arXiv subjects

Xiaolin Lin

Publications and source records attributed to Xiaolin Lin.

6 recordsLinked to original sources

arXiv2026

CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference

Long-context large language model (LLM) inference is increasingly constrained by the memory footprint and decoding cost of key-value (KV) caches, limiting sustainable deployment on resource-constrained hardware. Existing KV cache eviction methods typically apply heuristic token scoring over all heads in GQA-based LLMs. These methods ignore the different functionalities of attention heads, leading to the eviction of critical tokens and thus degrading the performance of LLMs. To address this issue, we propose CompressKV, a resource-efficient KV-cache compression framework for GQA-based LLMs. Instead of aggregating attention scores from all heads, CompressKV identifies Semantic Retrieval Heads (SRHs) that capture both the initial and final tokens of a prompt and semantically important mid-context evidence, and uses them to select tokens whose KV pairs should be retained. Furthermore, CompressKV allocates cache budgets across layers according to offline estimates of layer-wise eviction error. Experiments on LongBench and Needle-in-a-Haystack show that CompressKV consistently outperforms existing KV-cache eviction methods across memory budgets. Notably, it preserves over 97\% of full-cache performance using only 3\% of the KV cache on LongBench question-answering tasks and achieves 90\% accuracy with just 0.7\% KV storage on Needle-in-a-Haystack. These results demonstrate an improved resource--performance trade-off for long-context LLM inference. Our code is publicly available at: https://github.com/TUDa-HWAI/CompressKV

cs.AI↗

arXiv2026

Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding

While speculative decoding improves inference throughput for multi-batch long-context Large Language Models (LLMs), its efficiency is often limited by a verification bottleneck where Key-Value (KV) cache loading dominates latency. Existing compression methods fail in this regime: static eviction incurs accuracy loss due to saliency shift, while dynamic selection introduces prohibitive computational overhead during the verification path. We propose Dustin, a sparse verification framework designed for long-context speculative decoding. Dustin integrates lookahead signals from the draft model with historical attention from the target model to identify critical tokens with high fidelity across multi-step verification windows. To reduce recomputation latency, this approach further employs a sparse estimation scheme that restricts importance scoring to a minimal subset of attention heads. Evaluations on PG-19 and LongBench with Qwen2.5-72B demonstrate that Dustin achieves a 27.85x speedup in self-attention and a 9.17x end-to-end decoding speedup at a 32k sequence length, all with negligible accuracy degradation.

cs.CL↗

arXiv2026

AI Psychometrics: Evaluating the Psychological Reasoning of Large Language Models with Psychometric Validities

The immense number of parameters and deep neural networks make large language models (LLMs) rival the complexity of human brains, which also makes them opaque ``black box'' systems that are challenging to evaluate and interpret. AI Psychometrics is an emerging field that aims to tackle these challenges by applying psychometric methodologies to evaluate and interpret the psychological traits and processes of artificial intelligence (AI) systems. This paper investigates the application of AI Psychometrics to evaluate the psychological reasoning and overall psychometric validity of four prominent LLMs: GPT-3.5, GPT-4, LLaMA-2, and LLaMA-3. Using the Technology Acceptance Model (TAM), we examined convergent, discriminant, predictive, and external validity across these models. Our findings reveal that the responses from all these models generally met all validity criteria. Moreover, higher-performing models like GPT-4 and LLaMA-3 consistently demonstrated superior psychometric validity compared to their predecessors, GPT-3.5 and LLaMA-2. These results help to establish the validity of applying AI Psychometrics to evaluate and interpret large language models.

cs.AI↗

arXiv2025

CompressKV: Semantic Retrieval Heads Know What Tokens are Not Important Before Generation

Recent advances in large language models (LLMs) have significantly boosted long-context processing. However, the increasing key-value (KV) cache size poses critical challenges to memory and execution efficiency. Most KV cache compression methods rely on heuristic token eviction using all attention heads in Grouped Query Attention (GQA)-based LLMs. This method ignores the different functionalities of attention heads, leading to the eviction of critical tokens and thus degrades the performance of LLMs. To address the issue above, instead of using all the attention heads in GQA-based LLMs to determine important tokens as in the previous work, we first identify the attention heads in each layer that are not only capable of retrieving the initial and final tokens of a prompt, but also capable of retrieving important tokens within the text and attending to their surrounding semantic context. Afterwards, we exploit such heads to determine the important tokens and retain their corresponding KV cache pairs. Furthermore, we analyze the cache eviction error of each layer individually and introduce a layer-adaptive KV cache allocation strategy. Experimental results demonstrate the proposed CompressKV consistently outperforms state-of-the-art approaches under various memory budgets on LongBench and Needle-in-a-Haystack benchmarks. Our code is publicly available at: https://github.com/TUDa-HWAI/CompressKV.git.

cs.CL↗

arXiv2024

A Practice-Friendly LLM-Enhanced Paradigm with Preference Parsing for Sequential Recommendation

The training paradigm integrating large language models (LLM) is gradually reshaping sequential recommender systems (SRS) and has shown promising results. However, most existing LLM-enhanced methods rely on rich textual information on the item side and instance-level supervised fine-tuning (SFT) to inject collaborative information into LLM, which is inefficient and limited in many applications. To alleviate these problems, this paper proposes a practice-friendly LLM-enhanced paradigm with preference parsing (P2Rec) for SRS. Specifically, in the information reconstruction stage, we design a new user-level SFT task for collaborative information injection with the assistance of a pre-trained SRS model, which is more efficient and compatible with limited text information. Our goal is to let LLM learn to reconstruct a corresponding prior preference distribution from each user's interaction sequence, where LLM needs to effectively parse the latent category of each item and the relationship between different items to accomplish this task. In the information augmentation stage, we feed each item into LLM to obtain a set of enhanced embeddings that combine collaborative information and LLM inference capabilities. These embeddings can then be used to help train various future SRS models. Finally, we verify the effectiveness and efficiency of our TSLRec on three SRS benchmark datasets.

cs.IR↗

arXiv2023

HazardNet: Road Debris Detection by Augmentation of Synthetic Models

We present an algorithm to detect unseen road debris using a small set of synthetic models. Early detection of road debris is critical for safe autonomous or assisted driving, yet the development of a robust road debris detection model has not been widely discussed. There are two main challenges to building a road debris detector: first, data collection of road debris is challenging since hazardous objects on the road are rare to encounter in real driving scenarios; second, the variability of road debris is broad, ranging from a very small brick to a large fallen tree. To overcome these challenges, we propose a novel approach to few-shot learning of road debris that uses semantic augmentation and domain randomization to augment real road images with synthetic models. We constrain the problem domain to uncommon objects on the road and allow the deep neural network, HazardNet, to learn the semantic meaning of road debris to eventually detect unseen road debris. Our results demonstrate that HazardNet is able to accurately detect real road debris when only trained on synthetic objects in augmented images.

cs.CV↗