First-party MLX export of google/gemma-4-E2B-it, quantized to int4 (group size 64) for on-device use on Apple Silicon. Built by our own pipeline via mlx_lm.convert (mlx_lm >= 0.31.3).


Base	Gemma 4 E2B (Effective-2BB; Per-Layer Embeddings + cross-layer KV sharing)
Quantization	int4, group size 64
Runtime	MLX (`mlx_lm`)

Note: the source repo is multimodal (model_type: gemma4, vision + audio towers); this is the full model quantized. The conversion drops the 60 redundant k/v weights stored for the KV-shared layers (strict=False) — the architecture shares that KV, so those weights are unused.

License & attribution

Weights: derivative of Google's Gemma 4 — governed by the Gemma Terms of Use. Use is subject to Google's Prohibited Use Policy.
Conversion: mlx_lm.convert (Apple MLX) — MIT.

aufklarer/gemma-4-E2B-it-MLX-4bit

gemma-4-E2B-it-MLX-4bit

License & attribution

aufklarer/
gemma-4-E2B-it-MLX-4bit