Modelpublic

OsaurusAI/gemma-4-31B-it-qat-JANG_4M

sourceHugging Facegemmaupdated 27d agoView on Hugging Face

0likes162downloads

Model Card

OsaurusAI/gemma-4-31B-it-qat-JANG_4M

JANG4M MLX affine bundle converted from [google/gemma-4-31B-it-qat-q40-unquantized](https://huggingface.co/google/gemma-4-31B-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

Field	Value
Source	`google/gemma-4-31B-it-qat-q4_0-unquantized`
Architecture	`gemma4` / `Gemma4ForConditionalGeneration`
Text layers	60
Hidden size	5376
Weight format	`jang_affine`
Top-level quantization	bits=8, group_size=32, mode=affine
Tier bits	attention=8, router=8, mlp=4, embed=16, perlayermedia=16
Quantized modules	410 affine bases with `.scales` and `.biases` sidecars
Shards	25 safetensors shards
Capabilities	`vision`

Modalities

Modality	Status
Text	supported
Vision	source config present and preserved
Audio	`audio_config` is empty/not active for this variant; this card does not claim audio support.
Video	No `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.

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.