Kimi K3 2.8T Model Run on MacBook Pro via Four Streaming SSDs
Summary
The Deltafin project demonstrated running the 2.8-trillion parameter Kimi K3 model on a MacBook Pro at 1 token/s using four SSDs in parallel.
Prepared by the newsroom with AI assistance; the original link is always provided
The open-source Deltafin project has showcased a method for running massive artificial intelligence models on consumer-grade hardware. By streaming model weights across four SSDs, the 2.8-trillion parameter Kimi K3 model was executed directly on an Apple MacBook Pro.
The setup achieved an inference speed of approximately 1 token per second. This approach illustrates how high-capacity local storage arrays can enable the execution of ultra-large neural networks without relying on high-end enterprise GPU clusters.
Source: Hacker News
Read the original →Latest news
Bank CenterCredit Repurchases Shares for 18.2B Tenge on KASE
Bank CenterCredit JSC has repurchased 4.1 million of its common shares on the Kazakhstan Stock Exchange for 18.2 billion tenge.
Source: KASE
Home Credit Bank Bonds Added to KASE T+ List
Starting October 8, Home Credit Bank bonds (KZQ000001213, HCBNe1) will be included in the KASE T+ List.
Source: KASE
KASE Publishes Member Activity Rankings for September 2026
Kazakhstan Stock Exchange has released its exchange member activity rankings for September 2026 on its official website.
Source: KASE
KASE Index rises 0.25% to 7,862.81 points
On October 7, the KASE Index climbed by 0.25% to settle at 7,862.81 points.
Source: KASE