split_model.json

{
  "format": "split",
  "components": [
    "void_pass1",
    "void_pass2"
  ],
  "quantized": true,
  "quantization_bits": 8,
  "quantization_group_size": 64,
  "recipe": "void-model",
  "source": "netflix/void-model",
  "license": "apache-2.0",
  "quantization_scope": "transformer Linear weights only",
  "links": [
    "void-model-mlx (inference): https://github.com/dgrauet/void-model-mlx",
    "VideoX-Fun-mlx (engine): https://github.com/dgrauet/VideoX-Fun-mlx"
  ],
  "usage_url": "https://github.com/dgrauet/void-model-mlx",
  "extra_links": [
    "Base model weights (q8): https://huggingface.co/dgrauet/CogVideoX-Fun-V1.5-5b-InP-mlx-q8",
    "bf16 variant: https://huggingface.co/dgrauet/void-model-mlx",
    "q4 variant: https://huggingface.co/dgrauet/void-model-mlx-q4"
  ],
  "cli_snippet": "python -m void_mlx.infer \\\n    --sample sample/BigBen \\\n    --pass1 weights/q8/void_pass1.safetensors \\\n    --pass2 weights/q8/void_pass2.safetensors \\\n    --base-model /path/to/CogVideoX-Fun-V1.5-5b-InP-mlx-q8 \\\n    --steps 30 --max-frames 13 --height 352 --width 624 \\\n    --output result.gif",
  "build_note": "Good quality/memory balance (~48 GB RAM recommended for the full two-pass\npipeline). On 32 GB Macs use the q4 variant instead."
}