LauTttt commited on
Commit
91076eb
·
verified ·
1 Parent(s): fd57126

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +6 -0
  2. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/config.json +122 -0
  3. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/model.safetensors +3 -0
  4. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_postprocessor.json +38 -0
  5. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  6. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_preprocessor.json +103 -0
  7. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  8. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/train_config.json +264 -0
  9. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/config.json +122 -0
  10. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/model.safetensors +3 -0
  11. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_postprocessor.json +38 -0
  12. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  13. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_preprocessor.json +103 -0
  14. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  15. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/train_config.json +264 -0
  16. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/config.json +122 -0
  17. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/model.safetensors +3 -0
  18. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_postprocessor.json +38 -0
  19. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  20. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_preprocessor.json +103 -0
  21. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  22. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/train_config.json +264 -0
  23. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/debug-internal.log +11 -0
  24. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/debug.log +21 -0
  25. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/config.yaml +351 -0
  26. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/output.log +47 -0
  27. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/requirements.txt +167 -0
  28. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/wandb-metadata.json +117 -0
  29. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/wandb-summary.json +1 -0
  30. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug-core.log +15 -0
  31. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug-internal.log +11 -0
  32. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug.log +21 -0
  33. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/run-i0fuukqo.wandb +3 -0
  34. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/config.yaml +351 -0
  35. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/output.log +47 -0
  36. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/requirements.txt +167 -0
  37. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/wandb-metadata.json +117 -0
  38. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/wandb-summary.json +1 -0
  39. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-core.log +15 -0
  40. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-internal.log +11 -0
  41. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug.log +21 -0
  42. robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/run-i0fuukqo.wandb +3 -0
  43. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/config.json +122 -0
  44. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/model.safetensors +3 -0
  45. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_postprocessor.json +38 -0
  46. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  47. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_preprocessor.json +103 -0
