dataset:
  dataset: FMNIST
  image_size: 32
  channels: 1
sample:
  method: ${method}
  model_path: YOUR PATH of pretrained ckpt
  image_output_path: out
  batch_size: 5
  total_num: 100
work_dir: ${hydra:runtime.cwd}
method: F-PNDM
sample_speed: 50
seed: 42
model:
  _target_: src.model.scoresde.ddpm.DDPM
  nf: 128
  ch_mult:
  - 1
  - 2
  - 2
  - 2
  num_res_blocks: 2
  attn_resolutions:
  - 16
  dropout: 0.1
  num_channels: 1
  resamp_with_conv: true
  image_size: 32
  conditional: true
  centered: true
scheduler:
  _target_: src.trainer.scheduler.Schedule
  beta_start: 0.0001
  beta_end: 0.02
  diffusion_step: 1000
train:
  optim_cfg:
    weight_decay: 0.0
    optimizer: adam
    lr: 0.0002
    beta1: 0.9
    amsgrad: false
    eps: 1.0e-08
    grad_clip: 1.0
  epochs: 100
  ema_rate: 0.9999
  use_ema: true
  batch_size: 1024
  num_workers: 6
  save_path: ./save_ckpt