CoolFace
Modelpublic

JANGQ-AI/gemma-4-12B-it-qat-JANG_4M

sourceHugging Facegemmaupdated 19d agoView on Hugging Face
0likes82downloads
Model Card

<p align="center"><img src="./jangq-logo-dark.png" alt="JANGQ AI" width="180"></p>

<p align="center"> <a href="https://vmlx.net"><img src="vmlx-app.png" alt="vMLX — run JANG models on Apple Silicon" width="820"></a> </p>

<h3 align="center">⚡ All JANG models are meant to be run in <a href="https://vmlx.net">vMLX</a></h3>

JANGQ-AI/gemma-4-12B-it-qat-JANG_4M

JANG4M MLX affine bundle converted from [google/gemma-4-12B-it-qat-q40-unquantized](https://huggingface.co/google/gemma-4-12B-it-qat-q4_0-unquantized).

This bundle keeps Gemma 4 bookends and media-sensitive components coherent: token embeddings/output projection, norms, media towers/embedders, and Gemma 4 per-layer embedding/gate/projector tensors are fp16 passthrough; self-attention and MoE router projections are 8-bit affine; decoder MLP/expert bulk is 4-bit affine.

Bundle

FieldValue
Sourcegoogle/gemma-4-12B-it-qat-q4_0-unquantized
Architecturegemma4_unified / Gemma4UnifiedForConditionalGeneration
Text layers48
Hidden size3840
Weight formatjang_affine
Top-level quantizationbits=8, group_size=32, mode=affine
Tier bitsattention=8, router=8, mlp=4, embed=16, perlayermedia=16
Quantized modules328 affine bases with .scales and .biases sidecars
Shards10 safetensors shards
Preserved source modalitiestext, vision, audio
Verified current vMLX runtimetext-only

Modalities

ModalityStatus
Textverified through the vMLX JANG text loader
Visionsource config and weights are present and preserved; standard mlx_vlm does not yet provide gemma4_unified runtime support
Audioaudio_config and audio embedder weights are present and preserved; standard mlx_vlm does not yet provide gemma4_unified runtime support
VideoNo video_config is present, so this card does not claim a verified video runtime path.

Runtime support depends on a Gemma 4 compatible MLX/vMLX loader that understands config.json quantization overrides, jang_config.json, and Gemma 4 processor/chat-template files. The current verified path is text-only through vMLX's JANG loader; vision/audio require a real gemma4_unified VLM implementation.

Runtime Metadata

  • —config.json has source-derived has_vision, has_audio, has_video, modalities, and capabilities.
  • —tokenizer_config.json includes bos_token_id, eos_token_id, pad_token_id, and the patched Gemma 4 chat template.
  • —processor_config.json is preserved for Gemma 4 multimodal processing.
  • —MTP/speculative drafter weights are not present in the source checkpoint; metadata is mtp: none / mtp_policy: none.