mps: remove warmup passes (huggingface#2771)

* Remove warmup passes in mps tests.

* Update mps docs: no warmup pass in PyTorch 2

* Update imports.

Author: pcuenca

docs/source/en/optimization/mps.mdx

@@ -19,17 +19,22 @@ specific language governing permissions and limitations under the License.
 - Mac computer with Apple silicon (M1/M2) hardware.
 - macOS 12.6 or later (13.0 or later recommended).
 - arm64 version of Python.
-- PyTorch 1.13. You can install it with `pip` or `conda` using the instructions in https://pytorch.org/get-started/locally/.
+- PyTorch 2.0 (recommended) or 1.13 (minimum version supported for `mps`). You can install it with `pip` or `conda` using the instructions in https://pytorch.org/get-started/locally/.
 
 
 ## Inference Pipeline
 
 The snippet below demonstrates how to use the `mps` backend using the familiar `to()` interface to move the Stable Diffusion pipeline to your M1 or M2 device.
 
-We recommend to "prime" the pipeline using an additional one-time pass through it. This is a temporary workaround for a weird issue we have detected: the first inference pass produces slightly different results than subsequent ones. You only need to do this pass once, and it's ok to use just one inference step and discard the result.
+<Tip warning={true}>
+
+**If you are using PyTorch 1.13** you need to "prime" the pipeline using an additional one-time pass through it. This is a temporary workaround for a weird issue we detected: the first inference pass produces slightly different results than subsequent ones. You only need to do this pass once, and it's ok to use just one inference step and discard the result.
+
+</Tip>
+
+We strongly recommend you use PyTorch 2 or better, as it solves a number of problems like the one described in the previous tip.
 
 ```python
-# make sure you're logged in with `huggingface-cli login`
 from diffusers import StableDiffusionPipeline
 
 pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
@@ -40,7 +45,7 @@ pipe.enable_attention_slicing()
 
 prompt = "a photo of an astronaut riding a horse on mars"
 
-# First-time "warmup" pass (see explanation above)
+# First-time "warmup" pass if PyTorch version is 1.13 (see explanation above)
 _ = pipe(prompt, num_inference_steps=1)
 
 # Results match those from the CPU device after the warmup pass.
@@ -59,5 +64,4 @@ pipeline.enable_attention_slicing()
 
 ## Known Issues
 
-- As mentioned above, we are investigating a strange [first-time inference issue](https://github.com/huggingface/diffusers/issues/372).
 - Generating multiple prompts in a batch [crashes or doesn't work reliably](https://github.com/huggingface/diffusers/issues/363). We believe this is related to the [`mps` backend in PyTorch](https://github.com/pytorch/pytorch/issues/84039). This is being resolved, but for now we recommend to iterate instead of batching.

tests/models/test_models_vae.py

@@ -20,7 +20,6 @@
 from parameterized import parameterized
 
 from diffusers import AutoencoderKL
-from diffusers.models import ModelMixin
 from diffusers.utils import floats_tensor, load_hf_numpy, require_torch_gpu, slow, torch_all_close, torch_device
 
 from ..test_modeling_common import ModelTesterMixin
@@ -124,12 +123,7 @@ def test_output_pretrained(self):
         model = model.to(torch_device)
         model.eval()
 
-        # One-time warmup pass (see #372)
-        if torch_device == "mps" and isinstance(model, ModelMixin):
-            image = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
-            image = image.to(torch_device)
-            with torch.no_grad():
-                _ = model(image, sample_posterior=True).sample
+        if torch_device == "mps":
             generator = torch.manual_seed(0)
         else:
             generator = torch.Generator(device=torch_device).manual_seed(0)

tests/models/test_models_vq.py

@@ -85,9 +85,6 @@ def test_output_pretrained(self):
         image = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
         image = image.to(torch_device)
         with torch.no_grad():
-            # Warmup pass when using mps (see #372)
-            if torch_device == "mps":
-                _ = model(image)
             output = model(image).sample
 
         output_slice = output[0, -1, -3:, -3:].flatten().cpu()

tests/pipelines/ddpm/test_ddpm.py

@@ -74,10 +74,6 @@ def test_inference_predict_sample(self):
         ddpm.to(torch_device)
         ddpm.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = ddpm(num_inference_steps=1)
