暂无摘要(点击查看详情)
暂无摘要(点击查看详情)
This paper presents Mecellem models, a framework for developing specialized language models for the Turkish legal domain through domain adaptation strategies. We make two contributions: (1)Encoder Model Pre-trained from Scratch: ModernBERT-based bidirectional encoders pre-trained on a Turkish-dominant corpus of 112.7 billion tokens. We implement a checkpoint selection strategy that evaluates downstream retrieval performance throughout training, revealing that optimal checkpoints achieve best retrieval scores before pre-training loss reaches its minimum. Our encoder models achieve top-3 rankings on the Turkish retrieval leaderboard, with smaller models (155M parameters) achieving comparable performance to larger reference models (307M-567M parameters). Our approach achieves 92.36% production efficiency compared to state-of-the-art models (embeddinggemma-300m: 100.00%, BAAI/bge-m3: 99.54%, newmindai/bge-m3-stsb: 94.38%), ranking fourth overall despite requiring less computational resources. SOTA models rely on multi-stage, computationally intensive training pipelines, making our single-stage pre-training followed by efficient post-training approach a cost-effective alternative; (2)De
Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm for grounding large language model (LLM) outputs in retrieved evidence, thereby reducing hallucination and improving factual accuracy. Its efficacy, however, remains largely unexamined for low-resource, non-Latin-script languages such as Khmer. In this paper, we present a RAG-based question answering system for Khmer-language telecom-domain documents. We conduct a two-phase comparative evaluation. First, we benchmark three embedding models: BGE-M3 (567M), Jina-Embeddings-v3 (570M), and Qwen3-Embedding (597M), for dense retrieval over Khmer documents. BGE-M3 consistently performs best, achieving a Hit Rate@3 of 0.285, File Hit Rate@3 of 0.700, MRR@3 of 0.221, and Precision@3 of 0.112, substantially outperforming the other retrievers. Second, using BGE-M3 as the selected retriever, we evaluate five generator backends: Qwen3 (8B), Qwen3.5 (9B), Sailor2-8B-Chat, SeaLLMs-v3-7B-Chat, and Llama-SEA-LION-v2-8B-IT, on a curated golden dataset of 200 Khmer question-answer pairs. To quantify system performance, we apply six RAGAS-inspired metrics: faithfulness, answer relevance, context relevance, factual correctness, an
Virtual Private Networks (VPNs) offer an alternative solution using Internet Protocol (IP) tunnels to create secure, encrypted communication between geographically distant networks using a common shared medium such as the Internet. They use tunneling to establish end-to-end connectivity. OpenVPN is a cross-platform, secure, highly configurable VPN solution. Security in OpenVPN is handled by the OpenSSL cryptographic library which provides strong security over a Secure Socket Layer (SSL) using standard algorithms such as Advanced Encryption Standard (AES), Blowfish, or Triple DES (3DES). The Linksys WRT54GL router is a consumer-grade router made by Linksys, a division of Cisco Systems, capable of running under Linux. The Linux-based DD-WRT open-source router firmware can run OpenVPN on the Linksys WRT54GL router. For this case study, the performance of OpenVPN is measured and analyzed using a $2^{k-p}$ fractional factorial design for 5 minus 1 factors where $k=5$ and $p=1$. The results show that the throughput is mainly limited by the encryption cipher used, and that the round-trip time (RTT) is mostly dependent on the transport protocol selected.
Scientists have built the first all-optical photonic time crystal, allowing them to reshape the behavior of terahertz light at extraordinary speeds。 The breakthrough could open the door to ultrafast computing, smarter communication systems, advanced imaging, and a new generation of highly tunable lasers
A supermassive black hole has been found lurking about 30,000 light-years from the center of its galaxy—the first dormant black hole detected so far from a galactic core。 It revealed itself only after shredding a passing star and producing a brief, brilliant flare。 Astronomers think it may be the remnant of a devoured galaxy or a black hole flung o
DOGE's inflated "Wall of Receipts": 96% of grant savings unverifiable, GAO says
A person’s sleeping brain may reveal warning signs of dementia long before memory problems begin。 Researchers used machine learning to analyze EEG recordings from about 7,000 adults and found that an older-than-expected “brain age” was tied to a sharply higher dementia risk。 Every additional 10 years of brain aging raised that risk by nearly 40%
Photons traveling through a cloud of atoms can emerge so early that they appear to have spent a negative amount of time inside。 Researchers tested whether this was merely a misleading feature of the light pulse by making extremely weak measurements of the atoms。 Surprisingly, the atoms confirmed the same negative dwell time
A new nanostructured carbon design lets fuel-cell catalysts use tiny amounts of platinum while remaining remarkably stable and efficient。 The breakthrough could help hydrogen fuel cells become a more practical way to power data centers, vehicles, and other energy-intensive technologies
A surprisingly simple change could make sodium-ion batteries far more powerful while opening the door to turning seawater into drinking water。 Researchers at the University of Surrey found that sodium vanadium oxide performs much better when its naturally occurring water is left inside instead of being removed during manufacturing
Cassini data has revealed a surprising twist in Saturn’s magnetic shield: the planet’s rapid rotation appears to drag a key opening in its magnetosphere far toward the afternoon side instead of keeping it near noon, as happens on Earth。 The finding confirms that giant planets like Saturn operate under a fundamentally different magnetic regime, wher
Scientists traced a mysterious surge of low-energy gamma rays from zinc-70 to magnetic changes occurring inside its nucleus。 The breakthrough could improve models of how stars, supernovae, and neutron star mergers create heavy elements
NASA engineers have found a clever way to squeeze more life out of Voyager 2, nearly half a century after it left Earth。 In an operation nicknamed the “Big Bang,” the team simultaneously shut down certain power-hungry hardware and switched to lower-power alternatives while keeping the spacecraft warm enough to survive
A ten-month Antarctic experiment found that astronauts may struggle not only with loneliness but also with spending too much time around the same people。 Frequent contact was linked to greater tension and mistrust, while the crew increasingly divided into cultural and language-based groups
Scientists have demonstrated that heat can move through a crystal in focused, wave-like rays at room temperature instead of spreading randomly。 The breakthrough could make it possible to route heat around sensitive parts of next-generation chips and quantum devices