  48. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  49. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/train_config.json +264 -0
  50. robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/last/pretrained_model/config.json +122 -0
.gitattributes CHANGED
@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/run-i0fuukqo.wandb filter=lfs diff=lfs merge=lfs -text
37
+ robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/run-i0fuukqo.wandb filter=lfs diff=lfs merge=lfs -text
38
+ robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/wandb/latest-run/run-vmkt7kml.wandb filter=lfs diff=lfs merge=lfs -text
39
+ robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/wandb/run-20261027_224654-vmkt7kml/run-vmkt7kml.wandb filter=lfs diff=lfs merge=lfs -text
40
+ robotwin_grab_roller_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/run-ck732jyy.wandb filter=lfs diff=lfs merge=lfs -text
41
+ robotwin_grab_roller_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_110550-ck732jyy/run-ck732jyy.wandb filter=lfs diff=lfs merge=lfs -text
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "action_evidence_vla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.cam_high": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 480,
9
+ 640,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.cam_left_wrist": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 480,
17
+ 640,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.images.cam_right_wrist": {
22
+ "type": "VISUAL",
23
+ "shape": [
24
+ 480,
25
+ 640,
26
+ 3
27
+ ]
28
+ },
29
+ "observation.state": {
30
+ "type": "STATE",
31
+ "shape": [
32
+ 14
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": false,
48
+ "repo_id": null,
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": null,
53
+ "pi05_init_path": "lerobot/pi05_base",
54
+ "paligemma_variant": "gemma_2b",
55
+ "action_expert_variant": "gemma_300m",
56
+ "pi05_precision": "bfloat16",
57
+ "chunk_size": 50,
58
+ "n_action_steps": 10,
59
+ "max_state_dim": 32,
60
+ "max_action_dim": 32,
61
+ "tokenizer_max_length": 200,
62
+ "image_resolution": [
63
+ 224,
64
+ 224
65
+ ],
66
+ "empty_cameras": 0,
67
+ "normalization_mapping": {
68
+ "VISUAL": "IDENTITY",
69
+ "STATE": "QUANTILES",
70
+ "ACTION": "QUANTILES"
71
+ },
72
+ "num_inference_steps": 10,
73
+ "time_sampling_beta_alpha": 1.5,
74
+ "time_sampling_beta_beta": 1.0,
75
+ "time_sampling_scale": 0.999,
76
+ "time_sampling_offset": 0.001,
77
+ "min_period": 0.004,
78
+ "max_period": 4.0,
79
+ "freeze_vlm": false,
80
+ "freeze_vision_encoder": false,
81
+ "gradient_checkpointing": true,
82
+ "offload_frozen_prefix_activations": true,
83
+ "phase_state_camera_index": 0,
84
+ "phase_state_frame_stride": 2,
85
+ "phase_state_vision_chunk_size": 32,
86
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
87
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
88
+ "phase_state_cache_episodes": 2,
89
+ "phase_state_max_steps": 1000,
90
+ "phase_state_checkpoint_strict": true,
91
+ "freeze_phase_state_encoder": true,
92
+ "phase_state_action_normalization": "QUANTILES",
93
+ "phase_state_action_dim": 14,
94
+ "phase_state_dim": 256,
95
+ "phase_state_layers": 4,
96
+ "phase_state_dropout": 0.1,
97
+ "phase_state_vision_pretrained": false,
98
+ "selector_heads": 8,
99
+ "evidence_dim": 256,
100
+ "use_relative_actions": false,
101
+ "relative_exclude_joints": [
102
+ "gripper"
103
+ ],
104
+ "action_feature_names": [
105
+ "motors"
106
+ ],
107
+ "optimizer_lr": 2.5e-05,
108
+ "evidence_optimizer_lr": 0.0001,
109
+ "action_optimizer_lr": 2.5e-05,
110
+ "vlm_optimizer_lr": 2.5e-05,
111
