{"name":"tinyworldformer","repo":"SheepSeb/tinyworldformer","url":"https://midorreal.com/project/sheepseb-tinyworldformer","repository":"https://github.com/SheepSeb/tinyworldformer","description":"Tiny world model generates games and stream ac plays them all on Pico 2W","facts":{"language":{"value":"Rust","source":"https://github.com/SheepSeb/tinyworldformer","checked":"21 Nov 2026"},"homepage":{"value":"https://sheepseb.github.io/stealth/posts/tinyworldformer.html","source":"https://sheepseb.github.io/stealth/posts/tinyworldformer.html","checked":"21 Nov 2026"},"releases":{"value":"0","source":"https://github.com/SheepSeb/tinyworldformer/releases","checked":"21 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/SheepSeb/tinyworldformer/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/SheepSeb/tinyworldformer","checked":"21 Nov 2026"}},"stars":377.0,"writeup":{"what_it_is":"tinyworldformer is a project where two small neural networks run on a Raspberry Pi Pico 2 W. A looped transformer world model generates game frames and rewards from a frame and action, and a small actor network picks actions in that generated game. It includes several small numpy-based games: Breakout, Doom-lite, Flappy Bird, Pong and Pac-Man lite.","audience":"Developers and researchers interested in tiny neural networks on microcontrollers","claims":[{"kind":"specific","claim":"The world model and actor both run on the Pico 2 W's core 1, with core 0 handling USB and the display.","excerpt":"Both run on core 1 of the RP2350, while core 0 handles USB and the display.","status":"not_checked"},{"kind":"specific","claim":"The world model is between 26k and 131k parameters.","excerpt":"26k–131k parameters","status":"not_checked"},{"kind":"specific","claim":"The Breakout world model reaches 99.99% exact frames at 2 loops on held-out transitions.","excerpt":"Breakout | 20×16 | paddle, ball, 56 bricks, 5 lives | the serve after a miss | 26k params, 4.3M MACs | 99.99%","status":"not_checked"},{"kind":"specific","claim":"The agent uses Stream AC(λ) and learns one transition at a time without a replay buffer.","excerpt":"learns from one transition at a time: no replay buffer, no batches, no target network","status":"not_checked"},{"kind":"specific","claim":"The project provides an interface to play the real games in a terminal.","excerpt":"You can play the real games in a terminal","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-10-30"},"why_now":null,"owner_supplied":null,"score":{"verdict":"too_early","hype":88,"reality":19,"gap":69,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":{"estimate":true,"lean":"too_close","gap_low":null,"gap_high":null,"rule":"early-2.0"},"changes":[{"date":"2026-10-22T21:22:53+00:00","text":"Repository created","source":"https://github.com/SheepSeb/tinyworldformer"}],"cite":"tinyworldformer is a GitHub project at https://github.com/SheepSeb/tinyworldformer written mainly in Rust, described there as \"Tiny world model generates games and stream ac plays them all on Pico 2W\".","attribution":"Data from Mid or Real, https://midorreal.com/project/sheepseb-tinyworldformer, checked 22 November 2026. Attribution with a link is required.","still_real":null}