Skip to content

[tests] cache non lora pipeline outputs. #12298

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 26 commits into from
Oct 1, 2025
Merged

[tests] cache non lora pipeline outputs. #12298

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
26 commits
Select commit Hold shift + click to select a range
02fd92e
cache non lora pipeline outputs.
sayakpaul Sep 8, 2025
c8afd1c
up
sayakpaul Sep 8, 2025
6c0c72d
up
sayakpaul Sep 8, 2025
4256de9
up
sayakpaul Sep 8, 2025
772c32e
up
sayakpaul Sep 10, 2025
2c47a2f
Revert "up"
sayakpaul Sep 10, 2025
9c24d1f
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 11, 2025
cca03df
up
sayakpaul Sep 11, 2025
53ca186
Revert "up"
sayakpaul Sep 11, 2025
336efbd
up
sayakpaul Sep 11, 2025
69f2d5c
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 11, 2025
9d3f707
up
sayakpaul Sep 11, 2025
4923986
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 23, 2025
34d0aa2
resolve big conflicts.
sayakpaul Sep 28, 2025
1569fca
add .
sayakpaul Sep 28, 2025
8f405ed
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 29, 2025
ead2e04
up
sayakpaul Sep 29, 2025
fbcdf8b
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 29, 2025
8efb5c4
up
sayakpaul Sep 29, 2025
ca424e5
up
sayakpaul Sep 29, 2025
3e0a7f9
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 30, 2025
e442df0
up
sayakpaul Sep 30, 2025
dcfd979
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Sep 30, 2025
734b500
up
sayakpaul Sep 30, 2025
b01bf8e
up
sayakpaul Oct 1, 2025
c16d892
Merge branch 'main' into cache-non-lora-outputs
sayakpaul Oct 1, 2025
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
1 change: 1 addition & 0 deletions tests/lora/test_lora_layers_flux.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -331,6 +331,7 @@ def get_dummy_inputs(self, with_generator=True):
noise = floats_tensor((batch_size, num_channels) + sizes)
input_ids = torch.randint(1, sequence_length, size=(batch_size, sequence_length), generator=generator)

np.random.seed(0)
pipeline_inputs = {
"prompt": "A painting of a squirrel eating a burger",
"control_image": Image.fromarray(np.random.randint(0, 255, size=(32, 32, 3), dtype="uint8")),
Expand Down
84 changes: 53 additions & 31 deletions tests/lora/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -129,6 +129,34 @@ class PeftLoraLoaderMixinTests:
text_encoder_target_modules = ["q_proj", "k_proj", "v_proj", "out_proj"]
denoiser_target_modules = ["to_q", "to_k", "to_v", "to_out.0"]

cached_non_lora_outputs = {}

@pytest.fixture(scope="class", autouse=True)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Think this might be a leftover from previous refactoring. It's not used anywhere. Think it can be removed.

def cache_non_lora_outputs(self):
"""
This fixture will be executed once per test class and will populate
the cached_non_lora_outputs dictionary.
"""
for scheduler_cls in self.scheduler_classes:
# Check if the output for this scheduler is already cached to avoid re-running
if scheduler_cls.__name__ in self.cached_non_lora_outputs:
continue

components, _, _ = self.get_dummy_components(scheduler_cls)
pipe = self.pipeline_class(**components)
pipe = pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)

# Always ensure the inputs are without the `generator`. Make sure to pass the `generator`
# explicitly.
_, _, inputs = self.get_dummy_inputs(with_generator=False)
output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
self.cached_non_lora_outputs[scheduler_cls.__name__] = output_no_lora

# Ensures that there's no inconsistency when reusing the cache.
yield
self.cached_non_lora_outputs.clear()

def get_dummy_components(self, scheduler_cls=None, use_dora=False, lora_alpha=None):
if self.unet_kwargs and self.transformer_kwargs:
raise ValueError("Both `unet_kwargs` and `transformer_kwargs` cannot be specified.")
Expand Down Expand Up @@ -320,13 +348,7 @@ def test_simple_inference(self):
Tests a simple inference and makes sure it works as expected
"""
for scheduler_cls in self.scheduler_classes:
components, text_lora_config, _ = self.get_dummy_components(scheduler_cls)
pipe = self.pipeline_class(**components)
pipe = pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)

