Tag
A compact 2.7M-parameter world model trained to play NES Tetris, demonstrating that very small models can capture enough dynamics to master the game.
TinyDecide is a 10M-parameter Jev-like model (~6MB) that outperforms larger models on the Decision Index leaderboard and runs in browsers, Node.js, Python, Rust, and even on ESP32 microcontrollers.
This technical report introduces Harness-Aware Training (HAT) to train compact AI agents that can evolve with changing configurations, improving performance and reducing latency for real-time interactions in live e-commerce digital avatars.