Hugging Face's logo

Chengpengaaa
/
starvla-adapter

Model card Files
xet
 Community
Chengpengaaa commited on
Commit
fe6a639
·
verified ·
1 Parent(s): 81ac8f3

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. QwenAdapter-0130-pre-fintune-v/checkpoints/steps_36000_pytorch_model.pt +3 -0
  2. QwenAdapter-0130-pre-fintune-v/config.json +90 -0
  3. QwenAdapter-0130-pre-fintune-v/config.yaml +76 -0
  4. QwenAdapter-0130-pre-fintune-v/dataset_statistics.json +218 -0
QwenAdapter-0130-pre-fintune-v/checkpoints/steps_36000_pytorch_model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89956687beefe57e83ee3a63a3664a22dc68a581d829ff5ccd77ba15145b3d98
3
+ size 6870058775
QwenAdapter-0130-pre-fintune-v/config.json ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_id": "QwenAdapter-0130-pre-fintune+vit-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-fintune+vit-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
+ "framework_py": "QwenAdapter"
31
+ },
32
+ "datasets": {
33
+ "vla_data": {
34
+ "dataset_py": "lerobot_datasets",
35
+ "data_root_dir": "/lumos-vePFS/shenzhen/data/data_warehouse/data_warehouse_output",
36
+ "data_mix": "train_internAdapter",
37
+ "per_device_batch_size": 40,
38
+ "num_workers": 2,
39
+ "action_type": "abs_ee",
40
+ "image_size": 256,
41
+ "video_backend": "pyav"
42
+ }
43
+ },
44
+ "trainer": {
45
+ "epochs": 100,
46
+ "max_train_steps": 85000,
47
+ "num_warmup_steps": 1000,
48
+ "save_interval": 1000,
49
+ "eval_interval": 100000,
50
+ "learning_rate": {
51
+ "base": 4e-06,
52
+ "qwen_vl_interface": 4e-06,
53
+ "action_model": 1e-05
54
+ },
55
+ "lr_scheduler_type": "cosine_with_min_lr",
56
+ "scheduler_specific_kwargs": {
57
+ "min_lr": 5e-07
58
+ },
59
+ "freeze_modules": "qwen_vl_interface.model.model.language_model",
60
+ "loss_scale": {
61
+ "vla": 1.0,
62
+ "vlm": 0.05
63
+ },
64
+ "max_grad_norm": 1.0,
65
+ "warmup_ratio": 0.1,
66
+ "weight_decay": 0.0,
67
+ "logging_frequency": 10,
68
+ "gradient_clipping": 1.0,
69
+ "gradient_accumulation_steps": 1,
70
+ "optimizer": {
71
+ "name": "AdamW",
72
+ "betas": [
73
+ 0.9,
74
+ 0.95
75
+ ],
76
+ "eps": 1e-08,
77
+ "weight_decay": 0
78
+ },
79
+ "is_resume": true,
80
+ "resume_step": 36000
81
+ },
82
+ "is_resume": false,
83
+ "resume_epoch": null,
84
+ "resume_step": null,
85
+ "enable_gradient_checkpointing": true,
86
+ "enable_mixed_precision_training": true,
87
+ "report_to": "wandb",
88
+ "wandb_run_name": "v1",
89
+ "output_dir": "/root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-fintune+vit-eep"
90
+ }
QwenAdapter-0130-pre-fintune-v/config.yaml ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ run_id: QwenAdapter-0130-pre-fintune+vit-eep
2
+ run_root_dir: /root/jinchengp/LumosVLA/results
3
+ seed: 42
4
+ trackers:
5
+ - jsonl
6
+ - wandb
7
+ wandb_entity: xspadex-nanyang-technological-university-singapore
8
+ wandb_project: qwenvl-0130-pre-fintune+vit-eep
9
+ is_debug: false
10
+ framework:
11
+ name: QwenAdapter
12
+ qwenvl:
13
+ base_vlm: /root/jinchengp/LumosVLA/base_vl/Qwen3-VL-2B-Instruct
14
+ attn_implementation: flash_attention_2
15
+ vl_hidden_dim: 2048
16
+ action_model:
17
+ action_model_type: VLA-Adapter
18
+ hidden_dim: 2048
19
