Skip to content

fix: sdxl build oom #151

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 1 commit into from
Oct 28, 2025
Merged

fix: sdxl build oom #151

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion src/streamdiffusion/acceleration/tensorrt/builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,6 +60,7 @@ def build(
opt_batch_size=opt_batch_size,
onnx_opset=onnx_opset,
)
self.network = self.network.to("cpu")
del self.network
gc.collect()
torch.cuda.empty_cache()
Expand Down Expand Up @@ -89,7 +90,6 @@ def build(
build_all_tactics=build_all_tactics,
build_enable_refit=build_enable_refit,
)

for file in os.listdir(os.path.dirname(engine_path)):
if file.endswith('.engine'):
continue
Expand Down
2 changes: 1 addition & 1 deletion src/streamdiffusion/modules/controlnet_module.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -610,7 +610,7 @@ def _load_pytorch_controlnet_model(self, model_id: str, conditioning_channels: O
)
else:
controlnet = ControlNetModel.from_pretrained(model_id, **load_kwargs)
controlnet = controlnet.to(device=self.device, dtype=self.dtype)
controlnet = controlnet.to(dtype=self.dtype)
# Track model_id for updater diffing
try:
setattr(controlnet, 'model_id', model_id)
Expand Down
8 changes: 4 additions & 4 deletions src/streamdiffusion/wrapper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1041,9 +1041,9 @@ def _load_model(
traceback.print_exc()
raise RuntimeError(error_msg)
else:
if hasattr(pipe, "text_encoder") and pipe.text_encoder is not None:
if not compile_engines_only and hasattr(pipe, "text_encoder") and pipe.text_encoder is not None:
pipe.text_encoder = pipe.text_encoder.to(device=self.device)
if hasattr(pipe, "text_encoder_2") and pipe.text_encoder_2 is not None:
if not compile_engines_only and hasattr(pipe, "text_encoder_2") and pipe.text_encoder_2 is not None:
pipe.text_encoder_2 = pipe.text_encoder_2.to(device=self.device)

# If we get here, the model loaded successfully - break out of retry loop
Expand Down Expand Up @@ -1570,7 +1570,7 @@ def _load_model(
if self.use_safety_checker or safety_checker_engine_exists:
if not safety_checker_engine_exists:
from transformers import AutoModelForImageClassification
self.safety_checker = AutoModelForImageClassification.from_pretrained(safety_checker_model_id).to("cuda")
self.safety_checker = AutoModelForImageClassification.from_pretrained(safety_checker_model_id)

safety_checker_model = NSFWDetector(
device=self.device,
Expand All @@ -1585,7 +1585,7 @@ def _load_model(
model_config=safety_checker_model,
batch_size=self.batch_size if self.mode == "txt2img" else stream.frame_bff_size,
cuda_stream=None,
load_engine=load_engine,
load_engine=False,
)

if load_engine:
Expand Down