+ "vision_optimizer_lr": 2.5e-05,
112
+ "optimizer_betas": [
113
+ 0.9,
114
+ 0.95
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 0.01,
118
+ "optimizer_grad_clip_norm": 1.0,
119
+ "scheduler_warmup_steps": 33,
120
+ "scheduler_decay_steps": 1000,
121
+ "scheduler_decay_lr": 2.5e-06
122
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:994c07f7c29dfdbd54550494f37a8313c6b2a3e9896323597aad20e627812232
3
+ size 14437807336
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 14
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "IDENTITY",
18
+ "STATE": "QUANTILES",
19
+ "ACTION": "QUANTILES"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "absolute_actions_processor",
26
+ "config": {
27
+ "enabled": false
28
+ }
29
+ },
30
+ {
31
+ "registry_name": "device_processor",
32
+ "config": {
33
+ "device": "cpu",
34
+ "float_dtype": null
35
+ }
36
+ }
37
+ ]
38
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cce40a2ab3fca0b1c79fbc2a28185c37a6cd8529cbf3a6fde4ae07e804d867c
3
+ size 6792
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "delta_actions_processor",
16
+ "config": {
17
+ "enabled": false,
18
+ "exclude_joints": [
19
+ "gripper"
20
+ ],
21
+ "action_names": [
22
+ "motors"
23
+ ]
24
+ }
25
+ },
26
+ {
27
+ "registry_name": "normalizer_processor",
28
+ "config": {
29
+ "eps": 1e-08,
30
+ "features": {
31
+ "observation.images.cam_high": {
32
+ "type": "VISUAL",
33
+ "shape": [
34
+ 480,
35
+ 640,
36
+ 3
37
+ ]
38
+ },
39
+ "observation.images.cam_left_wrist": {
40
+ "type": "VISUAL",
41
+ "shape": [
42
+ 480,
43
+ 640,
44
+ 3
45
+ ]
46
+ },
47
+ "observation.images.cam_right_wrist": {
48
+ "type": "VISUAL",
49
+ "shape": [
50
+ 480,
51
+ 640,
52
+ 3
53
+ ]
54
+ },
55
+ "observation.state": {
56
+ "type": "STATE",
57
+ "shape": [
58
+ 14
59
+ ]
60
+ },
61
+ "action": {
62
+ "type": "ACTION",
63
+ "shape": [
64
+ 14
65
+ ]
66
+ }
67
+ },
68
+ "norm_map": {
69
+ "VISUAL": "IDENTITY",
70
+ "STATE": "QUANTILES",
71
+ "ACTION": "QUANTILES"
72
+ }
73
+ },
74
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
75
+ },
76
+ {
77
+ "registry_name": "action_evidence_prepare_anchor",
78
+ "config": {}
79
+ },
80
+ {
81
+ "registry_name": "action_evidence_prepare_pi05_prompt",
82
+ "config": {}
83
+ },
84
+ {
85
+ "registry_name": "tokenizer_processor",
86
+ "config": {
87
+ "max_length": 200,
88
+ "task_key": "task",
89
+ "padding_side": "right",
90
+ "padding": "max_length",
91
+ "truncation": true,
92
+ "tokenizer_name": "google/paligemma-3b-pt-224"
93
+ }
94
+ },
95
+ {
96
+ "registry_name": "device_processor",
97
+ "config": {
98
+ "device": "cuda",
99
+ "float_dtype": null
100
+ }
101
+ }
102
+ ]
103
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3590ff2a808dc009fed64697b14e573ac7b848b7e8164f37fa865fb6717bae93
3
+ size 6808
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/train_config.json ADDED
@@ -0,0 +1,264 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/robotwin_adjust_bottle_aevla50",
4
+ "root": "/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50",
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "v3.0",
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "pyav",
80
+ "return_uint8": false,
81
+ "streaming": false
82
+ },
83
+ "env": null,
84
+ "policy": {
85
+ "type": "action_evidence_vla",
86
+ "n_obs_steps": 1,
87
+ "input_features": {
88
+ "observation.images.cam_high": {
89
+ "type": "VISUAL",
90
+ "shape": [
91
+ 480,
92
+ 640,
93
+ 3
94
+ ]
95
+ },
96
+ "observation.images.cam_left_wrist": {
97
+ "type": "VISUAL",
98
+ "shape": [
99
+ 480,
100
+ 640,
101
+ 3
102
+ ]
103
+ },
104
+ "observation.images.cam_right_wrist": {
105
+ "type": "VISUAL",
106
+ "shape": [
