Hugging Face's logo

Chengpengaaa
/
starvla-adapter

Model card Files
xet
 Community
Chengpengaaa commited on
Commit
e1c3956
·
verified ·
1 Parent(s): 8fa8328

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. QwenAdapter-pretrain/checkpoints/steps_53000_pytorch_model.pt +3 -0
  2. QwenAdapter-pretrain/config.json +87 -0
  3. QwenAdapter-pretrain/config.yaml +47 -0
  4. QwenAdapter-pretrain/dataset_statistics.json +218 -0
QwenAdapter-pretrain/checkpoints/steps_53000_pytorch_model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2e363768893670bb0879ef6492e7da0c7f3dfce0bf5150bde25c6dddf83d3ab
3
+ size 6870058775
QwenAdapter-pretrain/config.json ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_id": "QwenAdapter-0130-pre-eep",
3
+ "run_root_dir": "/root/jinchengp/LumosVLA/results",
4
+ "seed": 42,
5
+ "trackers": [
6
+ "jsonl",
7
+ "wandb"
8
+ ],
9
+ "wandb_entity": "xspadex-nanyang-technological-university-singapore",
10
+ "wandb_project": "qwenvl--0130-pre-eep",
11
+ "is_debug": false,
12
+ "framework": {
13
+ "name": "QwenAdapter",
14
+ "qwenvl": {
15
+ "base_vlm": "/root/jinchengp/LumosVLA/base_vl/Qwen3-VL-2B-Instruct",
16
+ "attn_implementation": "flash_attention_2",
17
+ "vl_hidden_dim": 2048
18
+ },
19
+ "action_model": {
20
+ "action_model_type": "VLA-Adapter",
21
+ "hidden_dim": 2048,
22
+ "action_dim": 20,
23
+ "state_dim": 20,
24
+ "action_query_num": 64,
25
+ "use_pro_version": true,
26
+ "use_proprio": true,
27
+ "phase": "Training",
28
+ "num_actions_chunk": 30
29
+ }
30
+ },
31
+ "datasets": {
32
+ "vla_data": {
33
+ "dataset_py": "lerobot_datasets",
34
+ "data_root_dir": "/lumos-vePFS/shenzhen/data/opendata/RoboCoin_T2",
35
+ "data_mix": "pretrain1",
36
+ "per_device_batch_size": 32,
37
+ "num_workers": 2,
38
+ "action_type": "abs_ee",
39
+ "image_size": 256,
40
+ "video_backend": "pyav"
41
+ }
42
+ },
43
+ "trainer": {
44
+ "epochs": 100,
45
+ "max_train_steps": 85000,
46
+ "num_warmup_steps": 1000,
47
+ "save_interval": 1000,
48
+ "eval_interval": 100000,
49
+ "learning_rate": {
50
+ "base": 4e-06,
51
+ "qwen_vl_interface": 4e-06,
52
+ "action_model": 1e-05
53
+ },
54
+ "lr_scheduler_type": "cosine_with_min_lr",
55
+ "scheduler_specific_kwargs": {
56
+ "min_lr": 5e-07
57
+ },
58
+ "freeze_modules": "qwen_vl_interface",
59
+ "loss_scale": {
60
+ "vla": 1.0,
61
+ "vlm": 0.05
62
+ },
63
+ "max_grad_norm": 1.0,
64
+ "warmup_ratio": 0.1,
65
+ "weight_decay": 0.0,
66
+ "logging_frequency": 10,
67
+ "gradient_clipping": 1.0,
68
+ "gradient_accumulation_steps": 1,
69
+ "optimizer": {
70
+ "name": "AdamW",
71
+ "betas": [
72
+ 0.9,
73
+ 0.95
74
+ ],
75
+ "eps": 1e-08,
76
+ "weight_decay": 0
77
+ }
78
+ },
79
+ "is_resume": false,
80
+ "resume_epoch": null,
81
+ "resume_step": null,
82
+ "enable_gradient_checkpointing": true,
83
+ "enable_mixed_precision_training": true,
84
+ "report_to": "wandb",
85
+ "wandb_run_name": "v1",
86
+ "output_dir": "/root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-eep"
87
+ }
QwenAdapter-pretrain/config.yaml ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ datasets:
2
+ vla_data:
