[scheduler] support custom timesteps and sigmas (#7817)

* support custom sigmas and timesteps, dpm euler --------- Co-authored-by: Sayak Paul <[email protected]> Co-authored-by: Benjamin Bossan <[email protected]> Co-authored-by: Steven Liu <[email protected]>
huggingface · May 9, 2024 · b934215 · b934215
1 parent 5ed3abd
commit b934215
Show file tree

Hide file tree

Showing 33 changed files with 1,116 additions and 205 deletions.
diff --git a/docs/source/en/using-diffusers/schedulers.md b/docs/source/en/using-diffusers/schedulers.md
@@ -212,6 +212,62 @@ images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).
 images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
 ```
 
+## Custom Timestep Schedules
+
+With all our schedulers, you can choose one of the popular timestep schedules using configurations such as `timestep_spacing`, `interpolation_type`, and `use_karras_sigmas`. Some schedulers also provide the flexibility to use a custom timestep schedule. You can use any list of arbitrary timesteps, we will use the AYS timestep schedule here as example. It is a set of 10-step optimized timestep schedules released by researchers from Nvidia that can achieve significantly better quality compared to the preset timestep schedules. You can read more about their research [here](https://research.nvidia.com/labs/toronto-ai/AlignYourSteps/). 
+
+```python
+from diffusers.schedulers import AysSchedules
+sampling_schedule = AysSchedules["StableDiffusionXLTimesteps"]
+print(sampling_schedule)
+```
+```
+[999, 845, 730, 587, 443, 310, 193, 116, 53, 13]
+```
+
+You can then create a pipeline and pass this custom timestep schedule to it as `timesteps`.
+
+```python
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    "SG161222/RealVisXL_V4.0",
+    torch_dtype=torch.float16,
+    variant="fp16",
+).to("cuda")
+
+pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, algorithm_type="sde-dpmsolver++")
+
+prompt = "A cinematic shot of a cute little rabbit wearing a jacket and doing a thumbs up"
+
+generator = torch.Generator(device="cpu").manual_seed(2487854446)
+
+image = pipe(
+    prompt=prompt,
+    negative_prompt="",
+    generator=generator,
+    timesteps=sampling_schedule,
+).images[0]
+```
+The generated image has better quality than the default linear timestep schedule for the same number of steps, and it is similar to the default timestep scheduler when running for 25 steps.
+
+<div class="flex gap-4">
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/ays.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">AYS timestep schedule 10 steps</figcaption>
+  </div>
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/10.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">Linearly-spaced timestep schedule 10 steps</figcaption>
+  </div>
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/25.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">Linearly-spaced timestep schedule 25 steps</figcaption>
+  </div>
+</div>
+
+> [!TIP]
+> 🤗 Diffusers currently only supports `timesteps` and `sigmas` for a selected list of schedulers and pipelines, but feel free to open a [feature request](https://github.com/huggingface/diffusers/issues/new/choose) if you want to extend feature to a scheduler and pipeline that does not currently support it!
+
+
 ## Models
 
 Models are loaded from the [`ModelMixin.from_pretrained`] method, which downloads and caches the latest version of the model weights and configurations. If the latest files are available in the local cache, [`~ModelMixin.from_pretrained`] reuses files in the cache instead of re-downloading them.

diff --git a/src/diffusers/pipelines/animatediff/pipeline_animatediff_sdxl.py b/src/diffusers/pipelines/animatediff/pipeline_animatediff_sdxl.py
@@ -156,6 +156,7 @@ def retrieve_timesteps(
     num_inference_steps: Optional[int] = None,
     device: Optional[Union[str, torch.device]] = None,
     timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
     """
@@ -171,14 +172,18 @@ def retrieve_timesteps(
         device (`str` or `torch.device`, *optional*):
             The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
         timesteps (`List[int]`, *optional*):
-                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
-                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
-                must be `None`.
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
 
     Returns:
         `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
         second element is the number of inference steps.
     """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
     if timesteps is not None:
         accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
         if not accepts_timesteps:
@@ -189,6 +194,16 @@ def retrieve_timesteps(
         scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
         timesteps = scheduler.timesteps
         num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
     else:
         scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
         timesteps = scheduler.timesteps
@@ -865,6 +880,7 @@ def __call__(
         width: Optional[int] = None,
         num_inference_steps: int = 50,
         timesteps: List[int] = None,
+        sigmas: List[float] = None,
         denoising_end: Optional[float] = None,
         guidance_scale: float = 5.0,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -923,6 +939,10 @@ def __call__(
                 Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
                 in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
                 passed will be used. Must be in descending order.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
             denoising_end (`float`, *optional*):
                 When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
                 completed before it is intentionally prematurely terminated. As a result, the returned sample will
@@ -1104,7 +1124,9 @@ def __call__(
         )
 
         # 4. Prepare timesteps
-        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler, num_inference_steps, device, timesteps, sigmas
+        )
 
         # 5. Prepare latent variables
         num_channels_latents = self.unet.config.in_channels

diff --git a/src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py b/src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py
@@ -137,6 +137,7 @@ def retrieve_timesteps(
     num_inference_steps: Optional[int] = None,
     device: Optional[Union[str, torch.device]] = None,
     timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
     """
@@ -152,14 +153,18 @@ def retrieve_timesteps(
         device (`str` or `torch.device`, *optional*):
             The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
