PANews 1月21日消息,据量子位报道,DeepSeek在GitHub更新的FlashMLA代码中首次出现“MODEL1”名称,覆盖114个文件中28处提及,且与现有版本V32(DeepSeek-V3.2)并列,暗示MODEL1为下一代新架构模型。代码差异显示该模型在KV缓存布局、稀疏性处理及FP8解码等方面进行了优化,或将在春节前后正式发布。结合近期公开的mHC残差连接机制与Engram记忆模块,MODEL1有望整合多项自研创新。
DeepSeek新模型MODEL1代码曝光,疑为全新架构
Favorite
Share
Disclaimer: This article is copyrighted by the original author and does not represent MyToken’s views and positions. If you have any questions regarding content or copyright, please contact us.(www.mytokencap.com)contact
About MyToken:https://www.mytokencap.com/aboutusArticle Link:https://www.mytokencap.com/news/556093.html
Previous:顾景辞:1.21比特币/以太坊操作策略附行情分析
Related Reading



Altcoins Approach Historic Stress Levels as 38% of Tokens Near All-Time Lows
Altcoins have been under sustained pressure for months as the broader crypto market continues to gra...
NewsBTC2026-03-10 01:00:53

WORLD3 Partners PlaysOut to Bring AI Agents into Web3 Gaming
WORLD3 and PlaysOut teamup to integrate AI agents with gamified mini-game infrastructure to enhance ...
blockchainreporter2026-03-10 01:00:00
XRP Investors In Pain: $50 Billion Worth Of Supply Now In Loss
On-chain data shows the amount of XRP supply sitting underwater has shot up to historically high lev...
NewsBTC2026-03-09 23:30:57