the-Lin commited on
Commit
3639843
·
verified ·
1 Parent(s): 1d1b03b

Upload folder using huggingface_hub

Browse files
cumv2_r1.0/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "delta",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "dino",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
cumv2_r1.0/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29e45f5c10522556dd621a599a5c181386e0bbb3e8d0e2a65b21b17ff1d8ae2d
3
+ size 1627530056
cumv2_r2.0/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 2.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "delta",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "dino",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
cumv2_r2.0/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0da58cba34ba11a4f016b742f6606eef725126f2c75a46b29fb070c6b73f9d71
3
+ size 1627530056
dino_dz/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v1",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "delta",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "dino",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
dino_dz/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:822f67d04ad65c91fac8c894192d91c5b16c2438ee942d7661b1c4b9f94d6fa4
3
+ size 1627530056
geo_latent/config.json ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_keep_aspect": true,
38
+ "geo_loss_mode": "latent",
39
+ "geo_n_neg": 8,
40
+ "geo_nce_temp": 0.07,
41
+ "geo_neg_radius": 12.0,
42
+ "geo_ref_px": 20.0,
43
+ "geo_track_res": 224,
44
+ "group_loss_mode": "none",
45
+ "group_std_grip": 0.0572,
46
+ "group_std_pos": 0.1468,
47
+ "group_std_rot": 0.0097,
48
+ "hidden_size": 1024,
49
+ "k_high": 1.5,
50
+ "k_low": 0.5,
51
+ "linear_aug_mode": "none",
52
+ "max_action_dim": 32,
53
+ "max_num_views": 6,
54
+ "max_state_dim": 64,
55
+ "mm_projector_cfg": {
56
+ "_convert_": "object",
57
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
58
+ "config": {
59
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
60
+ "hidden_size": 1024,
61
+ "mm_hidden_size": 1024,
62
+ "mm_projector_type": "mlp_doubledownsample"
63
+ }
64
+ },
65
+ "mm_vision_select_layer": -2,
66
+ "model_dtype": "float32",
67
+ "noise_beta_alpha": 1.5,
68
+ "noise_beta_beta": 1.0,
69
+ "noise_s": 0.999,
70
+ "null_aug_mode": "none",
71
+ "null_detach": true,
72
+ "num_inference_timesteps": 16,
73
+ "num_timestep_buckets": 1000,
74
+ "siglip_hidden_size": 1024,
75
+ "siglip_model_cfg": {
76
+ "_convert_": "object",
77
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
78
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
79
+ },
80
+ "spatial_mask_mode": "none",
81
+ "spatial_mask_sigma": 1.0,
82
+ "tune_vision_tower": true,
83
+ "vggt_fusion_mode": "sub",
84
+ "vggt_mid_dim": 512,
85
+ "vggt_num_heads": 8,
86
+ "vggt_num_queries": 16,
87
+ "vision_source": "vggt",
88
+ "vl_self_attention_cfg": {
89
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
90
+ "attention_head_dim": 64,
91
+ "dropout": 0.2,
92
+ "final_dropout": true,
93
+ "num_attention_heads": 16,
94
+ "num_layers": 4,
95
+ "positional_embeddings": null
96
+ },
97
+ "w_fwd": 1.0,
98
+ "w_geo": 0.05,
99
+ "w_geo_latent": 0.05,
100
+ "w_grip": 1.0,
101
+ "w_hom": 1.0,
102
+ "w_null": 1.0,
103
+ "w_pos": 1.0,
104
+ "w_rot": 1.0
105
+ }
106
+ },
107
+ "action_horizon": 16,
108
+ "architectures": [
109
+ "IDM"
110
+ ],
111
+ "backbone_cfg": {
112
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
113
+ },
114
+ "compute_dtype": "bfloat16",
115
+ "hidden_size": 0,
116
+ "model_dtype": "float32",
117
+ "model_type": "idm",
118
+ "torch_dtype": "float32",
119
+ "transformers_version": "4.49.0"
120
+ }
geo_latent/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ee0ae689b13f422ff64e8962517a7e66afd266f487c2136818ed16f4675a0be
3
+ size 4895104392
geo_latent_w15/config.json ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_keep_aspect": true,
38
