Running an LLM on a 7MHz 68000: TINY 2.1 ported to a stock Amiga 500
azhadron · reddit · 2026-11-17
Reddit user azhadron ported WollyMonjac's tiny LLM TINY 2.1 to a stock Amiga 500: a Motorola 68000 (1979 architecture) at 7 MHz, no FPU, 1MB RAM.
- Everything runs in integer math since the 68000 lacks floating point; the C inference engine, 68000 assembly, and GUI were written mostly with Claude Code + Opus
- Every change was verified bit-for-bit against an integer spec in Python, an emulated 68000, and WinUAE before touching real hardware; in exact mode all 44 test answers match the original model
- Throughput is 0.2–0.4 letters/s, with KV cache and a few optimizations, but the original weights are intact
- Open source: the tiny68k repo, requiring TINY.BIN from the original repo
Related event: Dev Runs Tiny LLM on Original Amiga 500(3 posts)→
More from coding & agent
- Self-hosted MCP endpoint exposes 38 database engines, read-only, creds stay server-side — Green_Competition_21 · 2026-11-17
- SubCast: open-source bilingual podcast transcripts with per-sentence AI Q&A — vista8 · 2026-11-17
- "English was always meant to be the final programming language" — tlakomy · 2026-11-17
- MultiHopper ships Swarm for Hire: wallet-enforced budget caps for AI agents — EnigmaFund · 2026-11-17
- Vibe-coded with Claude Code: Roarway, a 20-level 3D jungle runner that plays in the browser — vinishkapoor · 2026-11-17
- Tasks vs Goals: A Reddit Take on Where Agents Are Heading Next — Any-Sport-3761 · 2026-11-17