onlyqt-gemma4-e4b-mlx-4bit

Google Gemma 4 E4B (text) converted for MLX and quantized to 4 bit, hosted here so the iOS app 온니큐티 (OnlyQT) can download it on first launch. The app runs the model fully on device.

This repository contains no fine-tuning — the weights are a straight re-quantization of Google's released model.

Contents

Files live under base_gemma4_e4b_mlx_q4/. manifest.json lists every file with its exact byte size; the app uses it to download and to resume safely.

File	Bytes
model.safetensors	4,198,905,174
tokenizer.json	32,169,626
model.safetensors.index.json	97,889
chat_template.jinja	16,804
config.json	2,914
tokenizer_config.json	2,773
generation_config.json	215
Total	4,231,195,395

Quantization

affine, 4 bit, group size 64. 42 layers, 18 of them KV-shared.

License

Gemma. Use of these weights is governed by the Gemma Terms of Use and the Gemma Prohibited Use Policy. By downloading them you agree to those terms.