Metadata-Version: 2.5
Name: vramfit
Version: 0.0.1
Summary: Fit large open models on a single GPU: per-layer quantization sensitivity, mixed-precision recipes under a VRAM budget. Name claim ahead of the first release.
Project-URL: Repository, https://github.com/Alberto-Codes/quantfit
Author-email: Alberto-Codes <alberto.codes.dev@gmail.com>
License: MIT
Classifier: Development Status :: 1 - Planning
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3 :: Only
Requires-Python: >=3.12
Description-Content-Type: text/markdown

# vramfit

Fit large open models on a single GPU.

vramfit measures per-layer quantization sensitivity, solves for a
mixed-precision recipe under a VRAM budget, validates the recipe
against the prediction, and packs the result for a target runtime.

This 0.0.1 release claims the name ahead of the first functional
release. The tool is real and public today under its former name:
<https://github.com/Alberto-Codes/quantfit>. The repository renames
to vramfit and the first functional release replaces this
placeholder shortly.
