DDIMScheduler
Denoising Diffusion Implicit Models (DDIM) by Jiaming Song, Chenlin Meng and Stefano Ermon.
The abstract from the paper is:
Denoising diffusion probabilistic models (DDPMs) have achieved high quality image generation without adversarial training, yet they require simulating a Markov chain for many steps to produce a sample. To accelerate sampling, we present denoising diffusion implicit models (DDIMs), a more efficient class of iterative implicit probabilistic models with the same training procedure as DDPMs. In DDPMs, the generative process is defined as the reverse of a Markovian diffusion process. We construct a class of non-Markovian diffusion processes that lead to the same training objective, but whose reverse process can be much faster to sample from. We empirically demonstrate that DDIMs can produce high quality samples 10× to 50× faster in terms of wall-clock time compared to DDPMs, allow us to trade off computation for sample quality, and can perform semantically meaningful image interpolation directly in the latent space.
The original codebase of this paper can be found at ermongroup/ddim, and you can contact the author on tsong.me.
Tips
The paper Common Diffusion Noise Schedules and Sample Steps are Flawed claims that a mismatch between the training and inference settings leads to suboptimal inference generation results for Stable Diffusion. To fix this, the authors propose:
🧪 This is an experimental feature!
- rescale the noise schedule to enforce zero terminal signal-to-noise ratio (SNR)
pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, rescale_betas_zero_snr=True)
- train a model with
v_prediction
(add the following argument to the train_text_to_image.py or train_text_to_image_lora.py scripts)
--prediction_type="v_prediction"
- change the sampler to always start from the last timestep
pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
- rescale classifier-free guidance to prevent over-exposure
image = pipe(prompt, guidance_rescale=0.7).images[0]
For example:
from diffusers import DiffusionPipeline, DDIMScheduler
import torch
pipe = DiffusionPipeline.from_pretrained("ptx0/pseudo-journey-v2", torch_dtype=torch.float16)
pipe.scheduler = DDIMScheduler.from_config(
pipe.scheduler.config, rescale_betas_zero_snr=True, timestep_spacing="trailing"
)
pipe.to("cuda")
prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intricate detail, octane render, 8k"
image = pipe(prompt, guidance_rescale=0.7).images[0]
image
DDIMScheduler
class diffusers.DDIMScheduler
< source >( num_train_timesteps: int = 1000 beta_start: float = 0.0001 beta_end: float = 0.02 beta_schedule: str = 'linear' trained_betas: typing.Union[numpy.ndarray, typing.List[float], NoneType] = None clip_sample: bool = True set_alpha_to_one: bool = True steps_offset: int = 0 prediction_type: str = 'epsilon' thresholding: bool = False dynamic_thresholding_ratio: float = 0.995 clip_sample_range: float = 1.0 sample_max_value: float = 1.0 timestep_spacing: str = 'leading' rescale_betas_zero_snr: bool = False )
Parameters
- num_train_timesteps (
int
, defaults to 1000) — The number of diffusion steps to train the model. - beta_start (
float
, defaults to 0.0001) — The startingbeta
value of inference. - beta_end (
float
, defaults to 0.02) — The finalbeta
value. - beta_schedule (
str
, defaults to"linear"
) — The beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose fromlinear
,scaled_linear
, orsquaredcos_cap_v2
. - trained_betas (
np.ndarray
, optional) — Pass an array of betas directly to the constructor to bypassbeta_start
andbeta_end
. - clip_sample (
bool
, defaults toTrue
) — Clip the predicted sample for numerical stability. - clip_sample_range (
float
, defaults to 1.0) — The maximum magnitude for sample clipping. Valid only whenclip_sample=True
. - set_alpha_to_one (
bool
, defaults toTrue
) — Each diffusion step uses the alphas product value at that step and at the previous one. For the final step there is no previous alpha. When this option isTrue
the previous alpha product is fixed to1
, otherwise it uses the alpha value at step 0. - steps_offset (
int
