A technical demonstration of running large mixture-of-experts models (GLM-5.3 744B and others) from SSDs on a MacBook Pro, achieving 4 tokens/second by optimizing expert routing and split reads across multiple drives. The key innovation is selective expert reading—routing to only necessary experts rather than all—combined with parallel SSD access to minimize read latency rather than maximize throughput.
The US Strategic Petroleum Reserve, created after the 1973 OAPEC oil embargo, is the world's largest crude oil storage facility holding up to 714 million barrels. It serves as a buffer against oil supply shocks and has become a key energy policy tool, though it has physical limits on how much oil can be safely removed.
Samsung's 990 PRO Gen4 2TB NVMe SSD is on sale for $385 on Amazon, a 40% discount from its regular $640 price, driven by global NAND flash memory shortages. The drive offers fast speeds, 2TB storage capacity, thermal control, and Samsung Magician Software for optimization, making it suitable for professionals and creatives working with large files.
Researchers introduce BITCOS, a distribution-adaptive compression layout for ternary LLMs that exploits the actual sparsity of weights (up to 51.5% zeros) to achieve storage below the theoretical 1.585-bit limit, reaching 1.485 bits per weight and improving inference throughput by up to 1.28× on production kernels across multiple platforms.
Backups are essential but complex, requiring more than simple mirroring due to risks like ransomware, data corruption, and user error. The article discusses backup principles including multiple copies, snapshots at appropriate intervals (RPO), and retention rotation strategies to balance storage costs with recovery needs.