Hugging Face's logo

yangyandan
/
ABot-M0-LIBERO

Model card Files
xet
 Community
yangyandan commited on
Commit
8c6e596
·
verified ·
1 Parent(s): ccc1d68

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. checkpoints/steps_40000_pytorch_model.pt +3 -0
  2. config.yaml +63 -0
  3. dataset_statistics.json +133 -0
  4. summary.jsonl +10 -0
checkpoints/steps_40000_pytorch_model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ca856f51e3e637f04ffe25dd2476e0b0fefa5220d76ad3d878ee721ce4034aa
3
+ size 15473526557
config.yaml ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ datasets:
2
+ vla_data:
3
+ data_mix: libero_all_ration
4
+ data_root_dir: /mnt/xlab-nas-2/vla_dataset/benchmark/libero
5
+ dataset_py: lerobot_datasets
6
+ num_workers: 4
7
+ per_device_batch_size: 8
8
+ framework:
9
+ action_model:
10
+ action_dim: 14
11
+ action_horizon: 10
12
+ action_model_type: DiT-B
13
+ add_pos_embed: true
14
+ diffusion_model_cfg:
15
+ cross_attention_dim: 2560
16
+ dropout: 0.2
17
+ final_dropout: true
18
+ interleave_self_attention: true
19
+ norm_type: ada_norm
20
+ num_layers: 16
21
+ output_dim: 1024
22
+ positional_embeddings: null
23
+ future_action_window_size: 9
24
+ hidden_size: 1024
25
+ max_seq_len: 1024
26
+ noise_beta_alpha: 1.5
27
+ noise_beta_beta: 1.0
28
+ noise_s: 0.999
29
+ num_inference_timesteps: 4
30
+ num_target_vision_tokens: 32
31
+ num_timestep_buckets: 1000
32
+ past_action_window_size: 0
33
+ state_dim: 7
34
+ name: ABot_M0
35
+ qwenvl:
36
+ base_vlm: /mnt/workspace/zengshuang.zs/checkpoints/Qwen3-VL-4B-Instruct-Action
37
+ run_root_dir: /mnt/workspace/zengshuang.zs/output/libero_all
38
+ seed: 42
39
+ trainer:
40
+ eval_interval: 5000
41
+ freeze_modules: spatial_model
42
+ gradient_accumulation_steps: 1
43
+ gradient_clipping: 1.0
44
+ is_resume: false
45
+ learning_rate:
46
+ action_model: 0.0001
47
+ base: 2.5e-05
48
+ qwen_vl_interface: 3.0e-05
49
+ logging_frequency: 100
50
+ lr_scheduler_type: cosine_with_min_lr
51
+ max_train_steps: 50000
52
+ num_warmup_steps: 5000
53
+ optimizer:
54
+ betas:
55
+ - 0.9
56
+ - 0.95
57
+ eps: 1.0e-08
58
+ weight_decay: 1.0e-08
59
+ save_interval: 5000
60
+ scheduler_specific_kwargs:
61
+ min_lr: 1.0e-06
62
+ wandb_entity: your_wandb_entity
63
+ wandb_project: llavavla
dataset_statistics.json ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "franka": {
3
+ "action": {
4
+ "mean": [
5
+ 0.06413520714984491,
6
+ 0.08330044630341807,
7
+ -0.09768853995662471,
8
+ 3.989756358071042e-05,
9
+ 0.0074605874788875775,
10
+ -0.0023408993505514594,
11
+ 0.5327441073380985
12
+ ],
13
+ "std": [
14
+ 0.34435420775153375,
15
+ 0.37664413619011156,
16
+ 0.4530659887664812,
17
+ 0.040994048642083246,
18
+ 0.06561666470959421,
19
+ 0.08020121465404138,
20
+ 0.4990030574544087
21
+ ],
22
+ "max": [
23
+ 0.9375,
24
+ 0.9375,
25
+ 0.9375,
26
+ 0.3557142913341522,
27
+ 0.375,
28
+ 0.375,
29
+ 1.0
30
+ ],
31
+ "min": [
32
+ -0.9375,
33
+ -0.9375,
34
+ -0.9375,
35
+ -0.2582142949104309,
36
+ -0.375,
37
+ -0.3675000071525574,
38
+ 0.0
39
+ ],
40
+ "q01": [
41
+ -0.8785714507102966,
42
+ -0.8758928775787354,
43
+ -0.9375,
44
+ -0.1510714292526245,
45
+ -0.20678570866584778,
46
+ -0.2742857038974762,
47
+ 0.0
48
+ ],
49
+ "q99": [
50
+ 0.9375,
51
+ 0.9107142686843872,
52
+ 0.9375,
53
+ 0.20357142388820648,
54
+ 0.26357144117355347,
55
+ 0.375,
56
+ 1.0
57
+ ],
58
+ "mask": [
59
+ true,
60
+ true,
61
+ true,
62
+ true,
63
+ true,
64
+ true,
65
+ false
66
+ ]
67
+ },
68
+ "state": {
69
+ "mean": [
70
+ -0.051236557416044756,
71
+ 0.03320404933765531,
72
+ 0.7923003721695681,
73
+ 2.9629008036393385,
74
+ -0.15774205785531267,
75
+ -0.13519181454410922,
76
+ 0.02689367747650697,
77
+ -0.027077807328448854
78
+ ],
79
+ "std": [
80
+ 0.10707255982690211,
81
+ 0.14929271152454146,
82
+ 0.37628407735102676,
83
+ 0.3693053742611154,
84
+ 0.8810664091949664,
85
+ 0.33025891946901687,
86
+ 0.01434538622885222,
87
+ 0.01422763564543161
88
+ ],
89
+ "max": [
90
+ 0.21031762659549713,
91
+ 0.39128610491752625,
92
+ 1.3660105466842651,
93
+ 3.6714255809783936,
94
+ 3.560650587081909,
95
+ 1.386339545249939,
96
+ 0.04233968257904053,
97
+ 0.0013633022317662835
98
+ ],
99
+ "min": [
100
+ -0.4828203022480011,
101
+ -0.3255046010017395,
102
+ 0.008128180168569088,
103
+ 0.35277295112609863,
104
+ -3.641430377960205,
105
+ -1.842738389968872,
106
+ -0.0013586411951109767,
107
+ -0.042040832340717316
108
+ ],
109
+ "q01": [
110
+ -0.42401049643754957,
111
+ -0.2838300323486328,
112
+ 0.009925739830359817,
113
+ 1.3085840785503386,
114
+ -2.886677579879761,
115
+ -1.1599004411697387,
116
+ 0.001503719249740243,
117
+ -0.040336399003863335
118
+ ],
119
+ "q99": [
120
+ 0.1530261474847791,
121
+ 0.3629165390133857,
122
+ 1.2910678112506866,
123
+ 3.303542451858519,
124
+ 2.7496529006957933,
125
+ 0.6893712210655194,
126
+ 0.040610933862626555,
127
+ -0.0015016929572448147
128
+ ]
129
+ },
130
+ "num_transitions": 272104,
131
+ "num_trajectories": 1693
132
+ }
133
+ }
summary.jsonl ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {"steps": 5000}
2
+ {"steps": 10000}
3
+ {"steps": 15000}
4
+ {"steps": 20000}
5
+ {"steps": 25000}
6
+ {"steps": 30000}
7
+ {"steps": 35000}
8
+ {"steps": 40000}
9
+ {"steps": 45000}
10
+ {"steps": 50000}