wangbo/ComfyUI-Manager
1
0
Fork 0
mirror of https://github.com/Comfy-Org/ComfyUI-Manager.git synced 2026-07-20 05:18:11 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Merge b066a920a1 into 619eae0cd6

Browse Source
This commit is contained in:
JWLHS 2026-07-11 18:58:43 +09:00 committed by GitHub
commit ea09316d50
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
custom-node-list.json
View File

@ -54236,6 +54236,18 @@
], ],
"install_type": "git-clone", "install_type": "git-clone",
"description": "Block Swap (RAM Offload) node for native ComfyUI MODEL - run DiT models larger than VRAM by streaming transformer blocks from system RAM." "description": "Block Swap (RAM Offload) node for native ComfyUI MODEL - run DiT models larger than VRAM by streaming transformer blocks from system RAM."
},
{
"author": "JWLHS",
"title": "TINT4 — INT4 Weight-Only Quantization via torchao",
"id": "tint4_xpu",
"reference": "https://github.com/JWLHS/ComfyUI-TINT4-XPU",
"files": [
"https://github.com/JWLHS/ComfyUI-TINT4-XPU"
],
"install_type": "git-clone",
"description": "INT4 weight-only quantization (W4A16) for diffusion models via torchao. Activations remain FP16 — zero quantization quality loss. Full LoRA support: standard + LoKr format compatible, multi-LoRA stack (≤5), IS_CHANGED fingerprint skip (0s repeat generation), SHA256 JSON disk cache (<1s hit), GPU pre-hook bake-in. Supports Intel Arc XPU, NVIDIA CUDA, AMD ROCm. Verified on Krea2 Turbo, Flux2 Klein 9B, Z-Image, Boogu. Includes model analysis CLI. Requires: pip install torchao>=0.17.0"
} }
] ]
} }