GitHub radar
7,200 Minecraft worlds trained on a single GPU with Netherite
A self-developed C/CUDA Minecraft simulator for large-scale reinforcement learning (RL) agent training.
A bit-to-bit reproduction of Minecraft 1.11.2 written in C/CUDA, which produces exactly the same frame outputs as its Java counterpart. Also included is a batched RL environment based on CUDA, enabling running 7,200 independent Minecraft instances concurrently on a single GPU card for efficient training of RL agents using a commodity gaming GPU.
Why a vibe-coder should care
Netherite enables you to perform large scale agent analysis with a single GPU card. RL, an essential method for creating modern AI, typically requires massive computing power.
▌ More finds