Gate News 消息，3月25日，谷歌研究院发布量化压缩算法TurboQuant，可将大语言模型的KV缓存压缩至3 bit，内存占用缩减至少6倍，无需训练或微调，不损失模型精度。在4 bit模式下，于英伟达H100 GPU上计算注意力的速度较32 bit未量化基线提升最高8倍。研究团队在LongBench、Needle In A Haystack、ZeroSCROLLS等长上下文基准上使用Gemma和Mistral模型进行验证，TurboQuant在所有测试中均达到最优表现。该算法由两个子算法组成：PolarQuant通过极坐标变换消除传统量化方法的内存开销，QJL仅用1 bit校正残余误差。该研究由谷歌研究院Amir Zandieh和副总裁兼Google Fellow Vahab Mirrokni主导，与韩国KAIST和纽约大学合作完成，将在ICLR 2026上发表。谷歌表示该技术的主要应用之一是解决Gemini等模型的KV缓存瓶颈。
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Articoli correlati
A16z 最新報告：為什麼區塊鏈是 AI 代理缺少的那塊基礎建設？
a16z crypto 最新報告指出，AI agents 正從輔助工具演變為經濟行為者，然而在身份、支付和跨平台協作等核心基礎設施方面仍存在重大缺口。報告強調，隨著AI在治理和交易中的參與，驗證機制成為信任的關鍵，區塊鏈技術可提供可驗證的基礎設施來解決這些挑戰。未来需要透過加密機制來確保AI agents 真正代表使用者意志，並改變傳統支付系統。
ChainNewsAbmedia6h fa
Three Major Platforms Control 75% of Stock Perpetual Futures Market in Q1 2026
TokenInsight’s Q1 2026 report reveals that the stock perpetual futures market is dominated by a few top platforms, which collectively hold about 75% market share. Major exchanges are increasingly offering U.S. stock and finance products to enhance cross-asset trading.
GateNews13h fa
Cross-Asset Hedging Emerges as Mainstream Strategy, Q1 Report Shows
A report by Block Scholes reveals rising correlations between crypto assets and traditional markets, noting increased demand for unified trading platforms as traders manage diverse assets. Trading volumes have surged, reflecting a shift towards multi-asset strategies.
GateNews14h fa
Digital Asset Investment Products Record $1.4B Net Inflows Last Week, Highest Since January
CoinShares reported $1.4 billion in net inflows for digital asset investment products last week, marking the largest increase since January. Bitcoin led with $1.116 billion, while Ethereum saw $328 million inflow. The U.S. contributed significantly, though Switzerland experienced outflows.
GateNews14h fa
DeFi 駭客 4 月盜走 6 億美元，Kelp DAO 和 Drift 佔月度損失 95%
2026 年 4 月僅 20 天內，加密協議因駭客攻擊損失超過 6.06 億美元，成為自 2025 年 2 月交易所 14 億美元資料外洩事件以來最嚴峻的單月損失紀錄。KelpDAO 和 Drift Protocol 兩起攻擊合計佔 4 月損失的 95%，以及 2026 年截至目前 7.718 億美元總損失的 75%。
MarketWhisper18h fa
Moody's: Stablecoin Market Exceeds $315.8B, but Near-Term Bank Threat Remains Limited
Moody's report shows stablecoins have reached a market value of $315.8 billion, primarily dominated by USDT. While near-term risks to banks are limited due to narrow adoption and regulation, long-term growth may challenge traditional banking.
GateNews18h fa