107
+ 480,
108
+ 640,
109
+ 3
110
+ ]
111
+ },
112
+ "observation.state": {
113
+ "type": "STATE",
114
+ "shape": [
115
+ 14
116
+ ]
117
+ }
118
+ },
119
+ "output_features": {
120
+ "action": {
121
+ "type": "ACTION",
122
+ "shape": [
123
+ 14
124
+ ]
125
+ }
126
+ },
127
+ "device": "cuda",
128
+ "use_amp": false,
129
+ "use_peft": false,
130
+ "push_to_hub": false,
131
+ "repo_id": null,
132
+ "private": null,
133
+ "tags": null,
134
+ "license": null,
135
+ "pretrained_path": null,
136
+ "pi05_init_path": "lerobot/pi05_base",
137
+ "paligemma_variant": "gemma_2b",
138
+ "action_expert_variant": "gemma_300m",
139
+ "pi05_precision": "bfloat16",
140
+ "chunk_size": 50,
141
+ "n_action_steps": 10,
142
+ "max_state_dim": 32,
143
+ "max_action_dim": 32,
144
+ "tokenizer_max_length": 200,
145
+ "image_resolution": [
146
+ 224,
147
+ 224
148
+ ],
149
+ "empty_cameras": 0,
150
+ "normalization_mapping": {
151
+ "VISUAL": "IDENTITY",
152
+ "STATE": "QUANTILES",
153
+ "ACTION": "QUANTILES"
154
+ },
155
+ "num_inference_steps": 10,
156
+ "time_sampling_beta_alpha": 1.5,
157
+ "time_sampling_beta_beta": 1.0,
158
+ "time_sampling_scale": 0.999,
159
+ "time_sampling_offset": 0.001,
160
+ "min_period": 0.004,
161
+ "max_period": 4.0,
162
+ "freeze_vlm": false,
163
+ "freeze_vision_encoder": false,
164
+ "gradient_checkpointing": true,
165
+ "offload_frozen_prefix_activations": true,
166
+ "phase_state_camera_index": 0,
167
+ "phase_state_frame_stride": 2,
168
+ "phase_state_vision_chunk_size": 32,
169
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
170
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
171
+ "phase_state_cache_episodes": 2,
172
+ "phase_state_max_steps": 1000,
173
+ "phase_state_checkpoint_strict": true,
174
+ "freeze_phase_state_encoder": true,
175
+ "phase_state_action_normalization": "QUANTILES",
176
+ "phase_state_action_dim": 14,
177
+ "phase_state_dim": 256,
178
+ "phase_state_layers": 4,
179
+ "phase_state_dropout": 0.1,
180
+ "phase_state_vision_pretrained": false,
181
+ "selector_heads": 8,
182
+ "evidence_dim": 256,
183
+ "use_relative_actions": false,
184
+ "relative_exclude_joints": [
185
+ "gripper"
186
+ ],
187
+ "action_feature_names": [
188
+ "motors"
189
+ ],
190
+ "optimizer_lr": 2.5e-05,
191
+ "evidence_optimizer_lr": 0.0001,
192
+ "action_optimizer_lr": 2.5e-05,
193
+ "vlm_optimizer_lr": 2.5e-05,
194
+ "vision_optimizer_lr": 2.5e-05,
195
+ "optimizer_betas": [
196
+ 0.9,
197
+ 0.95
198
+ ],
199
+ "optimizer_eps": 1e-08,
200
+ "optimizer_weight_decay": 0.01,
201
+ "optimizer_grad_clip_norm": 1.0,
202
+ "scheduler_warmup_steps": 33,
203
+ "scheduler_decay_steps": 1000,
204
+ "scheduler_decay_lr": 2.5e-06
205
+ },
206
+ "reward_model": null,
207
+ "output_dir": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
208
+ "job_name": "robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
209
+ "resume": false,
210
+ "seed": 42,
211
+ "cudnn_deterministic": false,
212
+ "num_workers": 4,
213
+ "batch_size": 16,
214
+ "fsdp": true,
215
+ "fsdp_save_training_state": false,
216
+ "prefetch_factor": 2,
217
+ "persistent_workers": true,
218
+ "steps": 1000,
219
+ "eval_freq": 0,
220
+ "log_freq": 100,
221
+ "tolerance_s": 0.0001,
222
+ "save_checkpoint": true,
223
+ "save_freq": 100,
224
+ "use_policy_training_preset": true,
225
+ "optimizer": {
226
+ "type": "adamw",
227
+ "lr": 2.5e-05,
228
+ "weight_decay": 0.01,
229
+ "grad_clip_norm": 1.0,
230
+ "betas": [
231
+ 0.9,
232
+ 0.95
233
+ ],
234
+ "eps": 1e-08
235
+ },
236
+ "scheduler": {
237
+ "type": "cosine_decay_with_warmup",
238
+ "num_warmup_steps": 33,
239
+ "num_decay_steps": 1000,
240
+ "peak_lr": 2.5e-05,
241
+ "decay_lr": 2.5e-06
