UniPC is a training-free framework designed for the fast sampling of diffusion models, which consists of a corrector (UniC) and a predictor (UniP) that share a unified analytical form and support arbitrary orders.
For more details about the method, please refer to the paper and the code.
Fast Sampling of Diffusion Models with Exponential Integrator.
( 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 solver_order: int = 2 prediction_type: str = 'epsilon' thresholding: bool = False dynamic_thresholding_ratio: float = 0.995 sample_max_value: float = 1.0 predict_x0: bool = True solver_type: str = 'bh2' lower_order_final: bool = True disable_corrector: typing.List[int] = [] solver_p: SchedulerMixin = None use_karras_sigmas: typing.Optional[bool] = False timestep_spacing: str = 'linspace' steps_offset: int = 0 )
Parameters
int
) — number of diffusion steps used to train the model.
float
) — the starting beta
value of inference.
float
) — the final beta
value.
str
) —
the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
linear
, scaled_linear
, or squaredcos_cap_v2
.
np.ndarray
, optional) —
option to pass an array of betas directly to the constructor to bypass beta_start
, beta_end
etc.
int
, default 2
) —
the order of UniPC, also the p in UniPC-p; can be any positive integer. Note that the effective order of
accuracy is solver_order + 1
due to the UniC. We recommend to use solver_order=2
for guided sampling,
and solver_order=3
for unconditional sampling.
str
, default epsilon
, optional) —
prediction type of the scheduler function, one of epsilon
(predicting the noise of the diffusion
process), sample
(directly predicting the noisy sample) or
v_prediction` (see section 2.4
https://imagen.research.google/video/paper.pdf)
bool
, default False
) —
whether to use the “dynamic thresholding” method (introduced by Imagen, https://arxiv.org/abs/2205.11487).
For pixel-space diffusion models, you can set both predict_x0=True
and thresholding=True
to use the
dynamic thresholding. Note that the thresholding method is unsuitable for latent-space diffusion models
(such as stable-diffusion).
float
, default 0.995
) —
the ratio for the dynamic thresholding method. Default is 0.995
, the same as Imagen
(https://arxiv.org/abs/2205.11487).
float
, default 1.0
) —
the threshold value for dynamic thresholding. Valid only when thresholding=True
and predict_x0=True
.
bool
, default True
) —
whether to use the updating algrithm on the predicted x0. See https://arxiv.org/abs/2211.01095 for details
str
, default bh2
) —
the solver type of UniPC. We recommend use bh1
for unconditional sampling when steps < 10, and use bh2
otherwise.
bool
, default True
) —
whether to use lower-order solvers in the final steps. Only valid for < 15 inference steps. We empirically
find this trick can stabilize the sampling of DPM-Solver for steps < 15, especially for steps <= 10.
list
, default []
) —
decide which step to disable the corrector. For large guidance scale, the misalignment between the
epsilon_theta(x_t, c)
and epsilon_theta(x_t^c, c)
might influence the convergence. This can be mitigated
by disable the corrector at the first few steps (e.g., disable_corrector=[0])
SchedulerMixin
, default None
) —
can be any other scheduler. If specified, the algorithm will become solver_p + UniC.
bool
, optional, defaults to False
) —
This parameter controls whether to use Karras sigmas (Karras et al. (2022) scheme) for step sizes in the
noise schedule during the sampling process. If True, the sigmas will be determined according to a sequence
of noise levels {σi} as defined in Equation (5) of the paper https://arxiv.org/pdf/2206.00364.pdf.
str
, default "linspace"
) —
The way the timesteps should be scaled. Refer to Table 2. of Common Diffusion Noise Schedules and Sample
Steps are Flawed for more information.
int
, default 0
) —
an offset added to the inference steps. You can use a combination of offset=1
and
set_alpha_to_one=False
, to make the last step use step 0 for the previous alpha product, as done in
stable diffusion.
UniPC is a training-free framework designed for the fast sampling of diffusion models, which consists of a corrector (UniC) and a predictor (UniP) that share a unified analytical form and support arbitrary orders. UniPC is by desinged model-agnostic, supporting pixel-space/latent-space DPMs on unconditional/conditional sampling. It can also be applied to both noise prediction model and data prediction model. The corrector UniC can be also applied after any off-the-shelf solvers to increase the order of accuracy.
For more details, see the original paper: https://arxiv.org/abs/2302.04867
Currently, we support the multistep UniPC for both noise prediction models and data prediction models. We recommend
to use solver_order=2
for guided sampling, and solver_order=3
for unconditional sampling.
We also support the “dynamic thresholding” method in Imagen (https://arxiv.org/abs/2205.11487). For pixel-space
diffusion models, you can set both predict_x0=True
and thresholding=True
to use the dynamic thresholding. Note
that the thresholding method is unsuitable for latent-space diffusion models (such as stable-diffusion).
~ConfigMixin takes care of storing all config attributes that are passed in the scheduler’s __init__
function, such as num_train_timesteps
. They can be accessed via scheduler.config.num_train_timesteps
.
SchedulerMixin provides general loading and saving functionality via the SchedulerMixin.save_pretrained() and
from_pretrained() functions.
(
model_output: FloatTensor
timestep: int
sample: FloatTensor
)
→
torch.FloatTensor
Parameters
torch.FloatTensor
) — direct output from learned diffusion model.
int
) — current discrete timestep in the diffusion chain.
torch.FloatTensor
) —
current instance of sample being created by diffusion process.
Returns
torch.FloatTensor
the converted model output.
Convert the model output to the corresponding type that the algorithm PC needs.
(
this_model_output: FloatTensor
this_timestep: int
last_sample: FloatTensor
this_sample: FloatTensor
order: int
)
→
torch.FloatTensor
Parameters
torch.FloatTensor
) — the model outputs at x_t
int
) — the current timestep t
torch.FloatTensor
) — the generated sample before the last predictor: x_{t-1}
torch.FloatTensor
) — the generated sample after the last predictor: x_{t}
int
) — the p
of UniC-p at this step. Note that the effective order of accuracy
should be order + 1
Returns
torch.FloatTensor
the corrected sample tensor at the current timestep.
One step for the UniC (B(h) version).
(
model_output: FloatTensor
prev_timestep: int
sample: FloatTensor
order: int
)
→
torch.FloatTensor
Parameters
torch.FloatTensor
) —
direct outputs from learned diffusion model at the current timestep.
int
) — previous discrete timestep in the diffusion chain.
torch.FloatTensor
) —
current instance of sample being created by diffusion process.
int
) — the order of UniP at this step, also the p in UniPC-p.
Returns
torch.FloatTensor
the sample tensor at the previous timestep.
One step for the UniP (B(h) version). Alternatively, self.solver_p
is used if is specified.
(
sample: FloatTensor
*args
**kwargs
)
→
torch.FloatTensor
Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep.
( num_inference_steps: int device: typing.Union[str, torch.device] = None )
Sets the timesteps used for the diffusion chain. Supporting function to be run before inference.
(
model_output: FloatTensor
timestep: int
sample: FloatTensor
return_dict: bool = True
)
→
~scheduling_utils.SchedulerOutput
or tuple
Parameters
torch.FloatTensor
) — direct output from learned diffusion model.
int
) — current discrete timestep in the diffusion chain.
torch.FloatTensor
) —
current instance of sample being created by diffusion process.
bool
) — option for returning tuple rather than SchedulerOutput class
Returns
~scheduling_utils.SchedulerOutput
or tuple
~scheduling_utils.SchedulerOutput
if return_dict
is
True, otherwise a tuple
. When returning a tuple, the first element is the sample tensor.
Step function propagating the sample with the multistep UniPC.