Sitelet https://github.com/mlx-node/mlx-node/pull/77/commits

feat(gemma4): load tied lm_head as packed-quantized (8-bit affine) - #77

Merged

Brooooooklyn merged 3 commits into

mainfrom

experiment/gemma4-lmhead8

Jun 25, 2026

Commits on Jun 25, 2026

feat(gemma4): load tied lm_head as packed-quantized instead of dense-dequant

Brooooooklyn
and
claude
committed
aceb554
fix(gemma4): force MX pack constants + validate mode/tensor consistency for packed embed

Brooooooklyn
and
claude
committed
d2af442
fix(gemma4): reject mxfp4/nvfp4 embed mis-resolved to mxfp8 (fail loud at load)

Brooooooklyn
and
claude
committed
9b34999