READER EDITION / ISSUE 002 / Revised 2026-11-20
Creative Agentic Computing
CodeFarmer / Issue 002
YOUR OWN AI WORKSHOP

Edge0 brings
a bigger experiment.

By Alvin

Edge0’s November 12 announcement reports a 145-billion-parameter local model, 29GB peak memory and generation above 80 tokens/s on a compatible setup. Treat these as vendor measurements tied to their hardware and evaluation conditions. [35]

Understand the moving parts

Parameters are a model’s learned numbers. This model uses selected groups per generated token; the guide reports 10.12B active parameters per token. Total model size, download space and working RAM measure different things. [36]

Our Mac download was about 101.6GB. The listing totals 105.4GB when alternative files are included. We used its Mac-compatible Q3_K package with Q4 projection files: labels for compressed model data. A large text-lookup table is read from storage as needed. File names and hashes are recorded in the lab. [36] [37]

Follow the current platform guide

The documented Mac route uses Apple Silicon and Metal, Apple’s graphics-computing system. edge0-strata performs the model calculations; edge0-infer starts it and provides a local address that other programs can call. The Mac default allows 32768 context tokens. It creates an extra draft-model file on first start. [37] [38]

The initial M1 server became ready in 20.1s, including draft-file preparation. The tuned follow-up startup took 10.1s. These are short local observations.