Skip to content

Add option to set dtype in pipeline.to() method #2317

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
Mar 21, 2023
Merged

Add option to set dtype in pipeline.to() method #2317

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
11 changes: 8 additions & 3 deletions src/diffusers/pipelines/pipeline_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,8 +342,13 @@ def is_saveable_module(name, value):

save_method(os.path.join(save_directory, pipeline_component_name), **save_kwargs)

def to(self, torch_device: Optional[Union[str, torch.device]] = None, silence_dtype_warnings: bool = False):
if torch_device is None:
def to(
self,
torch_device: Optional[Union[str, torch.device]] = None,
torch_dtype: Optional[torch.dtype] = None,
silence_dtype_warnings: bool = False,
):
if torch_device is None and torch_dtype is None:
return self

# throw warning if pipeline is in "offloaded"-mode but user tries to manually set to GPU.
Expand Down Expand Up @@ -380,6 +385,7 @@ def module_is_offloaded(module):
for name in module_names.keys():
module = getattr(self, name)
if isinstance(module, torch.nn.Module):
module.to(torch_device, torch_dtype)
if (
module.dtype == torch.float16
and str(torch_device) in ["cpu"]
Expand All @@ -393,7 +399,6 @@ def module_is_offloaded(module):
" support for`float16` operations on this device in PyTorch. Please, remove the"
" `torch_dtype=torch.float16` argument, or use another device for inference."
)
module.to(torch_device)
return self

@property
Expand Down
17 changes: 13 additions & 4 deletions tests/test_pipelines_common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -344,11 +344,8 @@ def test_float16_inference(self):
pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)

for name, module in components.items():
if hasattr(module, "half"):
components[name] = module.half()
pipe_fp16 = self.pipeline_class(**components)
pipe_fp16.to(torch_device)
pipe_fp16.to(torch_device, torch.float16)
pipe_fp16.set_progress_bar_config(disable=None)

output = pipe(**self.get_dummy_inputs(torch_device))[0]
Expand Down Expand Up @@ -447,6 +444,18 @@ def test_to_device(self):
output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
self.assertTrue(np.isnan(output_cuda).sum() == 0)

def test_to_dtype(self):
components = self.get_dummy_components()
pipe = self.pipeline_class(**components)
pipe.set_progress_bar_config(disable=None)

model_dtypes = [component.dtype for component in components.values() if hasattr(component, "dtype")]
self.assertTrue(all(dtype == torch.float32 for dtype in model_dtypes))

pipe.to(torch_dtype=torch.float16)
model_dtypes = [component.dtype for component in components.values() if hasattr(component, "dtype")]
self.assertTrue(all(dtype == torch.float16 for dtype in model_dtypes))

def test_attention_slicing_forward_pass(self):
self._test_attention_slicing_forward_pass()

Expand Down