+ action_dim: 20
20
+ state_dim: 20
21
+ action_query_num: 64
22
+ use_pro_version: true
23
+ use_proprio: true
24
+ phase: Training
25
+ num_actions_chunk: 30
26
+ framework_py: QwenAdapter
27
+ datasets:
28
+ vla_data:
29
+ dataset_py: lerobot_datasets
30
+ data_root_dir: /lumos-vePFS/shenzhen/data/data_warehouse/data_warehouse_output
31
+ data_mix: train_internAdapter
32
+ per_device_batch_size: 40
33
+ num_workers: 2
34
+ action_type: abs_ee
35
+ image_size: 256
36
+ video_backend: pyav
37
+ trainer:
38
+ epochs: 100
39
+ max_train_steps: 85000
40
+ num_warmup_steps: 1000
41
+ save_interval: 1000
42
+ eval_interval: 100000
43
+ learning_rate:
44
+ base: 4.0e-06
45
+ qwen_vl_interface: 4.0e-06
46
+ action_model: 1.0e-05
47
+ lr_scheduler_type: cosine_with_min_lr
48
+ scheduler_specific_kwargs:
49
+ min_lr: 5.0e-07
50
+ freeze_modules: qwen_vl_interface.model.model.language_model
51
+ loss_scale:
52
+ vla: 1.0
53
+ vlm: 0.05
54
+ max_grad_norm: 1.0
55
+ warmup_ratio: 0.1
56
+ weight_decay: 0.0
57
+ logging_frequency: 10
58
+ gradient_clipping: 1.0
59
+ gradient_accumulation_steps: 1
60
+ optimizer:
61
+ name: AdamW
62
+ betas:
63
+ - 0.9
64
+ - 0.95
65
+ eps: 1.0e-08
66
+ weight_decay: 0
67
+ is_resume: true
68
+ resume_step: 36000
69
+ is_resume: false
70
+ resume_epoch: null
71
+ resume_step: null
72
+ enable_gradient_checkpointing: true
73
+ enable_mixed_precision_training: true
74
+ report_to: wandb
75
+ wandb_run_name: v1
76
+ output_dir: /root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-fintune+vit-eep
QwenAdapter-0130-pre-fintune-v/dataset_statistics.json ADDED
@@ -0,0 +1,218 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "action": {
4
+ "mean": [
5
+ 0.2561876313252883,
6
+ -0.07762468829980318,
7
+ 0.19539409808137206,
8
+ -1.3202253191760092,
9
+ 0.40215678264697385,
10
+ 1.6105433865026992,
11
+ 0.03408320081617797,
12
+ 0.23211244651765528,
13
+ 0.09123119367568779,
14
+ 0.2078852373542208,
15
+ 1.5647810820603014,
16
+ 0.49727785406690656,
17
+ -1.669434749718868,
18
+ 0.0347955935267788
19
+ ],
20
+ "std": [
21
+ 0.09950804698192407,
22
+ 0.10764074013794674,
23
+ 0.06754211620391408,
24
+ 2.5930047780739156,
25
+ 0.31198643775887136,
26
+ 2.2838284773976234,
27
+ 0.02828675372041501,
28
+ 0.08766959345540473,
29
+ 0.07936882807492893,
30
+ 0.05428820254273273,
31
+ 2.4844969574766207,
32
+ 0.27824351328936514,
33
+ 2.273481347354871,
34
+ 0.028344366955933887
35
+ ],
36
+ "max": [
37
+ 0.5931400060653687,
38
+ 0.48442599177360535,
39
+ 0.6967939734458923,
40
+ 3.1415927410125732,
41
+ 1.5707963705062866,
42
+ 3.1415927410125732,
43
+ 0.0810599997639656,
44
+ 0.5580319762229919,
45
+ 0.45939499139785767,
46
+ 0.6456500291824341,
47
+ 3.1415927410125732,
48
+ 1.5707963705062866,
49
+ 3.1415927410125732,
50
+ 0.06069000065326691
51
+ ],
52
+ "min": [
53
+ -0.10788500308990479,
54
+ -0.5191159844398499,
55
+ 0.040369000285863876,
56
+ -3.1410515308380127,
57
+ -1.5302174091339111,
58
+ -3.1410515308380127,
59
+ -0.0017500000540167093,
60
+ -0.03119499981403351,
61
+ -0.32183900475502014,
62
+ 0.06903599947690964,
63
+ -3.1410515308380127,
64
+ -0.7706850171089172,
65
+ -3.1410515308380127,
66
+ 0.0
67
+ ],
68
+ "q01": [
69