_, _, inputs = self.get_dummy_inputs()
output_no_lora = pipe(**inputs)[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

def test_simple_inference_with_text_lora(self):
Expand All @@ -341,7 +363,7 @@ def test_simple_inference_with_text_lora(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -424,7 +446,7 @@ def test_low_cpu_mem_usage_with_loading(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -480,7 +502,7 @@ def test_simple_inference_with_text_lora_and_scale(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -518,7 +540,7 @@ def test_simple_inference_with_text_lora_fused(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -550,7 +572,7 @@ def test_simple_inference_with_text_lora_unloaded(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -585,7 +607,7 @@ def test_simple_inference_with_text_lora_save_load(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -636,7 +658,7 @@ def test_simple_inference_with_partial_text_lora(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -687,7 +709,7 @@ def test_simple_inference_save_pretrained_with_text_lora(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config=None)
Expand Down Expand Up @@ -730,7 +752,7 @@ def test_simple_inference_with_text_denoiser_lora_save_load(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -771,7 +793,7 @@ def test_simple_inference_with_text_denoiser_lora_and_scale(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -815,7 +837,7 @@ def test_simple_inference_with_text_lora_denoiser_fused(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, denoiser = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -853,7 +875,7 @@ def test_simple_inference_with_text_denoiser_lora_unloaded(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, denoiser = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -932,7 +954,7 @@ def test_simple_inference_with_text_denoiser_multi_adapter(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
Expand Down Expand Up @@ -1061,7 +1083,7 @@ def test_simple_inference_with_text_denoiser_block_scale(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
self.assertTrue(check_if_lora_correctly_set(pipe.text_encoder), "Lora not correctly set in text encoder")
Expand Down Expand Up @@ -1118,7 +1140,7 @@ def test_simple_inference_with_text_denoiser_multi_adapter_block_lora(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
Expand Down Expand Up @@ -1281,7 +1303,7 @@ def test_simple_inference_with_text_denoiser_multi_adapter_delete_adapter(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
Expand Down Expand Up @@ -1375,7 +1397,7 @@ def test_simple_inference_with_text_denoiser_multi_adapter_weighted(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
Expand Down Expand Up @@ -1619,7 +1641,7 @@ def test_simple_inference_with_text_lora_denoiser_fused_multi(
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
Expand Down Expand Up @@ -1700,7 +1722,7 @@ def test_lora_scale_kwargs_match_fusion(self, expected_atol: float = 1e-3, expec
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
Expand Down Expand Up @@ -1755,7 +1777,7 @@ def test_simple_inference_with_dora(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_dora_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_dora_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_dora_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -1887,7 +1909,7 @@ def test_logs_info_when_no_lora_keys_found(self):
pipe.set_progress_bar_config(disable=None)

_, _, inputs = self.get_dummy_inputs(with_generator=False)
original_out = pipe(**inputs, generator=torch.manual_seed(0))[0]
original_out = self.cached_non_lora_outputs[scheduler_cls.__name__]

no_op_state_dict = {"lora_foo": torch.tensor(2.0), "lora_bar": torch.tensor(3.0)}
logger = logging.get_logger("diffusers.loaders.peft")
Expand Down Expand Up @@ -1933,7 +1955,7 @@ def test_set_adapters_match_attention_kwargs(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(pipe, text_lora_config, denoiser_lora_config)
Expand Down Expand Up @@ -2287,7 +2309,7 @@ def test_lora_adapter_metadata_save_load_inference(self, lora_alpha):
pipe = self.pipeline_class(**components).to(torch_device)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]
self.assertTrue(output_no_lora.shape == self.output_shape)

pipe, _ = self.add_adapters_to_pipeline(
Expand Down Expand Up @@ -2337,7 +2359,7 @@ def test_inference_load_delete_load_adapters(self):
pipe.set_progress_bar_config(disable=None)
_, _, inputs = self.get_dummy_inputs(with_generator=False)

output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
output_no_lora = self.cached_non_lora_outputs[scheduler_cls.__name__]

if "text_encoder" in self.pipeline_class._lora_loadable_modules:
pipe.text_encoder.add_adapter(text_lora_config)
Expand Down
Loading