assistant/mtplx_artifact.json

{
  "note": "Converted locally with MTPLX Gemma4 assistant classes because stock mlx_lm.convert does not support model_type=gemma4_assistant.",
  "precision_policy": "Q8 affine G64 for all quantizable assistant modules, including tied embedding/LM-head path and projections.",
  "quantization": {
    "bits": 8,
    "group_size": 64,
    "mode": "affine"
  },
  "source_path": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-bf16-mlx",
  "source_repo": "google/gemma-4-31B-it-assistant",
  "source_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204"
}