peal.generators.ddpm_generator.DDPMConfig¶

class peal.generators.ddpm_generator.DDPMConfig[source]¶

Bases: GeneratorConfig

TODO actually implement this class properly This class defines the config of a DDPM.

Methods

Attributes

`attention_resolutions`
`base_path`	The path where the generator is stored.
`batch_size`	The batch size of the generator.
`best_fid`
`channel_mult`
`class_cond`
`data`	The config of the data.
`diffusion_steps`
`download_weights`
`dropout`
`ema_rate`
`fp16_scale_growth`
`generator_type`	The type of generator that shall be used.
`gpus`
`image_size`
`is_trained`
`learn_sigma`
`log_interval`
`lr`
`lr_anneal_steps`
`max_steps`
`microbatch`
`noise_schedule`
`num_channels`	The number of channels
`num_head_channels`
`num_heads`
`num_heads_upsample`
`num_res_blocks`
`output_path`
`predict_xstart`
`resblock_updown`
`rescale_learned_sigmas`
`rescale_timesteps`
`resume_checkpoint`
`save_interval`
`schedule_sampler`
`stochastic`
`timestep_respacing`
`use_checkpoint`
`use_fp16`
`use_hdf5`
`use_kl`
`use_new_attention_order`
`use_scale_shift_norm`
`weight_decay`
`x_selection`

generator_type: str = 'DDPM'¶: The type of generator that shall be used.

base_path: str = 'peal_runs/ddpm'¶: The path where the generator is stored.

data: DataConfig¶: The config of the data.

num_channels: int = 128¶: The number of channels

image_size: int | None = None¶

num_res_blocks: int = 2¶

num_heads: int = 4¶

num_heads_upsample: int = -1¶

num_head_channels: int = -1¶

attention_resolutions: str = '32,16,8'¶

channel_mult: str = ''¶

dropout: float = 0.0¶

class_cond: bool = False¶

use_checkpoint: bool = False¶

use_scale_shift_norm: bool = True¶

resblock_updown: bool = True¶

use_fp16: bool = False¶

use_new_attention_order: bool = False¶

schedule_sampler: str = 'uniform'¶

lr: float = 0.0001¶

weight_decay: float = 0.0¶

lr_anneal_steps: int = 0¶

batch_size: int = 1¶: The batch size of the generator.

microbatch: int = -1¶

ema_rate: str = '0.9999'¶

log_interval: int = 10¶

save_interval: int = 1000¶

max_steps: int = 10000¶

resume_checkpoint: str = ''¶

fp16_scale_growth: float = 0.001¶

output_path: str = 'peal_runs/ddpm/outputs'¶

gpus: str = ''¶

use_hdf5: bool = False¶

learn_sigma: bool = True¶

diffusion_steps: int = 1000¶

noise_schedule: str = 'linear'¶

timestep_respacing: str = '50'¶

use_kl: bool = False¶

predict_xstart: bool = False¶

rescale_timesteps: bool = False¶

rescale_learned_sigmas: bool = False¶

stochastic: bool = True¶

x_selection: list | None = None¶

is_trained: bool = False¶

best_fid: float = 1000000000.0¶

download_weights: str | None = None¶