         timesteps (`List[int]`, *optional*):
-                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
-                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
-                must be `None`.
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
 
     Returns:
         `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
         second element is the number of inference steps.
     """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
     if timesteps is not None:
         accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
         if not accepts_timesteps:
@@ -170,6 +175,16 @@ def retrieve_timesteps(
         scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
         timesteps = scheduler.timesteps
         num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
     else:
         scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
         timesteps = scheduler.timesteps
@@ -750,6 +765,7 @@ def __call__(
         width: Optional[int] = None,
         num_inference_steps: int = 50,
         timesteps: Optional[List[int]] = None,
+        sigmas: Optional[List[float]] = None,
         guidance_scale: float = 7.5,
         strength: float = 0.8,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -783,6 +799,14 @@ def __call__(
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality videos at the
                 expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
             strength (`float`, *optional*, defaults to 0.8):
                 Higher strength leads to more differences between original video and generated video.
             guidance_scale (`float`, *optional*, defaults to 7.5):
@@ -912,7 +936,9 @@ def __call__(
             )
 
         # 4. Prepare timesteps
-        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler, num_inference_steps, device, timesteps, sigmas
+        )
         timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, timesteps, strength, device)
         latent_timestep = timesteps[:1].repeat(batch_size * num_videos_per_prompt)
 

diff --git a/src/diffusers/pipelines/controlnet/pipeline_controlnet.py b/src/diffusers/pipelines/controlnet/pipeline_controlnet.py
@@ -97,6 +97,7 @@ def retrieve_timesteps(
     num_inference_steps: Optional[int] = None,
     device: Optional[Union[str, torch.device]] = None,
     timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
     """
@@ -112,14 +113,18 @@ def retrieve_timesteps(
         device (`str` or `torch.device`, *optional*):
             The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
         timesteps (`List[int]`, *optional*):
-                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
-                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
-                must be `None`.
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
 
     Returns:
         `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
         second element is the number of inference steps.
     """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
     if timesteps is not None:
         accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
         if not accepts_timesteps:
@@ -130,6 +135,16 @@ def retrieve_timesteps(
         scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
         timesteps = scheduler.timesteps
         num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
     else:
         scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
         timesteps = scheduler.timesteps
@@ -892,6 +907,7 @@ def __call__(
         width: Optional[int] = None,
         num_inference_steps: int = 50,
         timesteps: List[int] = None,
+        sigmas: List[float] = None,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         num_images_per_prompt: Optional[int] = 1,
@@ -941,6 +957,10 @@ def __call__(
                 Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
                 in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
                 passed will be used. Must be in descending order.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
             guidance_scale (`float`, *optional*, defaults to 7.5):
                 A higher guidance scale value encourages the model to generate images closely linked to the text
                 `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
@@ -1162,7 +1182,9 @@ def __call__(
             assert False
 
         # 5. Prepare timesteps
-        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler, num_inference_steps, device, timesteps, sigmas
+        )
         self._num_timesteps = len(timesteps)
 
         # 6. Prepare latent variables

diff --git a/src/diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion.py b/src/diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion.py
@@ -79,6 +79,7 @@ def retrieve_timesteps(
     num_inference_steps: Optional[int] = None,
     device: Optional[Union[str, torch.device]] = None,
     timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
     """
@@ -94,14 +95,18 @@ def retrieve_timesteps(
         device (`str` or `torch.device`, *optional*):
             The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
         timesteps (`List[int]`, *optional*):
-                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
-                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
-                must be `None`.
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
 
     Returns:
         `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
         second element is the number of inference steps.
     """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
     if timesteps is not None:
         accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
         if not accepts_timesteps:
@@ -112,6 +117,16 @@ def retrieve_timesteps(
         scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
         timesteps = scheduler.timesteps
         num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
     else:
         scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
         timesteps = scheduler.timesteps
@@ -673,6 +688,7 @@ def __call__(
         width: Optional[int] = None,
         num_inference_steps: int = 50,
         timesteps: List[int] = None,
+        sigmas: List[float] = None,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         num_images_per_prompt: Optional[int] = 1,
@@ -848,7 +864,9 @@ def __call__(
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 
         # 4. Prepare timesteps
-        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler, num_inference_steps, device, timesteps, sigmas
+        )
 
         # 5. Prepare latent variables
         num_channels_latents = self.unet.config.in_channels