3
+ data_mix: pretrain1
4
+ data_root_dir: /lumos-vePFS/shenzhen/data/opendata/RoboCoin_T2
5
+ dataset_py: lerobot_datasets
6
+ num_workers: 2
7
+ per_device_batch_size: 32
8
+ framework:
9
+ action_model:
10
+ action_dim: 20
11
+ action_query_num: 64
12
+ hidden_dim: 2048
13
+ num_actions_chunk: 30
14
+ phase: Training
15
+ state_dim: 20
16
+ use_pro_version: true
17
+ use_proprio: true
18
+ name: QwenAdapter
19
+ qwenvl:
20
+ base_vlm: /root/jinchengp/LumosVLA/base_vl/Qwen3-VL-2B-Instruct
21
+ vl_hidden_dim: 2048
22
+ output_dir: /root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-eep
23
+ run_id: QwenAdapter-0130-pre-eep
24
+ run_root_dir: /root/jinchengp/LumosVLA/results
25
+ seed: 42
26
+ trainer:
27
+ eval_interval: 100000
28
+ freeze_modules: qwen_vl_interface
29
+ gradient_accumulation_steps: 1
30
+ gradient_clipping: 1.0
31
+ learning_rate:
32
+ action_model: 1.0e-05
33
+ base: 4.0e-06
34
+ qwen_vl_interface: 4.0e-06
35
+ logging_frequency: 10
36
+ lr_scheduler_type: cosine_with_min_lr
37
+ max_train_steps: 85000
38
+ num_warmup_steps: 1000
39
+ optimizer:
40
+ betas:
41
+ - 0.9
42
+ - 0.95
43
+ eps: 1.0e-08
44
+ weight_decay: 0
45
+ save_interval: 1000
46
+ scheduler_specific_kwargs:
47
+ min_lr: 5.0e-07
QwenAdapter-pretrain/dataset_statistics.json ADDED
@@ -0,0 +1,218 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "action": {
4
+ "mean": [
5
+ 0.30040545809653507,
6
+ 0.15762381433253933,
7
+ 0.7458027630083024,
8
+ 1.6618751739301993,
9
+ 0.37386574503034353,
10
+ 1.5729571561659539,
11
+ 0.1324957077717221,
12
+ 0.3232377693057061,
13
+ -0.12770604029778512,
14
+ 0.7663230199006297,
15
+ 1.5535530388775856,
16
+ 0.19679608126922,
17
+ 0.7855641188160067,
18
+ 0.17707381911966355
19
+ ],
20
+ "std": [
21
+ 0.720805842563276,
22
+ 0.1614626625306384,
23
+ 1.2188527178068638,
24
+ 1.7018175276694938,
25
+ 0.5481052190649864,
26
+ 1.064226008174834,
27
+ 0.19446469866871616,
28
+ 0.6037348466886293,
29
+ 0.1776571629073011,
30
+ 1.4588972017725848,
31
+ 1.8805491974593895,
32
+ 0.5124725626744526,
33
+ 2.0440070562341166,
34
+ 0.21305108089362437
35
+ ],
36
+ "max": [
37
+ 1895.825439453125,
38
+ 0.5397133827209473,
39
+ 1895.825439453125,
40
+ 3.141589403152466,
41
+ 1.570796251296997,
42
+ 3.141251802444458,
43
+ 0.6944000124931335,
44
+ 1895.825439453125,
45
+ 0.4470269978046417,
46
+ 1895.825439453125,
47
+ 3.141589403152466,
48
+ 1.570796251296997,
49
+ 3.1415886878967285,
50
+ 11.852590560913086
51
+ ],
52
+ "min": [
53
+ -0.11240299791097641,
54
+ -16.7772159576416,
55
+ 0.0,
56
+ -3.1415927410125732,
57
+ -1.5053287744522095,
58
+ -3.141592502593994,
59
+ -0.06369999796152115,
60
+ -0.06729000061750412,
61
+ -0.5366361737251282,
62
+ 0.0,
63
+ -3.1415927410125732,
64
+ -1.5556800365447998,
65
+ -3.141592502593994,
66
+ -10.500527381896973
67
+ ],
68
+ "q01": [
69
+ -0.013505999930202961,
70
+ -0.3756061708927155,
71
+ 0.038277868777513505,
72
+ -3.1395506858825684,
73
+ -1.3926945877075196,
74
+ -3.1397950649261475,
75
+ -0.02240000106394291,
76
+ -0.01436999998986721,
77
+ -0.4361535876989365,
78
