dgrauet commited on
Commit
29ee4d4
·
verified ·
1 Parent(s): bce7de6

Rename the build note key so it cannot collide with the per-component notes table

Browse files
Files changed (1) hide show
  1. split_model.json +1 -1
split_model.json CHANGED
@@ -22,5 +22,5 @@
22
  "q8 variant: https://huggingface.co/dgrauet/void-model-mlx-q8"
23
  ],
24
  "cli_snippet": "python -m void_mlx.infer \\\n --sample sample/BigBen \\\n --pass1 weights/q4/void_pass1.safetensors \\\n --pass2 weights/q4/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",
25
- "notes": "**This is the 32 GB configuration**: paired with the q8 base model, a full\ntwo-pass BigBen run (30 steps, 13 frames, 352\u00d7624) peaks at ~23.7 GB \u2014\nunder the 26.8 GB recommended working set of a 32 GB Apple Silicon Mac.\nQuality vs the bf16 weights: PSNR \u2248 35.5 dB on the same seed."
26
  }
 
22
  "q8 variant: https://huggingface.co/dgrauet/void-model-mlx-q8"
23
  ],
24
  "cli_snippet": "python -m void_mlx.infer \\\n --sample sample/BigBen \\\n --pass1 weights/q4/void_pass1.safetensors \\\n --pass2 weights/q4/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",
25
+ "build_note": "**This is the 32 GB configuration**: paired with the q8 base model, a full\ntwo-pass BigBen run (30 steps, 13 frames, 352\u00d7624) peaks at ~23.7 GB \u2014\nunder the 26.8 GB recommended working set of a 32 GB Apple Silicon Mac.\nQuality vs the bf16 weights: PSNR \u2248 35.5 dB on the same seed."
26
  }