Search

Forge — LLM training from scratch in pure C++20 + Metal on Apple Silicon.

🔥 Forge LLM training from scratch — pure C++20 + Metal on Apple Silicon No PyTorch. No MLX. No ML dependencies. Just C++, Metal kernels, and a JSON parser. 📄 Paper · 🔬 Research notes · 📐 Spec ✨ What this is A complet…

…s. swap dealers) conditions the reaction of energy (crude oil, natural gas) and metals (gold, silver, copper, palladium) futures to standardized macro surprises. Key findings 🎯 🧯 Increased speculative trading dampens t…

…PU (LAPACK) or GPU (MLX) automatically. Large bootstrap runs execute as batched Metal solves; you never choose a backend. Reproducibility is a feature, not an accident. All randomness flows through a counter based Philox…

…on · zero third party dependencies ✨ What it is Forge is a from scratch C++20 + Metal LLM training framework: its own tensors, autograd, flash attention, Muon optimizer, MoE, QAT, and the zero copy weight format. Forge S…

…is never required. Two prerequisites beyond the Command Line Tools: 1. Apple's Metal Toolchain on — MLX compiles its GPU kernels at build time. On a machine with Xcode 26+: ; a standalone on also works. 2. SDK pin on ma…