermodynamic metastability and intrinsic plasticity. Through atomistic simulations guided by machine-learning interatomic potentials and replica-exchange molecular dynamics, we reveal that clustering oxygen contents, driven by reaction-diffusion-coupled pattern dynamics, act as localized pinning sites. These motifs drastically slow structural relaxation, yielding kinetic stability comparable to crystal-like ultrastable glasses while retaining an energetic as-cast state. Remarkably, the thermodynamically metastable state preserves heterogeneous atomic mobility, allowing strain delocalization under mechanical stress. By tailoring oxygen modulation via geometric patterning, we achieve an approximately 200 K increase in the onset temperature of the glass transition (Tonset) while maintaining fracture toughness akin to conventional MGs. This work establishes a paradigm of kinetic stabilization without thermodynamic compromise, offering a roadmap to additively manufacture bulk amorphous materials with combined hyperstability and plasticity.

cond-mat.mtrl-sci↗

arXiv2025

A$^2$ATS: Retrieval-Based KV Cache Reduction via Windowed Rotary Position Embedding and Query-Aware Vector Quantization

Long context large language models (LLMs) pose significant challenges for efficient serving due to the large memory footprint and high access overhead of KV cache. Retrieval-based KV cache reduction methods can mitigate these challenges, typically by offloading the complete KV cache to CPU and retrieving necessary tokens on demand during inference. However, these methods still suffer from unsatisfactory accuracy degradation and extra retrieval overhead. To address these limitations, this paper proposes A$^2$ATS, a novel retrieval-based KV cache reduction method. A$^2$ATS aims to obtain an accurate approximation of attention scores by applying the vector quantization technique to key states, thereby enabling efficient and precise retrieval of the top-K tokens. First, we propose Windowed Rotary Position Embedding, which decouples the positional dependency from query and key states after position embedding. Then, we propose query-aware vector quantization that optimizes the objective of attention score approximation directly. Finally, we design the heterogeneous inference architecture for KV cache offloading, enabling long context serving with larger batch sizes. Experimental results demonstrate that A$^2$ATS can achieve a lower performance degradation with similar or lower overhead compared to existing methods, thereby increasing long context serving throughput by up to $2.7 \times$.

cs.CL↗

arXiv2024

CHESS: Optimizing LLM Inference via Channel-Wise Thresholding and Selective Sparsification

Deploying large language models (LLMs) on edge devices presents significant challenges due to the substantial computational overhead and memory requirements. Activation sparsification can mitigate these resource challenges by reducing the number of activated neurons during inference. Existing methods typically employ thresholding-based sparsification based on the statistics of activation tensors. However, they do not model the impact of activation sparsification on performance, resulting in suboptimal performance degradation. To address the limitations, this paper reformulates the activation sparsification problem to explicitly capture the relationship between activation sparsity and model performance. Then, this paper proposes CHESS, a general activation sparsification approach via CHannel-wise thrEsholding and Selective Sparsification. First, channel-wise thresholding assigns a unique threshold to each activation channel in the feed-forward network (FFN) layers. Then, selective sparsification involves applying thresholding-based activation sparsification to specific layers within the attention modules. Finally, we detail the implementation of sparse kernels to accelerate LLM inference. Experimental results demonstrate that the proposed CHESS achieves lower performance degradation over eight downstream tasks while activating fewer parameters than existing methods, thus speeding up the LLM inference by up to 1.27x.

cs.CL↗

arXiv2021

TinyAKE: A More Practicable and Trustable Scheme for Authenticated Key Establishment in WSNs

The characteristics of high loss rate, resource constraint, being eager for good security haven't been fully considered in the existing key establishment protocols of wireless sensor networks. Analyzing the key establishing problem from the MAC and physical layers, existing protocols are not practicable enough due to their overlong agreement packets and single round key establishment. To mitigate the impact of these problems, a group of design principles for secure sensor networks has been presented and TinyAKE, an authenticated key transport protocol based on lightweight certificate, is proposed in this paper. The security of TinyAKE are proved with the theory of indistinguishability, meanwhile, the correctness is also proved, the performance is analyzed and compared with the existing similar protocols. Finally TinyAKE is implemented in the TinyOS with TinyECC. Our evaluation shows that TinyAKE is a more practicable and trustable authenticated key establishment protocol than existing protocols. The experimental result shows that the key transport with certificate mechanism is feasible in WSNs. Moreover, the simulation results show that the optimal number of repeated negotiation is one when the secure connectivity rate of TinyAKE is improved by using the repeated key negotiation.

cs.CR↗

arXiv2015

Evidence of photo-induced dynamic competition of metallic and insulating phase in a layered manganite

We show evidence that the competition between the antiferromagetic metallic phase and the charge- and orbital-ordered insulating phase at the reentrant phase boundary of a layered manganite, LaSr2Mn2O7, can be manipulated using ultrafast optical excitation. The time-dependent evolution of the Jahn-Teller superlattice reflection, which indicates the formation of the charge and orbital order, was measured at different laser fluences. The laser-induced enhancement and reduction the Jahn-Teller reflection intensity shows a reversal of sign between earlier (~10 ns) and later (~150 ns) time delays during the relaxation after photo excitation. This effect is consistent with a scenario whereby the laser excitation modulates the local competition between the metallic and the insulating phases.

cond-mat.mtrl-sci↗

arXiv2014

Brownmillerite Ca2Co2O5: Synthesis, Stability, and Re-entrant Single-Crystal-to-Single-Crystal Structural Transitions

Ca2Co2O5 in the brownmillerite form was synthesized using a high-pressure optical-image floating zone furnace, and single crystals with dimensions up to 1.4x0.8x0.5 mm3 were obtained. At room temperature, Ca2Co2O5 crystallizes as a fully ordered brownmillerite variant in the orthorhombic space group Pcmb (No. 57) with unit cell parameters a=5.28960(10) Å, b=14.9240(2) Å, and c=10.9547(2) Å. With decreasing temperature, it undergoes a re-entrant sequence of first-order structural phase transitions (Pcmb to P2/c11 to P121/m1 to Pcmb) that is unprecedented among brownmillerites, broadening the family of space groups available to these materials and challenging current approaches for sorting the myriad variants of brownmillerite structures. Magnetic susceptibility data indicate antiferromagnetic ordering in Ca2Co2O5 occurs near 240 K, corroborated by neutron powder diffraction. Below 140 K, the specimen exhibits a weak ferromagnetic component directed primarily along the b axis that shows a pronounced thermal and magnetic history dependence.

cond-mat.mtrl-sci↗