Kimi K3 runs on a single Mac Studio after MLX port cuts it to 350GB
pcuenq · x · 2026-07-29
- A team says it has made Kimi K3 run on a single Mac Studio using MLX.
- The model is described as 2.8T parameters and 1.6TB on disk in its original form; the port reduces it to about 350GB.
- Two engineering tricks are highlighted: a streaming converter that processes the model layer by layer, and REAP pruning, which scores all 896 experts against a calibration corpus and keeps only the ones needed for the workload.
- The result is an open-source MLX port that brings a very large model into Apple Silicon territory.
More from coding & agent
- 'LLM code is sloppy' rebutted with real ffmpeg code: identical if/else branches in a top library — kingslyj · 2026-10-16
- Office-editing tool calls need file, worktree and unit: lessons from dsh-univer-office — FunDistribution6537 · 2026-10-16
- Architect MD workflow: Gemini one-shot outputs that beat Opus, via 5-lens plan critique — oikk01 · 2026-10-16
- Prove a coding agent actually ran the tests: check artifacts, not approvals — ComprehensiveShake76 · 2026-10-16
- Building an acceptance harness for AI spreadsheet edits: allowlists, recalc checks, idempotency — ukdicha-modakkk · 2026-10-16
- Agent actions need checks outside the model: NVIDIA demos lean on guardrails amid wave of real agent incidents — Creamy-And-Crowded · 2026-10-16