Hugging Face's logo

shauray
/
mira-doom

Model card Files
xet
 Community
mira-doom
File size: 2,716 Bytes
80163b1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
actions:
  target_fps: 35
  valid_keys:
  - forward
  - backward
  - strafe_right
  - strafe_left
  - weapon1
  - weapon2
  - weapon3
  - weapon4
  - weapon5
  - weapon6
  - weapon7
  - attack
  - speed
wandb:
  entity: null
  project: mira-codec
  name: doom-codec-260721-1912
  group: null
  mode: online
  remove_old_runs: false
run:
  seed: 28
  steps: 20000
  batch_size: 3
  checkpoint_every: 2000
  checkpoint_keep_recent: 3
  checkpoint_keep_permanent_every: 25000
  log_every: 50
  output_dir: /runs/doom_codec
  compile: true
  viz_every: 1000
  continue_from: null
  finetune_from: null
  latents_ema_decay: 0.99
dataloader:
  num_workers: 12
  shuffle_buffer_size: 100
validation:
  batch_size: null
  val_first: true
  val_n_samples: 512
  val_every: 2%
optim:
  optimizer:
    lr: 0.0001
    betas:
    - 0.9
    - 0.95
    weight_decay: 0.1
  scheduler:
    warmup_steps: 1000
    constant_steps: 0
    decay_steps: 19000
    min_lr: 1.0e-06
  model_ema_decay: 0.9999
model:
  loss:
    _target_: mira.codec.CodecLoss
    _convert_: all
    weights:
      _target_: mira.codec.CodecLossWeights
      loss_mae: 1.0
      loss_lpips_perceptual: 1.0
      lpips_perceptual_frame_frac: 0.25
      loss_dino_latent_consistency: 1.0
      dino_latent_consistency_frame_frac: 0.25
      compile_dino: true
      auto_weight: true
  architecture:
    _target_: mira.codec.VideoCodec
    _convert_: all
    config:
      _target_: mira.codec.VideoCodecConfig
      encoder:
        latent_dim: 32
        rae_model: dinov3_vitl16
        aggregation_layers:
        - 11
        - 13
        - 15
        - 17
        - 19
        - 21
        - 23
        bottleneck:
          stride: 2
          temporal_stride: 2
          noise_tau: 0.0
        compile_dino: true
        video:
          height: 384
          width: 512
          channels: 3
          timesteps: 40
          fps: 35
      decoder:
        latent_dim: 32
        activation_checkpointing: true
        bottleneck:
          stride: 2
        vit_width: 1152
        vit_depth: 28
        vit_num_heads: 16
        mlp_dim_multiplier: 4
        qk_norm: layernorm
        patch_size: 16
        patch_size_t: 2
        video:
          height: 384
          width: 512
          channels: 3
          timesteps: 40
          fps: 35
dataset:
  actions:
    target_fps: 35
    valid_keys:
    - forward
    - backward
    - strafe_right
    - strafe_left
    - weapon1
    - weapon2
    - weapon3
    - weapon4
    - weapon5
    - weapon6
    - weapon7
    - attack
    - speed
  train_index: /data/doom_mira/train
  test_index: /data/doom_mira/test
  n_players: 1
  target_fps: 35
  frame_size:
  - 384
  - 512
  exclude_replays: false