AI/ML●●Solid
Gemma 3 inference in pure C++ with Metal acceleration
Pure C++23 Metal inference when llama.cpp already dominates this space.
WizardryNiche Gem
ybubnov
621mo ago
Back to browse
GitHub Repository
Yet Another Llama 3.2 implementation (in pure numpy)
5 starsPython
MinLlama – Llama 3.2 inference in ~100 lines of NumPy
by timothygao·Jun 23, 2026·2 points·0 comments
AI Analysis
●●SolidCozyNiche Gem
Pure NumPy Llama 3.2 inference in 100 lines for hacking KV cache compression.
Strengths
- •Pure NumPy implementation makes every tensor operation visible and hackable for research.
- •Includes PyTorch and Jax variants with statically-shaped KV cache streaming.
- •Clear setup scripts using uv for dependency management across all backends.
Weaknesses
- •Many minimal transformer implementations already exist via llama2.c and tinygrad.
- •Requires downloading the gated HuggingFace weights before running any code locally.
Category
Target Audience
ML researchers, students, hobbyists
Similar To
llama2.c · minGPT · tinygrad
Post Description
I built minLlama because I wanted a Llama implementation that was easy to understand and hack for KV cache compression research. There is also a PyTorch and Jax version in ~140 lines.
Would be interested in feedback from people who have written transformer implementations before, are there any implementation "tricks" that I'm missing (e.g, cleaner KV cache for PyTorch/Jax or rope tricks)?
Similar Projects
AI/ML●●●Banger
Luna Agent – Custom AI agent in ~2300 lines of Python, no frameworks
Custom agent framework in 2300 lines beats 400K-line bloatware; auditable and runs fully local.
Big BrainWizardryNiche Gem
nonatofabio
105mo ago
Developer Tools●●Solid
MicroECS – entity component system library in Python/NumPy
NumPy-backed ECS with archetype pooling beats OOP inheritance for Python game loops.
Big BrainNiche Gem
meehai
501mo ago
AI/ML●●Solid
100cc - Roll your own Claude in 100 lines
Self-bootstrapping agent writes its own improvements in 100 lines of TypeScript.
Big BrainCozy
rapiz
1042mo ago
AI/ML●●●Banger
GPT-2 inference in pure C#, 0 bytes allocated per token
GPT-2 inference in pure C# allocating zero bytes per token beats ONNX Runtime.
WizardryBig Brain
dev-on-bike
112mo ago
Education●Mid
Fast Chladni figure simulation in Python with NumPy vectorization
Beautiful physics visualization, but Chladni simulations exist in MATLAB, Mathematica, and other interactive platforms.
Rabbit HoleBig Brain
ratwolf
415mo ago