mavis-ai/Gemma4-E2B-QAT-Q5

This repository contains an MLX-ready experimental QAT Q5 distribution of Google's Gemma 4 E2B instruction-tuned model, prepared for local inference on Apple Silicon Macs.

This is not a newly fine-tuned model and it is not an official Google QAT Q5 release. It is a quantized MLX redistribution converted from Google's official Quantization-Aware Training (QAT) unquantized checkpoint. No structural modifications or additional training have been applied to the model architecture.

Important Notice

This repository is hosted primarily as a dedicated engine source for the R.E.V.I.S. application ecosystem. You are free to download and use this model package for your own local MLX projects or workflows, subject to the Apache License 2.0 and Google's Gemma terms.

For the original model cards, architecture details, intended usage, limitations, and evaluation information, refer to the official upstream models:

• Base model: https://huggingface.co/google/gemma-4-E2B-it
• Official QAT source checkpoint: https://huggingface.co/google/gemma-4-E2B-it-qat-q4_0-unquantized
• Official QAT Q4 GGUF reference: https://huggingface.co/google/gemma-4-E2B-it-qat-q4_0-gguf

Quantization

This package uses MLX affine quantization with a mixed layout chosen to follow the tensor boundaries of Google's official QAT Q4 GGUF release as closely as practical in MLX, while raising the default quantized language tensors to Q5 for this experiment:

• Source checkpoint: google/gemma-4-E2B-it-qat-q4_0-unquantized
• Reference layout: google/gemma-4-E2B-it-qat-q4_0-gguf
• Default language tensor quantization: MLX affine Q5
• Group size: 64
• Mode: affine
• Token embeddings: MLX affine Q6
• Audio, vision, multimodal projection components, MoE routers where present, and other keep-list tensors: BF16
• Norms, biases, and small non-linear tensors: preserved in their source floating-point representation

In practical terms, this is not a plain "quantize every linear layer to Q5" package. The embedding table is intentionally kept at higher precision, and the multimodal / routing side of the model is not aggressively quantized. This keeps the MLX package close to the official Google QAT Q4 tensor policy while testing whether Q5 gives a useful quality-size tradeoff.

Optimized for R.E.V.I.S. (Local Cognitive OS)

We host and test this model package to serve as a local reasoning and judgment engine for R.E.V.I.S.

R.E.V.I.S. is a 100% local Cognitive OS for Multi-Agentic AI. It transforms your Mac devices into a distributed Agentic Swarm via zero-config Wi-Fi clustering, allowing you to run heavy AI workloads like recursive web research, dynamic RAG generation, and multi-step logic without killing single-machine performance.

If you are interested in pushing the limits of local AI and open-weight models, check out our project.

• Official Website: https://mavis-ai.co.jp/revis/
• Watch the 13-min Raw Demo (Multi-node Dynamic RAG): https://x.gd/LxaBF
• Follow our updates on X: https://x.com/mavis_ai_jp

Usage

Install or update the MLX runtime you use for Gemma 4 / multimodal models:

pip install -U mlx mlx-lm mlx-vlm huggingface_hub hf_xet

Download the model:

hf download mavis-ai/Gemma4-E2B-QAT-Q5 \
  --local-dir ~/Models/mlx/Gemma4-E2B-QAT-Q5

Run a quick text generation test:

python -m mlx_vlm.generate \
  --model ~/Models/mlx/Gemma4-E2B-QAT-Q5 \
  --max-tokens 256 \
  --temperature 0.0 \
  --prompt "日本語で短く自己紹介してください。"

Usage Notes

Gemma 4 E2B is the lightest member in this local set and is useful for repeated routing, extraction, and short judgment tasks where latency matters. This QAT Q5 build is experimental and should be compared against the corresponding QAT Q4 and regular Q5/Q6 builds before production use.

Files

Recommended repository files:

.gitattributes
LICENSE
README.md
chat_template.jinja
config.json
generation_config.json
model.safetensors
model.safetensors.index.json
processor_config.json
tokenizer.json
tokenizer_config.json

License

This repository redistributes a quantized derivative of Google's Gemma 4 instruction-tuned QAT release, which is distributed by Google under the Apache License 2.0.

This derivative is likewise distributed under the Apache License 2.0. A copy of the license is included in the LICENSE file in this repository, and can also be found at https://www.apache.org/licenses/LICENSE-2.0.

Modification Notice

Compared with the official Google QAT source checkpoint, this repository applies the following packaging modification:

The source checkpoint was converted to MLX format and quantized with a GGUF-like mixed Q5/Q6/BF16 policy for local MLX inference.

No fine-tuning, additional training, or architecture-level modification has been applied.