-
         generator = torch.manual_seed(0)
         image = ddpm(generator=generator, num_inference_steps=2, output_type="numpy").images
 

tests/pipelines/latent_diffusion/test_latent_diffusion_uncond.py

@@ -79,11 +79,6 @@ def test_inference_uncond(self):
         ldm.to(torch_device)
         ldm.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            generator = torch.manual_seed(0)
-            _ = ldm(generator=generator, num_inference_steps=1, output_type="numpy").images
-
         generator = torch.manual_seed(0)
         image = ldm(generator=generator, num_inference_steps=2, output_type="numpy").images
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py

@@ -265,10 +265,6 @@ def test_dict_tuple_outputs_equivalent(self):
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = pipe(**self.get_dummy_inputs(torch_device))
-
         output = pipe(**self.get_dummy_inputs(torch_device))[0]
         output_tuple = pipe(**self.get_dummy_inputs(torch_device), return_dict=False)[0]
 

tests/test_modeling_common.py

@@ -24,7 +24,7 @@
 import torch
 from requests.exceptions import HTTPError
 
-from diffusers.models import ModelMixin, UNet2DConditionModel
+from diffusers.models import UNet2DConditionModel
 from diffusers.models.attention_processor import AttnProcessor
 from diffusers.training_utils import EMAModel
 from diffusers.utils import torch_device
@@ -119,11 +119,6 @@ def test_from_save_pretrained(self):
             new_model.to(torch_device)
 
         with torch.no_grad():
-            # Warmup pass when using mps (see #372)
-            if torch_device == "mps" and isinstance(model, ModelMixin):
-                _ = model(**self.dummy_input)
-                _ = new_model(**self.dummy_input)
-
             image = model(**inputs_dict)
             if isinstance(image, dict):
                 image = image.sample
@@ -161,11 +156,6 @@ def test_from_save_pretrained_variant(self):
             new_model.to(torch_device)
 
         with torch.no_grad():
-            # Warmup pass when using mps (see #372)
-            if torch_device == "mps" and isinstance(model, ModelMixin):
-                _ = model(**self.dummy_input)
-                _ = new_model(**self.dummy_input)
-
             image = model(**inputs_dict)
             if isinstance(image, dict):
                 image = image.sample
@@ -203,10 +193,6 @@ def test_determinism(self):
         model.eval()
 
         with torch.no_grad():
-            # Warmup pass when using mps (see #372)
-            if torch_device == "mps" and isinstance(model, ModelMixin):
-                model(**self.dummy_input)
-
             first = model(**inputs_dict)
             if isinstance(first, dict):
                 first = first.sample
@@ -377,10 +363,6 @@ def recursive_check(tuple_object, dict_object):
         model.eval()
 
         with torch.no_grad():
-            # Warmup pass when using mps (see #372)
-            if torch_device == "mps" and isinstance(model, ModelMixin):
-                model(**self.dummy_input)
-
             outputs_dict = model(**inputs_dict)
             outputs_tuple = model(**inputs_dict, return_dict=False)
 

tests/test_pipelines_common.py

@@ -121,10 +121,6 @@ def test_save_load_local(self):
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = pipe(**self.get_dummy_inputs(torch_device))
-
         inputs = self.get_dummy_inputs(torch_device)
         output = pipe(**inputs)[0]
 
@@ -327,10 +323,6 @@ def test_dict_tuple_outputs_equivalent(self):
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = pipe(**self.get_dummy_inputs(torch_device))
-
         output = pipe(**self.get_dummy_inputs(torch_device))[0]
         output_tuple = pipe(**self.get_dummy_inputs(torch_device), return_dict=False)[0]
 
@@ -402,10 +394,6 @@ def test_save_load_optional_components(self):
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = pipe(**self.get_dummy_inputs(torch_device))
-
         # set all optional components to None
         for optional_component in pipe._optional_components:
             setattr(pipe, optional_component, None)
@@ -477,10 +465,6 @@ def _test_attention_slicing_forward_pass(
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        # Warmup pass when using mps (see #372)
-        if torch_device == "mps":
-            _ = pipe(**self.get_dummy_inputs(torch_device))
-
         inputs = self.get_dummy_inputs(torch_device)
         output_without_slicing = pipe(**inputs)[0]