Skip to content

Only unpin tensor if it was pinned by ComfyUI #10677

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
comfyanonymous merged 1 commit into from
Nov 7, 2025
Merged

Only unpin tensor if it was pinned by ComfyUI #10677

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 10 additions & 5 deletions comfy/model_management.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1129,13 +1129,18 @@ def unpin_memory(tensor):
if not is_device_cpu(tensor.device):
return False

if not tensor.is_pinned():
#NOTE: Cuda does detect when a tensor is already pinned and would
#error below, but there are proven cases where this also queues an error
#on the GPU async. So dont trust the CUDA API and guard here
ptr = tensor.data_ptr()
size = tensor.numel() * tensor.element_size()

size_stored = PINNED_MEMORY.get(ptr, None)
if size_stored is None:
logging.warning("Tried to unpin tensor not pinned by ComfyUI")
return False

if size != size_stored:
logging.warning("Size of pinned tensor changed")
return False

ptr = tensor.data_ptr()
if torch.cuda.cudart().cudaHostUnregister(ptr) == 0:
TOTAL_PINNED_MEMORY -= PINNED_MEMORY.pop(ptr)
if len(PINNED_MEMORY) == 0:
Expand Down