Microsoft 公佈 MAI-Base-2:總參數量 2.8T、啟用參數量 80B 的 MoE 模型,聲稱預訓練 FLOPs 效率較 DeepSeek-V4-Pro-Base 高 4x-12x
Microsoft unveils MAI-Base-2, a MoE model with 2.8T total and 80B active parameters, claiming 4x-12x more pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base (Microsoft AI)
Microsoft 公佈 MAI-Base-2,並稱其預訓練 FLOPs 效率較 DeepSeek-V4-Pro-Base 高 4x-12x。這款 MoE 模型的總參數量為 2.8T,啟用參數量為 80B。
Featured Podcasts
The $7B Startup Mapping Your Health In Just One Hour | Neko's Hjalmar Nilsonne
Veteran tech reporter Alex Konrad sits down with breakout entrepreneurs taking on the status quo to shake up their fields in AI, design, nuclear energy, space, and more.
Subscribe to The Upstarts Podcast.
The company building underwater drones for $50,000
Stripe cofounder John Collison interviews founders, builders, and leaders over a pint.
Subscribe to Cheeky Pint.
Jensen Huang | Does AI Revenue Justify Capex, '27 AI Bottlenecks, Singularity, Open vs Closed & more
Open-source podcast on all things tech, markets, investing, and capitalism, hosted by Brad Gerstner.
Subscribe to BG2 Pod.
The Bumpy Road To AGI — With Kevin Roose
The Big Technology Podcast takes you behind the scenes in the tech world featuring interviews with plugged-in insiders and outside agitators.
Subscribe to Big Technology Podcast.
How to Run a Political Campaign in the TikTok Era (And How Not To)
Media and tech aren't just intersecting - they're fully intertwined. To understand how those worlds work, Peter Kafka talks to industry leaders, upstarts and observers.
Subscribe to Channels with Peter Kafka.
Robinhood's CEO on letting AI trade your money
Candid conversations with the leaders shaping what's next in AI and the broader tech industry. Hosted by Alex Heath, who has covered Silicon Valley for more than a decade.
Subscribe to Sources.
來源:Techmeme · techmeme.com





