SPB Git
spboucher.ai

spb/forge

Public MIT

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

llmtrainingcppmetalapple-silicon
History
NameLast commitUpdated
configs Add architecture-variant waves 1+2: train Mistral/Qwen/Gemma/OLMo-class models...
models Add wave 3: DeepSeek-V3-style MoE routing, all config-selected
paper Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
src Add elapsed_s column to log.csv for structured metrics consumers
tests Add wave 3: DeepSeek-V3-style MoE routing, all config-selected
third_party Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
tools Add HF streaming data pipeline and three research reports
.gitignore Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
ARCHITECTURES.md Add wave 3: DeepSeek-V3-style MoE routing, all config-selected
CLAUDE.md Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
CMakeLists.txt Add .forge — Apple-native, git-style weight format with zero-copy loading
INFERENCE_RESEARCH.md Add HF streaming data pipeline and three research reports
LICENSE Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
README.md Add HF streaming data pipeline and three research reports
RESEARCH.md Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
SMALL_MODELS_RESEARCH.md Add HF streaming data pipeline and three research reports

Clone forge

Anonymous clone and fetch. Pushing requires the owner's personal access token.

ZIP tar.gz snapshot of 147560ec423d00ad5f58daaf762e57546c819b00
Files · 147560ec423d00ad5f58daaf762e57546c819b00
Menu

Theme
↑↓ navigate ↵ open esc close / or ⌘K anywhere