SPB Git
spboucher.ai

Repository Public MIT

spbforge

History

Latest commit Add architecture-variant waves 1+2: train Mistral/Qwen/Gemma/OLMo-class models by config f7777e9

NameLast commitUpdated
..
gpt-100m.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
gpt-10m-1epoch.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
gpt-10m.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
gpt-200m-cluster.json 200M cluster run: one epoch at micro-batch 8
gpt-200m.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
gpt-25m.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon
gpt-50m-deep.json Add configurable training modes: Muon optimizer and WSD schedule
gpt-50m-gemma.json Add architecture-variant waves 1+2: train Mistral/Qwen/Gemma/OLMo-class models by config
gpt-50m-mistral.json Add architecture-variant waves 1+2: train Mistral/Qwen/Gemma/OLMo-class models by config
gpt-50m-moe.json Add QAT, MoE, and architecture-variant knobs — all config-selected
gpt-50m-muon.json Add configurable training modes: Muon optimizer and WSD schedule
gpt-50m-ternary.json Add QAT, MoE, and architecture-variant knobs — all config-selected
gpt-50m.json Add configurable training modes: Muon optimizer and WSD schedule
gpt-smoke.json Forge: LLM training from scratch in C++20 + Metal on Apple Silicon

Clone forge

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

Snapshot of 2e570b0640b23ceb0074540c3f682f6814e8d28d

ZIP tar.gz
Files 2e570b0640b23ceb0074540c3f682f6814e8d28d
Menu

Theme
↑↓navigate ↵open escclose /⌘Kanywhere