Metadata-Version: 2.4
Name: metalforge
Version: 0.0.1a1
Summary: An MLX-native framework for pretraining 50M-3B parameter LLMs from scratch on Apple Silicon (early pre-release).
Author: The MetalForge Authors
License-Expression: Apache-2.0
License-File: LICENSE
Classifier: Development Status :: 1 - Planning
Classifier: Intended Audience :: Science/Research
Classifier: Operating System :: MacOS
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.12
Project-URL: Homepage, https://github.com/BhodiSea/MetalForge
Project-URL: Issues, https://github.com/BhodiSea/MetalForge/issues
Description-Content-Type: text/markdown

# MetalForge

An MLX-native framework for pretraining 50M–3B parameter LLMs from scratch on Apple Silicon, built and verified under an autonomous agent harness.

This is an early pre-release. Version 0.0.1a1 on PyPI installs the `metalforge` package and the `bake` and `metalforge` commands, which so far only print their version and help; the repository also holds the research behind the plan. The plan lives in the issues: the [roadmap index](https://github.com/BhodiSea/MetalForge/issues/195) lists all 194 steps in order, and [issue 1](https://github.com/BhodiSea/MetalForge/issues/1) is the first.