242
+ },
243
+ "eval": {
244
+ "n_episodes": 50,
245
+ "batch_size": 50,
246
+ "use_async_envs": true,
247
+ "max_episodes_rendered": 10,
248
+ "reseed_policy_each_episode": false
249
+ },
250
+ "wandb": {
251
+ "enable": true,
252
+ "disable_artifact": true,
253
+ "project": "action-evidence-vla-pi05-phase",
254
+ "entity": null,
255
+ "notes": null,
256
+ "run_id": "i0fuukqo",
257
+ "mode": "online",
258
+ "add_tags": true
259
+ },
260
+ "peft": null,
261
+ "sample_weighting": null,
262
+ "rename_map": {},
263
+ "checkpoint_path": null
264
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "action_evidence_vla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.cam_high": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 480,
9
+ 640,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.cam_left_wrist": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 480,
17
+ 640,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.images.cam_right_wrist": {
22
+ "type": "VISUAL",
23
+ "shape": [
24
+ 480,
25
+ 640,
26
+ 3
27
+ ]
28
+ },
29
+ "observation.state": {
30
+ "type": "STATE",
31
+ "shape": [
32
+ 14
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": false,
48
+ "repo_id": null,
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": null,
53
+ "pi05_init_path": "lerobot/pi05_base",
54
+ "paligemma_variant": "gemma_2b",
55
+ "action_expert_variant": "gemma_300m",
56
+ "pi05_precision": "bfloat16",
57
+ "chunk_size": 50,
58
+ "n_action_steps": 10,
59
+ "max_state_dim": 32,
60
+ "max_action_dim": 32,
61
+ "tokenizer_max_length": 200,
62
+ "image_resolution": [
63
+ 224,
64
+ 224
65
+ ],
66
+ "empty_cameras": 0,
67
+ "normalization_mapping": {
68
+ "VISUAL": "IDENTITY",
69
+ "STATE": "QUANTILES",
70
+ "ACTION": "QUANTILES"
71
+ },
72
+ "num_inference_steps": 10,
73
+ "time_sampling_beta_alpha": 1.5,
74
+ "time_sampling_beta_beta": 1.0,
75
+ "time_sampling_scale": 0.999,
76
+ "time_sampling_offset": 0.001,
77
+ "min_period": 0.004,
78
+ "max_period": 4.0,
79
+ "freeze_vlm": false,
80
+ "freeze_vision_encoder": false,
81
+ "gradient_checkpointing": true,
82
+ "offload_frozen_prefix_activations": true,
83
+ "phase_state_camera_index": 0,
84
+ "phase_state_frame_stride": 2,
85
+ "phase_state_vision_chunk_size": 32,
86
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
87
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
88
+ "phase_state_cache_episodes": 2,
89
+ "phase_state_max_steps": 1000,
90
+ "phase_state_checkpoint_strict": true,
91
+ "freeze_phase_state_encoder": true,
92
+ "phase_state_action_normalization": "QUANTILES",
93
+ "phase_state_action_dim": 14,
94
+ "phase_state_dim": 256,
95
+ "phase_state_layers": 4,
96
+ "phase_state_dropout": 0.1,
97
+ "phase_state_vision_pretrained": false,
98
+ "selector_heads": 8,
99
+ "evidence_dim": 256,
100
+ "use_relative_actions": false,
101
+ "relative_exclude_joints": [
102
+ "gripper"
103
+ ],
104
+ "action_feature_names": [
105
+ "motors"
106
+ ],
107
+ "optimizer_lr": 2.5e-05,
108
+ "evidence_optimizer_lr": 0.0001,
109
+ "action_optimizer_lr": 2.5e-05,
110
+ "vlm_optimizer_lr": 2.5e-05,
111
+ "vision_optimizer_lr": 2.5e-05,
112
+ "optimizer_betas": [
113
+ 0.9,
114
+ 0.95
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 0.01,
118
+ "optimizer_grad_clip_norm": 1.0,
119
+ "scheduler_warmup_steps": 33,
120
+ "scheduler_decay_steps": 1000,
121
+ "scheduler_decay_lr": 2.5e-06
122
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65f704b0acbfda7232571b8de76bac07dd4868cd16e5b2176bf58ba21ea2575c
3
+ size 14437807336
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 14
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "IDENTITY",
18
