PISA: block sparse attention with log-linear complexity
ResearchRelated articles
Mac Studio M5 Ultra: 512GB of unified memory on the desktop
Four dies, 512GB of unified memory at 1.2 TB/s: the Mac Studio M5 Ultra runs 100-560B MoE models locally. The speed math, the cost comparison with the DGX Station, and when the Mac is the right choice.
Mistral Large 4 in open weights: a 1T MoE (52B active) for on-premise inference
One trillion parameters, 52B active, 1M context, open-weight license: the quant, RAM and speed math for running it on-premise.
SynthAPI: synthetic environments and scaling laws for training tool-use agents
Controlled synthetic environments with failures on a dial to train local tool-use agents: SynthAPI's scaling laws and the math — ~€1.60 of energy vs ~€158 of API for 10,000 tasks.
Get the next local-AI briefing
Once a week, the top selection from iaLocale: articles, news and resources on local AI, and more.
After signing up you receive a confirmation email (double opt-in). Without confirmation we send nothing. No spam, we promise.

