File size: 2,394 Bytes
952993c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
# Model config matching the FlashWAM M1_FusedKV_RopeFixed LIBERO run's final
# checkpoint runs/libero/M1_FusedKV_RopeFixed/2026-07-03_16-13-56/
# checkpoints/weights/step_021700.pt: FasterWAM decoupled MoT with a 1-layer
# action expert, fused_kv KV source (kv_fusing_layer + k/v channel
# projections), and the action-aligned video RoPE fix (fixed_rope: true ->
# MoTDecoupledActionAlignedVideoRoPE).
# Copied VERBATIM from that run's resolved config.yaml model block — the
# architecture must match the checkpoint exactly for the finetune leg, and the
# scratch leg reuses the identical architecture for a clean comparison.
# NOTE action_dit_pretrained_path is null exactly as in the LIBERO run: fused
# KV modes cannot use layer-selective init, so with a null path the 1-layer
# action expert starts RANDOM (fasterwam_decoupled.py from_wan22_pretrained);
# the video expert always initializes from the Wan2.2-TI2V-5B base.
_target_: fastwam.models.wan22.fasterwam_decoupled.create_fasterwam_decoupled
model_id: Wan-AI/Wan2.2-TI2V-5B
tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
tokenizer_max_len: 128
load_text_encoder: false
proprio_dim: 8
redirect_common_files: true
mot_checkpoint_mixed_attn: false
action_dit_pretrained_path: null
skip_dit_load_from_pretrain: false
decoupled: true
kv_source_mode: fused_kv
fusion_hidden_dim: 64
fusion_use_norm: true
fixed_rope: true
video_dit_config:
  has_image_input: false
  patch_size:
  - 1
  - 2
  - 2
  in_dim: 48
  hidden_dim: 3072
  ffn_dim: 14336
  freq_dim: 256
  text_dim: 4096
  out_dim: 48
  num_heads: 24
  attn_head_dim: 128
  num_layers: 30
  eps: 1.0e-06
  seperated_timestep: true
  require_clip_embedding: false
  require_vae_embedding: false
  fuse_vae_embedding_in_latents: true
  use_gradient_checkpointing: false
  video_attention_mask_mode: first_frame_causal
  action_conditioned: false
  action_dim: 7
  action_group_causal_mask_mode: group_diagonal
action_dit_config:
  action_dim: 7
  hidden_dim: 1024
  ffn_dim: 4096
  num_heads: 24
  attn_head_dim: 128
  num_layers: 1
  layer_selected: null
  text_dim: 4096
  freq_dim: 256
  eps: 1.0e-06
  use_gradient_checkpointing: false
  action_text_cross_attn: true
video_scheduler:
  train_shift: 5.0
  infer_shift: 5.0
  num_train_timesteps: 1000
action_scheduler:
  train_shift: 5.0
  infer_shift: 5.0
  num_train_timesteps: 1000
loss:
  lambda_video: 1.0
  lambda_action: 1.0