upload configs/kermany/sample_sdvae.yaml

configs/kermany/sample_sdvae.yaml

@@ -0,0 +1,89 @@
+stage_1:
+  target: stage1.VAE
+  params:
+    vae_type: sdvae-mse
+    resolution: 256
+    sample_mode: mode
+stage_2:
+  target: stage2.models.DDT.DiTwDDTHeadIG
+  params:
+    input_size: 32
+    patch_size:
+    - 2
+    - 2
+    in_channels: 4
+    hidden_size:
+    - 1440
+    - 2048
+    depth:
+    - 28
+    - 2
+    num_heads:
+    - 20
+    - 16
+    mlp_ratio: 4.0
+    base_model_depth: 8
+  ckpt: /data/temp/qinshengqian/c3/Code/RAEv2/results/stage2/KERMANY_sdvae/checkpoints/ep-0000050.pt
+conditioning:
+  type: label
+  cfg_dropout_prob: 0.1
+  arch:
+    num_t_tokens: 4
+    num_c_tokens: 8
+transport:
+  prediction: x
+  time_dist_type: logit-normal_0_1
+sampler:
+  num_steps: 50
+guidance:
+  cfg:
+    scale: 1.0
+    t_min: 0.0
+    t_max: 1.0
+dataset:
+  target: imagenet
+  type: hf
+  data_dir: /data/temp/qinshengqian/c3/data/kermany2018-arrow-256
+  split: train
+  condition_type: label
+  shared_tmpdir: ~/tmp
+training:
+  epochs: 200
+  global_batch_size: 256
+  grad_accum_steps: 4
+  ema_decay: 0.9995
+  num_workers: 4
+  log_interval: 50
+  checkpoint_interval: 5
+  sample_every: 100000
+  clip_grad: 1.0
+  global_seed: 42
+  optimizer:
+    type: gmuon
+    lr: 0.0002
+    momentum: 0.95
+    nesterov: true
+    ns_coefficients_preset: POLAR_EXPRESS_COEFFICIENTS
+    ns_use_kernels: false
+    weight_decay: 0.0
+  scheduler:
+    type: linear
+    warmup_epochs: 10
+    decay_end_epoch: 150
+    base_lr: 0.0002
+    final_lr: 2.0e-05
+    warmup_from_zero: false
+  image_size: 256
+eval:
+  eval_interval: 100000
+  eval_model: false
+  eval_dir: results/stage2/eval
+  datasets: {}
+misc:
+  latent_size:
+  - 4
+  - 32
+  - 32
+  num_classes: 4
+  time_dist_shift_dim: 4096
+  time_dist_shift_base: 4096