+ "STATE": "QUANTILES",
19
+ "ACTION": "QUANTILES"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "absolute_actions_processor",
26
+ "config": {
27
+ "enabled": false
28
+ }
29
+ },
30
+ {
31
+ "registry_name": "device_processor",
32
+ "config": {
33
+ "device": "cpu",
34
+ "float_dtype": null
35
+ }
36
+ }
37
+ ]
38
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cce40a2ab3fca0b1c79fbc2a28185c37a6cd8529cbf3a6fde4ae07e804d867c
3
+ size 6792
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "delta_actions_processor",
16
+ "config": {
17
+ "enabled": false,
18
+ "exclude_joints": [
19
+ "gripper"
20
+ ],
21
+ "action_names": [
22
+ "motors"
23
+ ]
24
+ }
25
+ },
26
+ {
27
+ "registry_name": "normalizer_processor",
28
+ "config": {
29
+ "eps": 1e-08,
30
+ "features": {
31
+ "observation.images.cam_high": {
32
+ "type": "VISUAL",
33
+ "shape": [
34
+ 480,
35
+ 640,
36
+ 3
37
+ ]
38
+ },
39
+ "observation.images.cam_left_wrist": {
40
+ "type": "VISUAL",
41
+ "shape": [
42
+ 480,
43
+ 640,
44
+ 3
45
+ ]
46
+ },
47
+ "observation.images.cam_right_wrist": {
48
+ "type": "VISUAL",
49
+ "shape": [
50
+ 480,
51
+ 640,
52
+ 3
53
+ ]
54
+ },
55
+ "observation.state": {
56
+ "type": "STATE",
57
+ "shape": [
58
+ 14
59
+ ]
60
+ },
61
+ "action": {
62
+ "type": "ACTION",
63
+ "shape": [
64
+ 14
65
+ ]
66
+ }
67
+ },
68
+ "norm_map": {
69
+ "VISUAL": "IDENTITY",
70
+ "STATE": "QUANTILES",
71
+ "ACTION": "QUANTILES"
72
+ }
73
+ },
74
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
75
+ },
76
+ {
77
+ "registry_name": "action_evidence_prepare_anchor",
78
+ "config": {}
79
+ },
80
+ {
81
+ "registry_name": "action_evidence_prepare_pi05_prompt",
82
+ "config": {}
83
+ },
84
+ {
85
+ "registry_name": "tokenizer_processor",
86
+ "config": {
87
+ "max_length": 200,
88
+ "task_key": "task",
89
+ "padding_side": "right",
90
+ "padding": "max_length",
91
+ "truncation": true,
92
+ "tokenizer_name": "google/paligemma-3b-pt-224"
93
+ }
94
+ },
95
+ {
96
+ "registry_name": "device_processor",
97
+ "config": {
98
+ "device": "cuda",
99
+ "float_dtype": null
100
+ }
101
+ }
102
+ ]
103
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3590ff2a808dc009fed64697b14e573ac7b848b7e8164f37fa865fb6717bae93
3
+ size 6808
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,264 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/robotwin_adjust_bottle_aevla50",
4
+ "root": "/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50",
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "v3.0",
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "pyav",
80
+ "return_uint8": false,
81
+ "streaming": false
82
+ },
83
+ "env": null,
84
+ "policy": {
85
+ "type": "action_evidence_vla",
86
+ "n_obs_steps": 1,
87
+ "input_features": {
88
+ "observation.images.cam_high": {
89
+ "type": "VISUAL",
90
+ "shape": [
91
+ 480,
92
+ 640,
93
+ 3
94
+ ]
95
+ },
96
+ "observation.images.cam_left_wrist": {
97
+ "type": "VISUAL",
98
+ "shape": [
99
+ 480,
100
+ 640,
101
+ 3
102
+ ]
103
+ },
104
+ "observation.images.cam_right_wrist": {
105
+ "type": "VISUAL",
106
+ "shape": [
107
+ 480,
108
+ 640,
109
+ 3
110
+ ]
111
+ },
112
+ "observation.state": {
113
+ "type": "STATE",
114
+ "shape": [
115
+ 14
116
+ ]
117
+ }
118
+ },
119
+ "output_features": {
120
+ "action": {
121
+ "type": "ACTION",
122
+ "shape": [
123
+ 14
124
+ ]
125
+ }
126
+ },
127
+ "device": "cuda",
128
+ "use_amp": false,
129
+ "use_peft": false,
130
+ "push_to_hub": false,
131
+ "repo_id": null,
132
+ "private": null,
133
+ "tags": null,
134
+ "license": null,
135
+ "pretrained_path": null,
136
+ "pi05_init_path": "lerobot/pi05_base",
137
+ "paligemma_variant": "gemma_2b",
138