+ "geo_loss_mode": "latent",
39
+ "geo_n_neg": 8,
40
+ "geo_nce_temp": 0.07,
41
+ "geo_neg_radius": 12.0,
42
+ "geo_ref_px": 20.0,
43
+ "geo_track_res": 224,
44
+ "group_loss_mode": "none",
45
+ "group_std_grip": 0.0572,
46
+ "group_std_pos": 0.1468,
47
+ "group_std_rot": 0.0097,
48
+ "hidden_size": 1024,
49
+ "k_high": 1.5,
50
+ "k_low": 0.5,
51
+ "linear_aug_mode": "none",
52
+ "max_action_dim": 32,
53
+ "max_num_views": 6,
54
+ "max_state_dim": 64,
55
+ "mm_projector_cfg": {
56
+ "_convert_": "object",
57
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
58
+ "config": {
59
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
60
+ "hidden_size": 1024,
61
+ "mm_hidden_size": 1024,
62
+ "mm_projector_type": "mlp_doubledownsample"
63
+ }
64
+ },
65
+ "mm_vision_select_layer": -2,
66
+ "model_dtype": "float32",
67
+ "noise_beta_alpha": 1.5,
68
+ "noise_beta_beta": 1.0,
69
+ "noise_s": 0.999,
70
+ "null_aug_mode": "none",
71
+ "null_detach": true,
72
+ "num_inference_timesteps": 16,
73
+ "num_timestep_buckets": 1000,
74
+ "siglip_hidden_size": 1024,
75
+ "siglip_model_cfg": {
76
+ "_convert_": "object",
77
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
78
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
79
+ },
80
+ "spatial_mask_mode": "none",
81
+ "spatial_mask_sigma": 1.0,
82
+ "tune_vision_tower": true,
83
+ "vggt_fusion_mode": "sub",
84
+ "vggt_mid_dim": 512,
85
+ "vggt_num_heads": 8,
86
+ "vggt_num_queries": 16,
87
+ "vision_source": "vggt",
88
+ "vl_self_attention_cfg": {
89
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
90
+ "attention_head_dim": 64,
91
+ "dropout": 0.2,
92
+ "final_dropout": true,
93
+ "num_attention_heads": 16,
94
+ "num_layers": 4,
95
+ "positional_embeddings": null
96
+ },
97
+ "w_fwd": 1.0,
98
+ "w_geo": 0.05,
99
+ "w_geo_latent": 0.15,
100
+ "w_grip": 1.0,
101
+ "w_hom": 1.0,
102
+ "w_null": 1.0,
103
+ "w_pos": 1.0,
104
+ "w_rot": 1.0
105
+ }
106
+ },
107
+ "action_horizon": 16,
108
+ "architectures": [
109
+ "IDM"
110
+ ],
111
+ "backbone_cfg": {
112
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
113
+ },
114
+ "compute_dtype": "bfloat16",
115
+ "hidden_size": 0,
116
+ "model_dtype": "float32",
117
+ "model_type": "idm",
118
+ "torch_dtype": "float32",
119
+ "transformers_version": "4.49.0"
120
+ }
geo_latent_w15/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51de0ce830bb341dd8b11b60a4723fcadc02d00552aa5250e45a98d9ef65a524
3
+ size 4895104392
siglip_abs/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v1",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "siglip",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
siglip_abs/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fdf58b7f23526e2ab675b28b65b17b3dc4711498bfe2258fdd5bed8ac08a906e
3
+ size 2549650952
siglip_delta/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v1",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "siglip",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
siglip_delta/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d31b962d649e41def232ec8a8f6fab9b496be607b5649a38ab1f638cf6c50afe
3
+ size 2549650952
vggt_none/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v1",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "vggt",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
vggt_none/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f75cab6b7d9924832de407f09afd2dac2da45ba67e3361f3b132a8283ee449b5
3
+ size 4895104392
vggt_none_cum/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "none",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "vggt",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
vggt_none_cum/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d1f4fc57d4470569b4046db2ebd0f167e51565e2369223746073e040e1d97a2
3
+ size 4895104392
