File convert.py

GQLA_preprint/src/convert.py:None–None · view source on GitHub ↗

Source from the content-addressed store, hash-verified

1	#!/usr/bin/env python3
2	"""Convert GLM-4.7-Flash (MLA) to GQLA via per-group K/V PCA + absorption.
3
4	Data-driven: a single MLA forward over the calibration set captures every

nothing calls this directly

mainFunction · 0.70

no test coverage detected