Elliot Arledge

cuda-course. Cuda

fcc-intro-to-llms. Jupyter Notebook

mnist-cuda. Cuda

x-cli. CLI for X/Twitter API v2 -- post, search, like, bookmark from your terminal

all-of-it. Jupyter Notebook

metal-mc-terrain. Native Metal terrain renderer for Minecraft 1.16.5 on Apple Silicon. 10x FPS by replacing macOS OpenGL with Metal.

MegaQwen. Qwen3-0.6B megakernel: 527 tok/s decode on RTX 3090 (3.8x faster than PyTorch)

docs.md.

x-mcp. MCP server for the X (Twitter) API -- give AI agents the ability to post, search, read, and engage on X

driftin. Single-step image generation at 306 FPS. Drifting vs Diffusion head-to-head on CIFAR-10.

txtscrpr. Python

batmobile. High-performance CUDA kernels for equivariant graph neural networks (MACE, NequIP, Allegro). 10-20x faster than e3nn.

intro-to-mojo. Jupyter Notebook

grassmann-flows. Reproduction study of Grassmann Flows for sequence modeling (arXiv 2512.19428). Shows 22.6% gap vs claimed 10-15%, includes CUDA kernels with 2x speedup.

prompts.

mine.cu. High-performance batched voxel RL environment with custom CUDA kernels

physics-llm-inference. Companion code for The Physics of LLM Inference book

grokking-megakernels. Companion code for Grokking Megakernels: fuse an entire LLM forward pass into a single CUDA kernel

KernelBench-v3. Python

timelapse-maker. Python

book.cu. Cuda

rl-handbook. Code companion for the RL Post-Training Handbook - training reasoning models on a single GPU

emrakul. Agent orchestration framework - delegate to Cursor/Codex/Kimi/OpenCode without burning Claude quota

mc-labeller. Java

moltbook-tutorial. Beginner guide for adding AI bots to Moltbook - the social network for agents

pre-training-for-dummies. Train GPT-2 from scratch on a single GPU. Companion code for the Pre-Training from Scratch book.

KernelBench-v3. Python

dendrite. GPU-accelerated CUDA kernels for lithium-ion battery simulation. 133x faster than NumPy.

excalidraw.py. Render Excalidraw files to PNG with proper text rendering

CLAUDE.md. Claude Code CLI configuration dotfiles

exllama-ish. Apple Silicon port of exllamav3 - Efficient 4-bit quantized LLM inference using Metal and MLX

giga. TypeScript

llama-va. Python

nanoNext. Python

thinkingcap. Multi-agent research MCP server with parallel LLM providers

reasoning.

mnist.rs. Rust

nanoDSV3.2. Python

gpuup.sh. Shell

nvim. My NeoVim configuration (0.11+)

matmul-perf. Matrix multiplication performance benchmarks in Rust - comparing naive, parallel, blocked, and optimized implementations

spec-dec. Python

nanoMuon. Python

rusteze. Rust

vpt. Python

MetalMarlin. Marlin-style 4-bit quantized LLM inference kernels for Apple Metal

rusteze. High-performance chat platform. Rust backend (PostgreSQL + Redis), Rust-native client.