Metadata-Version: 2.3
Name: nn-chisel
Version: 0.0.0
Summary: Skill-targeted pruning toolkit for neural networks (placeholder release — real code lands soon).
Keywords: pruning,sparsity,sparsegpt,wanda,compression,neural-networks
Author: Julius Arkenberg
Author-email: Julius Arkenberg <julius@rkenberg.de>
License: Apache-2.0
Classifier: Development Status :: 1 - Planning
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Project-URL: Homepage, https://github.com/arki05/chisel
Project-URL: Source, https://github.com/arki05/chisel
Description-Content-Type: text/markdown

# nn-chisel

Skill-targeted pruning toolkit for neural networks.

This `0.0.0` release is a **placeholder** — it reserves the name on PyPI
while the real package is in development. Source lives at
https://github.com/arki05/chisel; the first functional release will be
`0.1.0`.

## What it's going to be

A small library for pruning large language models in a *skill-targeted*
way — calibrate the pruner on data representative of one capability and
measure how cleanly the resulting model preserves that capability versus
others. Includes SparseGPT- and Wanda-family methods with a composable
metric / corrector design.
