HobbyLM-1B: a 305M-active-param MoE agent that does computer use fully offline on a MacBook Air
lmoroney · x · 2026-10-25
Laurence Moroney highlights HobbyLM-1B Computer Use: a LoRA adapter on a 1B sparse MoE base (8 of 64 experts active, 305M params per token) that turns plain-English requests into step-by-step Mac actions, fully offline on a MacBook Air — the demo books a Sunday 7pm call with WiFi off.
Key facts:
- 14MB MLX LoRA (rank 16, 83 modules); median 0.26s per decision on Apple MLX, FP32
- Completed 90 of 125 sealed live macOS requests (72%)
- Trained in 4 rounds totaling just 60 H100-minutes; Apache-2.0 with training code, evals, and the companion app
- Skills include calculator, brightness, volume, dark mode; arithmetic is delegated to system apps
A concrete data point for how small a useful on-device computer-use assistant can get.
Related event: HobbyLM 1B Sparse MoE Model Runs Computer Use Offline on MacBook(2 posts)→
More from coding & agent
- Coding agent vetoes Effect rewrite: plain TypeScript with 250 lines of helpers wins — gabe_ragland · 2026-10-25
- Leslie Lamport's "Claude, make no mistakes" hailed as predicting LLMs 30 years early — akbirthko · 2026-10-25
- Three Markdown files boost coding agents: 26% smaller diffs, 27% faster runs — lmoroney · 2026-10-25
- Dev bets on Proof-of-Agent experiment to build fully agent-to-agent trading — adamamcbride · 2026-10-25
- Developer vibe-codes his own Jaeger game with Claude — kenan_wyz · 2026-10-25
- Creator builds year-long video pipeline pairing Hailuo H3 with Magnific MCP — TinfoilTricorn · 2026-10-25