Large Language Models (LLMs) are widely adopted, but their high training cost leads many developers to fine-tune existing open-source models. While most adhere to open-source licenses, some falsely claim original training despite clear derivation from public models, raising pressing concerns about intellectual property protection and the need to verify model provenance. In this paper, we propose GhostSpec, a lightweight yet effective method for verifying LLM lineage without access to training data or modification of model behavior. Our approach constructs compact and robust fingerprints by applying singular value decomposition (SVD) to invariant products of internal attention weight matrices. Unlike watermarking or output-based methods, GhostSpec is fully data-free, non-invasive, and computationally efficient. Extensive experiments show it is robust to fine-tuning, pruning, expansion, and adversarial transformations, reliably tracing lineage with minimal overhead. By offering a practical solution for model verification, our method contributes to intellectual property protection and fosters a transparent, trustworthy LLM ecosystem.

Large language models (LLMs) have achieved remarkable success across various domains, driving significant technological advancements and innovations.Despite the rapid growth in model scale and capability, systematic, data-driven research on how structural configurations affect performance remains scarce.To address this gap, we present a large-scale dataset encompassing diverse open-source LLM structures and their performance across multiple benchmarks.Leveraging this dataset, we conduct a systematic, data mining-driven analysis to validate and quantify the relationship between structural configurations and performance.Our study begins with a review of the historical development of LLMs and an exploration of potential future trends.We then analyze how various structural choices impact performance across benchmarks and further corroborate our findings using mechanistic interpretability techniques.By providing data-driven insights into LLM optimization, our work aims to guide the targeted development and application of future models.We release our dataset at

Attribute-based encryption (ABE) is a promising technique for fine-grained access control of encrypted data in a cloud storage, however, decryption involved in the ABEs is usually too expensive for resource-constrained front-end users, which greatly hinders its practical popularity. In order to reduce the decryption overhead for a user to recover the plaintext, Green et al. suggested to outsource the majority of the decryption work without revealing actually data or private keys. To ensure the third-party service honestly computes the outsourced work, Lai et al. provided a requirement of verifiability to the decryption of ABE, but their scheme doubled the size of the underlying ABE ciphertext and the computation costs. Roughly speaking, their main idea is to use a parallel encryption technique, while one of the encryption components is used for the verification purpose. Hence, the bandwidth and the computation cost are doubled. In this paper, we investigate the same problem. In particular, we propose a more efficient and generic construction of ABE with verifiable outsourced decryption based on an attribute-based key encapsulation mechanism, a symmetric-key encryption scheme and a commitment scheme. Then, we prove the security and the verification soundness of our constructed ABE scheme in the standard model. Finally, we instantiate our scheme with concrete building blocks. Compared with Lai et al.'s scheme, our scheme reduces the bandwidth and the computation costs almost by half.