vggt_sub/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v1",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "sub",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "vggt",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
vggt_sub/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a58616c8d404c071138ca3440327fcdce12b7a3830df7e6e499a75e787fdc512
3
+ size 4895104392
vggt_sub_cum/config.json ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "_convert_": "object",
5
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDM",
6
+ "config": {
7
+ "_recursive_": false,
8
+ "_target_": "gr00t.model.action_head.flow_matching_action_head_idm.FlowMatchingActionHeadIDMConfig",
9
+ "action_dim": 32,
10
+ "action_horizon": 16,
11
+ "add_pos_embed": true,
12
+ "add_seperator_token": true,
13
+ "add_view_embed": true,
14
+ "alpha_high": 2.0,
15
+ "alpha_low": 0.0,
16
+ "backbone_features_projector_cfg": null,
17
+ "cum_loss_mode": "v2",
18
+ "cum_ref_pos_mm": 60.0,
19
+ "cum_ref_rot_deg": 6.0,
20
+ "cum_w_pos": 1.0,
21
+ "cum_w_rot": 1.0,
22
+ "diffusion_model_cfg": {
23
+ "_target_": "gr00t.model.action_head.cross_attention_dit.DiT",
24
+ "attention_head_dim": 64,
25
+ "dropout": 0.2,
26
+ "final_dropout": true,
27
+ "interleave_self_attention": true,
28
+ "norm_type": "ada_norm",
29
+ "num_attention_heads": 16,
30
+ "num_layers": 8,
31
+ "output_dim": 1024,
32
+ "positional_embeddings": null
33
+ },
34
+ "feature_mode": "frame",
35
+ "geo_gate_slope": 1.0,
36
+ "geo_gate_tau": 2.0,
37
+ "geo_loss_mode": "none",
38
+ "geo_ref_px": 20.0,
39
+ "geo_track_res": 224,
40
+ "group_loss_mode": "none",
41
+ "group_std_grip": 0.0572,
42
+ "group_std_pos": 0.1468,
43
+ "group_std_rot": 0.0097,
44
+ "hidden_size": 1024,
45
+ "k_high": 1.5,
46
+ "k_low": 0.5,
47
+ "linear_aug_mode": "none",
48
+ "max_action_dim": 32,
49
+ "max_num_views": 6,
50
+ "max_state_dim": 64,
51
+ "mm_projector_cfg": {
52
+ "_convert_": "object",
53
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjector",
54
+ "config": {
55
+ "_target_": "gr00t.model.action_head.multimodal_projector.MultimodalProjectorConfig",
56
+ "hidden_size": 1024,
57
+ "mm_hidden_size": 1024,
58
+ "mm_projector_type": "mlp_doubledownsample"
59
+ }
60
+ },
61
+ "mm_vision_select_layer": -2,
62
+ "model_dtype": "float32",
63
+ "noise_beta_alpha": 1.5,
64
+ "noise_beta_beta": 1.0,
65
+ "noise_s": 0.999,
66
+ "null_aug_mode": "none",
67
+ "null_detach": true,
68
+ "num_inference_timesteps": 16,
69
+ "num_timestep_buckets": 1000,
70
+ "siglip_hidden_size": 1024,
71
+ "siglip_model_cfg": {
72
+ "_convert_": "object",
73
+ "_target_": "gr00t.model.action_head.siglip.SiglipModel.from_pretrained",
74
+ "pretrained_model_name_or_path": "${oc.env:SIGLIP_PATH,checkpoints/google/siglip2-large-patch16-256}"
75
+ },
76
+ "spatial_mask_mode": "none",
77
+ "spatial_mask_sigma": 1.0,
78
+ "tune_vision_tower": true,
79
+ "vggt_fusion_mode": "sub",
80
+ "vggt_mid_dim": 512,
81
+ "vggt_num_heads": 8,
82
+ "vggt_num_queries": 16,
83
+ "vision_source": "vggt",
84
+ "vl_self_attention_cfg": {
85
+ "_target_": "gr00t.model.action_head.cross_attention_dit.SelfAttentionTransformer",
86
+ "attention_head_dim": 64,
87
+ "dropout": 0.2,
88
+ "final_dropout": true,
89
+ "num_attention_heads": 16,
90
+ "num_layers": 4,
91
+ "positional_embeddings": null
92
+ },
93
+ "w_fwd": 1.0,
94
+ "w_geo": 0.05,
95
+ "w_grip": 1.0,
96
+ "w_hom": 1.0,
97
+ "w_null": 1.0,
98
+ "w_pos": 1.0,
99
+ "w_rot": 1.0
100
+ }
101
+ },
102
+ "action_horizon": 16,
103
+ "architectures": [
104
+ "IDM"
105
+ ],
106
+ "backbone_cfg": {
107
+ "_target_": "gr00t.model.backbone.IdentityBackbone"
108
+ },
109
+ "compute_dtype": "bfloat16",
110
+ "hidden_size": 0,
111
+ "model_dtype": "float32",
112
+ "model_type": "idm",
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.49.0"
115
+ }
vggt_sub_cum/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bf902285be05ae5b0ad8f8342a259ed5a542f405e9ae161f8a058ecfd8b926e
3
+ size 4895104392