wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-02-08 12:32:32 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

mm: Dont GPU load models

Browse Source 
Aimdo will do this on demand as 0 copy. Remove the special case for
vram > ram.
This commit is contained in:
Rattus 2026-01-27 14:10:54 +10:00
parent db99ab48c2
commit 26dc3a20c6
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
comfy/model_management.py
View File

@ -823,7 +823,7 @@ def unet_inital_load_device(parameters, dtype):
mem_dev = get_free_memory(torch_dev) mem_dev = get_free_memory(torch_dev)
mem_cpu = get_free_memory(cpu_dev) mem_cpu = get_free_memory(cpu_dev)
if mem_dev > mem_cpu and model_size < mem_dev: if mem_dev > mem_cpu and model_size < mem_dev and comfy.memory_management.aimdo_allocator is None:
return torch_dev return torch_dev
else: else:
return cpu_dev return cpu_dev