+ 0.0,
79
+ -3.140266180038452,
80
+ -0.9948101663589477,
81
+ -3.139096736907959,
82
+ -0.12600000202655792
83
+ ],
84
+ "q99": [
85
+ 0.7761559200286865,
86
+ 0.47395932048559186,
87
+ 1.2645205318927755,
88
+ 3.137927532196045,
89
+ 1.5381237268447876,
90
+ 3.1396028995513916,
91
+ 0.6901999711990356,
92
+ 0.7505726778507233,
93
+ 0.39113301038742065,
94
+ 1.3503002834320068,
95
+ 3.139064886569977,
96
+ 1.5247894525527954,
97
+ 3.137979745864868,
98
+ 0.6761999726295471
99
+ ],
100
+ "mask": [
101
+ true,
102
+ true,
103
+ true,
104
+ true,
105
+ true,
106
+ true,
107
+ false,
108
+ true,
109
+ true,
110
+ true,
111
+ true,
112
+ true,
113
+ true,
114
+ false
115
+ ]
116
+ },
117
+ "state": {
118
+ "mean": [
119
+ 0.30040545809653507,
120
+ 0.15762381433253933,
121
+ 0.7458027630083024,
122
+ 1.6618751739301993,
123
+ 0.37386574503034353,
124
+ 1.5729571561659539,
125
+ 0.1324957077717221,
126
+ 0.3232377693057061,
127
+ -0.12770604029778512,
128
+ 0.7663230199006297,
129
+ 1.5535530388775856,
130
+ 0.19679608126922,
131
+ 0.7855641188160067,
132
+ 0.17707381911966355
133
+ ],
134
+ "std": [
135
+ 0.720805842563276,
136
+ 0.1614626625306384,
137
+ 1.2188527178068638,
138
+ 1.7018175276694938,
139
+ 0.5481052190649864,
140
+ 1.064226008174834,
141
+ 0.19446469866871616,
142
+ 0.6037348466886293,
143
+ 0.1776571629073011,
144
+ 1.4588972017725848,
145
+ 1.8805491974593895,
146
+ 0.5124725626744526,
147
+ 2.0440070562341166,
148
+ 0.21305108089362437
149
+ ],
150
+ "max": [
151
+ 1895.825439453125,
152
+ 0.5397133827209473,
153
+ 1895.825439453125,
154
+ 3.141589403152466,
155
+ 1.570796251296997,
156
+ 3.141251802444458,
157
+ 0.6944000124931335,
158
+ 1895.825439453125,
159
+ 0.4470269978046417,
160
+ 1895.825439453125,
161
+ 3.141589403152466,
162
+ 1.570796251296997,
163
+ 3.1415886878967285,
164
+ 11.852590560913086
165
+ ],
166
+ "min": [
167
+ -0.11240299791097641,
168
+ -16.7772159576416,
169
+ 0.0,
170
+ -3.1415927410125732,
171
+ -1.5053287744522095,
172
+ -3.141592502593994,
173
+ -0.06369999796152115,
174
+ -0.06729000061750412,
175
+ -0.5366361737251282,
176
+ 0.0,
177
+ -3.1415927410125732,
178
+ -1.5556800365447998,
179
+ -3.141592502593994,
180
+ -10.500527381896973
181
+ ],
182
+ "q01": [
183
+ -0.013505999930202961,
184
+ -0.3756061708927155,
185
+ 0.038277868777513505,
186
+ -3.1395506858825684,
187
+ -1.3926945877075196,
188
+ -3.1397950649261475,
189
+ -0.02240000106394291,
190
+ -0.01436999998986721,
191
+ -0.4361535876989365,
192
+ 0.0,
193
+ -3.140266180038452,
194
+ -0.9948101663589477,
195
+ -3.139096736907959,
196
+ -0.12600000202655792
197
+ ],
198
+ "q99": [
199
+ 0.7761559200286865,
200
+ 0.47395932048559186,
201
+ 1.2645205318927755,
202
+ 3.137927532196045,
203
+ 1.5381237268447876,
204
+ 3.1396028995513916,
205
+ 0.6901999711990356,
206
+ 0.7505726778507233,
207
+ 0.39113301038742065,
208
+ 1.3503002834320068,
209
+ 3.139064886569977,
210
+ 1.5247894525527954,
211
+ 3.137979745864868,
212
+ 0.6761999726295471
213
+ ]
214
+ },
215
+ "num_transitions": 7626529,
216
+ "num_trajectories": 13566
217
+ }
218
+ }