+ "action_expert_variant": "gemma_300m",
139
+ "pi05_precision": "bfloat16",
140
+ "chunk_size": 50,
141
+ "n_action_steps": 10,
142
+ "max_state_dim": 32,
143
+ "max_action_dim": 32,
144
+ "tokenizer_max_length": 200,
145
+ "image_resolution": [
146
+ 224,
147
+ 224
148
+ ],
149
+ "empty_cameras": 0,
150
+ "normalization_mapping": {
151
+ "VISUAL": "IDENTITY",
152
+ "STATE": "QUANTILES",
153
+ "ACTION": "QUANTILES"
154
+ },
155
+ "num_inference_steps": 10,
156
+ "time_sampling_beta_alpha": 1.5,
157
+ "time_sampling_beta_beta": 1.0,
158
+ "time_sampling_scale": 0.999,
159
+ "time_sampling_offset": 0.001,
160
+ "min_period": 0.004,
161
+ "max_period": 4.0,
162
+ "freeze_vlm": false,
163
+ "freeze_vision_encoder": false,
164
+ "gradient_checkpointing": true,
165
+ "offload_frozen_prefix_activations": true,
166
+ "phase_state_camera_index": 0,
167
+ "phase_state_frame_stride": 2,
168
+ "phase_state_vision_chunk_size": 32,
169
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
170
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
171
+ "phase_state_cache_episodes": 2,
172
+ "phase_state_max_steps": 1000,
173
+ "phase_state_checkpoint_strict": true,
174
+ "freeze_phase_state_encoder": true,
175
+ "phase_state_action_normalization": "QUANTILES",
176
+ "phase_state_action_dim": 14,
177
+ "phase_state_dim": 256,
178
+ "phase_state_layers": 4,
179
+ "phase_state_dropout": 0.1,
180
+ "phase_state_vision_pretrained": false,
181
+ "selector_heads": 8,
182
+ "evidence_dim": 256,
183
+ "use_relative_actions": false,
184
+ "relative_exclude_joints": [
185
+ "gripper"
186
+ ],
187
+ "action_feature_names": [
188
+ "motors"
189
+ ],
190
+ "optimizer_lr": 2.5e-05,
191
+ "evidence_optimizer_lr": 0.0001,
192
+ "action_optimizer_lr": 2.5e-05,
193
+ "vlm_optimizer_lr": 2.5e-05,
194
+ "vision_optimizer_lr": 2.5e-05,
195
+ "optimizer_betas": [
196
+ 0.9,
197
+ 0.95
198
+ ],
199
+ "optimizer_eps": 1e-08,
200
+ "optimizer_weight_decay": 0.01,
201
+ "optimizer_grad_clip_norm": 1.0,
202
+ "scheduler_warmup_steps": 33,
203
+ "scheduler_decay_steps": 1000,
204
+ "scheduler_decay_lr": 2.5e-06
205
+ },
206
+ "reward_model": null,
207
+ "output_dir": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
208
+ "job_name": "robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
209
+ "resume": false,
210
+ "seed": 42,
211
+ "cudnn_deterministic": false,
212
+ "num_workers": 4,
213
+ "batch_size": 16,
214
+ "fsdp": true,
215
+ "fsdp_save_training_state": false,
216
+ "prefetch_factor": 2,
217
+ "persistent_workers": true,
218
+ "steps": 1000,
219
+ "eval_freq": 0,
220
+ "log_freq": 100,
221
+ "tolerance_s": 0.0001,
222
+ "save_checkpoint": true,
223
+ "save_freq": 100,
224
+ "use_policy_training_preset": true,
225
+ "optimizer": {
226
+ "type": "adamw",
227
+ "lr": 2.5e-05,
228
+ "weight_decay": 0.01,
229
+ "grad_clip_norm": 1.0,
230
+ "betas": [
231
+ 0.9,
232
+ 0.95
233
+ ],
234
+ "eps": 1e-08
235
+ },
236
+ "scheduler": {
237
+ "type": "cosine_decay_with_warmup",
238
+ "num_warmup_steps": 33,
239
+ "num_decay_steps": 1000,
240
+ "peak_lr": 2.5e-05,
241
+ "decay_lr": 2.5e-06
242
+ },
243
+ "eval": {
244
+ "n_episodes": 50,
245
+ "batch_size": 50,
246
+ "use_async_envs": true,
247
+ "max_episodes_rendered": 10,
248
+ "reseed_policy_each_episode": false
249
+ },
250
+ "wandb": {
251
+ "enable": true,
252
+ "disable_artifact": true,
253
+ "project": "action-evidence-vla-pi05-phase",
254
+ "entity": null,
255
+ "notes": null,
256
+ "run_id": "i0fuukqo",
257
+ "mode": "online",
258
+ "add_tags": true
259
+ },
260
+ "peft": null,
261
+ "sample_weighting": null,
262
+ "rename_map": {},
263
+ "checkpoint_path": null
264
