Skip to content

Cleanup vram after models offloading #3826

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
merged 2 commits into from
Jul 18, 2023
Merged

Cleanup vram after models offloading #3826

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bda0000
Cleanup vram after models offloading, tweak to cleanup local variable…
StAlKeR7779 Jul 18, 2023
186e98d
Merge branch 'main' into fix/mem_cleanup
blessedcoolant Jul 18, 2023
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
22 changes: 22 additions & 0 deletions invokeai/backend/model_management/model_cache.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -328,6 +328,25 @@ def _make_cache_room(self, model_size):

refs = sys.getrefcount(cache_entry.model)

# manualy clear local variable references of just finished function calls
# for some reason python don't want to collect it even by gc.collect() immidiately
if refs > 2:
while True:
cleared = False
for referrer in gc.get_referrers(cache_entry.model):
if type(referrer).__name__ == "frame":
# RuntimeError: cannot clear an executing frame
with suppress(RuntimeError):
referrer.clear()
cleared = True
#break

# repeat if referrers changes(due to frame clear), else exit loop
if cleared:
gc.collect()
else:
break

device = cache_entry.model.device if hasattr(cache_entry.model, "device") else None
self.logger.debug(f"Model: {model_key}, locks: {cache_entry._locks}, device: {device}, loaded: {cache_entry.loaded}, refs: {refs}")

Expand Down Expand Up @@ -363,6 +382,9 @@ def _offload_unlocked_models(self, size_needed: int=0):
self.logger.debug(f'GPU VRAM freed: {(mem.vram_used/GIG):.2f} GB')
vram_in_use += mem.vram_used # note vram_used is negative
self.logger.debug(f'{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB')

gc.collect()
torch.cuda.empty_cache()

def _local_model_hash(self, model_path: Union[str, Path]) -> str:
sha = hashlib.sha256()
Expand Down