slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It uniquely combines a bit-perfect game replica with high-performance CUDA RL training, enabling RL research in a controlled yet realistic Minecraft environment without relying on Java.
Who it is for
- Reinforcement learning researchers
- Game AI developers
- Minecraft modders and reverse engineers
- CUDA programmers interested in game simulation
Use cases
- Training RL agents with batched CUDA rollouts in Minecraft
- Verifying game mechanics by comparing pixel output to the Java oracle
- Side-by-side visual comparison of game behavior (oracle vs netherite)
- Research on bit-level game reconstruction and RL integration
Strengths
- Bit-verified fidelity to Minecraft 1.11.2 against the real Java game
- Native CUDA acceleration for batched RL training
- One-command setup and build for Linux x86_64
- Shipped canonical test tape for reproducible pixel-level verification
Considerations
- Full functionality only on Linux x86_64; macOS limited to viewer/SSH, Windows not supported as build host
- Requires a legal Minecraft ownership and JDK 8
- No Mojang assets shipped; depends on user's own game installation
README quick start
netherite
From-scratch C/CUDA Minecraft 1.11.2 (bit-verified vs the real Java game) + batched CUDA RL.
Platforms
| Support | |
|---|---|
| Linux x86_64 | Full stack (build, CUDA train, Java oracle). Canonical: anvil. |
| macOS | Viewer / SSH only (Moonlight, mcwindow). No native game or CUDA train. |
| Windows | Not supported as a build host. |
No Mojang content is shipped. You need a legal Minecraft ownership and JDK 8.
Using an LLM on this repo
Open AGENTS.md (Claude also loads CLAUDE.md). Or paste:
Read AGENTS.md in this repo and follow it. Task:
Clean Linux box (one command)
bash scripts/setup_and_verify.sh # bootstrap + build + --quick sweep
bash scripts/setup_and_verify.sh --demo # + physics/pixel tape replay + SBS MP4
# -> demos/pixel_match_sbs.mp4 (oracle | magma side-by-side)
Prism is optional. Bootstrap uses ForgeGradle; details in docs/BOOTSTRAP.md.
Pixel demo uses the shipped canonical tape under c/magma/raster/verify/demo/.
Related repositories
Similar projects matched by category, topics, and programming language.
jamesob
Featured
local-llm
A comprehensive guide for building and configuring a high-end local machine to run state-of-the-art LLMs, with detailed hardware choices, BIOS tuning, and Docker-based model serving.
makecindy
Featured
cindy
Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.