+ 0.03489213060587645,
70
+ -0.36912800788879396,
71
+ 0.07184209674596786,
72
+ -3.137246325016022,
73
+ -0.20070483744144438,
74
+ -3.137392144203186,
75
+ 0.0,
76
+ 0.030881709717214108,
77
+ -0.15914200246334076,
78
+ 0.10170399993658066,
79
+ -3.138237874507904,
80
+ -0.11282174646854401,
81
+ -3.1362587141990663,
82
+ 7.000000186963007e-05
83
+ ],
84
+ "q99": [
85
+ 0.5304324626922607,
86
+ 0.41213971734046934,
87
+ 0.47792605459690085,
88
+ 3.1415927410125732,
89
+ 1.5365502309799193,
90
+ 3.1415927410125732,
91
+ 0.06019999831914902,
92
+ 0.4840700030326843,
93
+ 0.34843780994415285,
94
+ 0.42531779676675785,
95
+ 3.1415927410125732,
96
+ 1.5435859155654907,
97
+ 3.1415927410125732,
98
+ 0.05992000177502632
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
+ 0.2561876313252883,
120
+ -0.07762468829980318,
121
+ 0.19539409808137206,
122
+ -1.3202253191760092,
123
+ 0.40215678264697385,
124
+ 1.6105433865026992,
125
+ 0.03408320081617797,
126
+ 0.23211244651765528,
127
+ 0.09123119367568779,
128
+ 0.2078852373542208,
129
+ 1.5647810820603014,
130
+ 0.49727785406690656,
131
+ -1.669434749718868,
132
+ 0.0347955935267788
133
+ ],
134
+ "std": [
135
+ 0.09950804698192407,
136
+ 0.10764074013794674,
137
+ 0.06754211620391408,
138
+ 2.5930047780739156,
139
+ 0.31198643775887136,
140
+ 2.2838284773976234,
141
+ 0.02828675372041501,
142
+ 0.08766959345540473,
143
+ 0.07936882807492893,
144
+ 0.05428820254273273,
145
+ 2.4844969574766207,
146
+ 0.27824351328936514,
147
+ 2.273481347354871,
148
+ 0.028344366955933887
149
+ ],
150
+ "max": [
151
+ 0.5931400060653687,
152
+ 0.48442599177360535,
153
+ 0.6967939734458923,
154
+ 3.1415927410125732,
155
+ 1.5707963705062866,
156
+ 3.1415927410125732,
157
+ 0.0810599997639656,
158
+ 0.5580319762229919,
159
+ 0.45939499139785767,
160
+ 0.6456500291824341,
161
+ 3.1415927410125732,
162
+ 1.5707963705062866,
163
+ 3.1415927410125732,
164
+ 0.06069000065326691
165
+ ],
166
+ "min": [
167
+ -0.10788500308990479,
168
+ -0.5191159844398499,
169
+ 0.040369000285863876,
170
+ -3.1410515308380127,
171
+ -1.5302174091339111,
172
+ -3.1410515308380127,
173
+ -0.0017500000540167093,
174
+ -0.03119499981403351,
175
+ -0.32183900475502014,
176
+ 0.06903599947690964,
177
+ -3.1410515308380127,
178
+ -0.7706850171089172,
179
+ -3.1410515308380127,
180
+ 0.0
181
+ ],
182
+ "q01": [
183
+ 0.03489213060587645,
184
+ -0.36912800788879396,
185
+ 0.07184209674596786,
186
+ -3.137246325016022,
187
+ -0.20070483744144438,
188
+ -3.137392144203186,
189
+ 0.0,
190
+ 0.030881709717214108,
191
+ -0.15914200246334076,
192
+ 0.10170399993658066,
193
+ -3.138237874507904,
194
+ -0.11282174646854401,
195
+ -3.1362587141990663,
196
+ 7.000000186963007e-05
197
+ ],
198
+ "q99": [
199
+ 0.5304324626922607,
200
+ 0.41213971734046934,
201
+ 0.47792605459690085,
202
+ 3.1415927410125732,
203
+ 1.5365502309799193,
204
+ 3.1415927410125732,
205
+ 0.06019999831914902,
206
+ 0.4840700030326843,
207
+ 0.34843780994415285,
208
+ 0.42531779676675785,
209
+ 3.1415927410125732,
210
+ 1.5435859155654907,
211
+ 3.1415927410125732,
212
+ 0.05992000177502632
213
+ ]
214
+ },
215
+ "num_transitions": 2404578,
216
+ "num_trajectories": 1923
217
+ }
218
+ }