, defaults to 0) — An offset added to the inference steps, as required by some model families. - prediction_type (
str
, defaults toepsilon
, optional) — Prediction type of the scheduler function; can beepsilon
(predicts the noise of the diffusion process),sample
(directly predicts the noisy sample) or
v_prediction` (see section 2.4 of Imagen Video paper). - thresholding (
bool
, defaults toFalse
) — Whether to use the “dynamic thresholding” method. This is unsuitable for latent-space diffusion models such as Stable Diffusion. - dynamic_thresholding_ratio (
float
, defaults to 0.995) — The ratio for the dynamic thresholding method. Valid only whenthresholding=True
. - sample_max_value (
float
, defaults to 1.0) — The threshold value for dynamic thresholding. Valid only whenthresholding=True
. - timestep_spacing (
str
, defaults to"leading"
) — The way the timesteps should be scaled. Refer to Table 2 of the Common Diffusion Noise Schedules and Sample Steps are Flawed for more information. - rescale_betas_zero_snr (
bool
, defaults toFalse
) — Whether to rescale the betas to have zero terminal SNR. This enables the model to generate very bright and dark samples instead of limiting it to samples with medium brightness. Loosely related to--offset_noise
.
DDIMScheduler
extends the denoising procedure introduced in denoising diffusion probabilistic models (DDPMs) with
non-Markovian guidance.
This model inherits from SchedulerMixin and ConfigMixin. Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving.
scale_model_input
< source >( sample: Tensor timestep: typing.Optional[int] = None ) → torch.Tensor
Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep.
set_timesteps
< source >( num_inference_steps: int device: typing.Union[str, torch.device] = None )
Sets the discrete timesteps used for the diffusion chain (to be run before inference).
step
< source >( model_output: Tensor timestep: int sample: Tensor eta: float = 0.0 use_clipped_model_output: bool = False generator = None variance_noise: typing.Optional[torch.Tensor] = None return_dict: bool = True ) → DDIMSchedulerOutput or tuple
Parameters
- model_output (
torch.Tensor
) — The direct output from learned diffusion model. - timestep (
float
) — The current discrete timestep in the diffusion chain. - sample (
torch.Tensor
) — A current instance of a sample created by the diffusion process. - eta (
float
) — The weight of noise for added noise in diffusion step. - use_clipped_model_output (
bool
, defaults toFalse
) — IfTrue
, computes “corrected”model_output
from the clipped predicted original sample. Necessary because predicted original sample is clipped to [-1, 1] whenself.config.clip_sample
isTrue
. If no clipping has happened, “corrected”model_output
would coincide with the one provided as input anduse_clipped_model_output
has no effect. - generator (
torch.Generator
, optional) — A random number generator. - variance_noise (
torch.Tensor
) — Alternative to generating noise withgenerator
by directly providing the noise for the variance itself. Useful for methods such asCycleDiffusion
. - return_dict (
bool
, optional, defaults toTrue
) — Whether or not to return a DDIMSchedulerOutput ortuple
.
Returns
DDIMSchedulerOutput or tuple
If return_dict is True
, DDIMSchedulerOutput is returned, otherwise a
tuple is returned where the first element is the sample tensor.
Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise).
DDIMSchedulerOutput
class diffusers.schedulers.scheduling_ddim.DDIMSchedulerOutput
< source >( prev_sample: Tensor pred_original_sample: typing.Optional[torch.Tensor] = None )
Parameters
- prev_sample (
torch.Tensor
of shape(batch_size, num_channels, height, width)
for images) — Computed sample(x_{t-1})
of previous timestep.prev_sample
should be used as next model input in the denoising loop. - pred_original_sample (
torch.Tensor
of shape(batch_size, num_channels, height, width)
for images) — The predicted denoised sample(x_{0})
based on the model output from the current timestep.pred_original_sample
can be used to preview progress or for guidance.
Output class for the scheduler’s step
function output.