ttts/diffusion/config.yaml

train: 
  train_steps: 1000000
  val_freq: 100
  save_freq: 1000
  keep_ckpts: 3
  lr: 1e-4
  logs_folder: 'ttts/diffusion/logs'
  accumulate_num: 1
  unconditioned_percentage: 0.1
  timesteps: 1000
dataset:
  path: /home/hyc/tortoise_plus_zh/ttts/datasets/filtered_paths.jsonl
  gpt_path: /home/hyc/tortoise_plus_zh/ttts/gpt/logs/2023-12-24-14-22-14/model-70.pt
dataloader:
  batch_size : 64
  shuffle: false
  num_workers : 64
  drop_last: true 
  pin_memory: true

diffusion:
  model_channels: 512
  num_layers: 10 
  in_channels: 100 
  out_channels: 200
  in_latent_channels: 1024 
  in_tokens: 8193 
  dropout: 0 
  use_fp16: false 
  num_heads: 16
  layer_drop: 0.2 
  unconditioned_percentage: 0.15

base_diffusion:
  in_channels: 100
  out_channels: 200
  model_channels: 512
  attention_resolutions: [ 4, 2, 1 ]
  num_res_blocks: 2
  channel_mult: [ 1, 1 ]
  num_heads: 8
  use_spatial_transformer: True
  transformer_depth: 1
  context_dim: 512
  use_checkpoint: True
  dims: 1
  legacy: False
refer_diffusion:
  in_channels: 100
  out_channels: 200
  model_channels: 512
  attention_resolutions: [ 4, 2, 1 ]
  num_res_blocks: 2
  channel_mult: [ 1, 1]
  num_heads: 8
  use_spatial_transformer: True
  transformer_depth: 1
  context_dim: 512
  use_checkpoint: True
  legacy: False
clip:
  embed_dim: 512
  vision_cfg: 
    layers: 6
    width: 512
    head_width: 64
    mlp_ratio: 4.0
    patch_dropout: 0.4
    attentional_pool: False
    patch_size: 32
    image_size: 1000
    in_channels: 100
    pool_type: 'tok'
    pos_embed_type: 'learnable'
    final_ln_after_pool: false