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "action_evidence_vla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.cam_high": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 480,
9
+ 640,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.cam_left_wrist": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 480,
17
+ 640,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.images.cam_right_wrist": {
22
+ "type": "VISUAL",
23
+ "shape": [
24
+ 480,
25
+ 640,
26
+ 3
27
+ ]
28
+ },
29
+ "observation.state": {
30
+ "type": "STATE",
31
+ "shape": [
32
+ 14
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": false,
48
+ "repo_id": null,
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": null,
53
+ "pi05_init_path": "lerobot/pi05_base",
54
+ "paligemma_variant": "gemma_2b",
55
+ "action_expert_variant": "gemma_300m",
56
+ "pi05_precision": "bfloat16",
57
+ "chunk_size": 50,
58
+ "n_action_steps": 10,
59
+ "max_state_dim": 32,
60
+ "max_action_dim": 32,
61
+ "tokenizer_max_length": 200,
62
+ "image_resolution": [
63
+ 224,
64
+ 224
65
+ ],
66
+ "empty_cameras": 0,
67
+ "normalization_mapping": {
68
+ "VISUAL": "IDENTITY",
69
+ "STATE": "QUANTILES",
70
+ "ACTION": "QUANTILES"
71
+ },
72
+ "num_inference_steps": 10,
73
+ "time_sampling_beta_alpha": 1.5,
74
+ "time_sampling_beta_beta": 1.0,
75
+ "time_sampling_scale": 0.999,
76
+ "time_sampling_offset": 0.001,
77
+ "min_period": 0.004,
78
+ "max_period": 4.0,
79
+ "freeze_vlm": false,
80
+ "freeze_vision_encoder": false,
81
+ "gradient_checkpointing": true,
82
+ "offload_frozen_prefix_activations": true,
83
+ "phase_state_camera_index": 0,
84
+ "phase_state_frame_stride": 2,
85
+ "phase_state_vision_chunk_size": 32,
86
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
87
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
88
+ "phase_state_cache_episodes": 2,
89
+ "phase_state_max_steps": 1000,
90
+ "phase_state_checkpoint_strict": true,
91
+ "freeze_phase_state_encoder": true,
92
+ "phase_state_action_normalization": "QUANTILES",
93
+ "phase_state_action_dim": 14,
94
+ "phase_state_dim": 256,
95
+ "phase_state_layers": 4,
96
+ "phase_state_dropout": 0.1,
97
+ "phase_state_vision_pretrained": false,
98
+ "selector_heads": 8,
99
+ "evidence_dim": 256,
100
+ "use_relative_actions": false,
101
+ "relative_exclude_joints": [
102
+ "gripper"
103
+ ],
104
+ "action_feature_names": [
105
+ "motors"
106
+ ],
107
+ "optimizer_lr": 2.5e-05,
108
+ "evidence_optimizer_lr": 0.0001,
109
+ "action_optimizer_lr": 2.5e-05,
110
+ "vlm_optimizer_lr": 2.5e-05,
111
+ "vision_optimizer_lr": 2.5e-05,
112
+ "optimizer_betas": [
113
+ 0.9,
114
+ 0.95
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 0.01,
118
+ "optimizer_grad_clip_norm": 1.0,
119
+ "scheduler_warmup_steps": 33,
120
+ "scheduler_decay_steps": 1000,
121
+ "scheduler_decay_lr": 2.5e-06
122
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65f704b0acbfda7232571b8de76bac07dd4868cd16e5b2176bf58ba21ea2575c
3
+ size 14437807336
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 14
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "IDENTITY",
18
+ "STATE": "QUANTILES",
19
+ "ACTION": "QUANTILES"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "absolute_actions_processor",
26
+ "config": {
27
+ "enabled": false
28
+ }
29
+ },
30
+ {
31
+ "registry_name": "device_processor",
32
+ "config": {
33
+ "device": "cpu",
34
+ "float_dtype": null
35
+ }
36
+ }
37
+ ]
38
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cce40a2ab3fca0b1c79fbc2a28185c37a6cd8529cbf3a6fde4ae07e804d867c
3
+ size 6792
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "delta_actions_processor",
16
+ "config": {
17
+ "enabled": false,
18
+ "exclude_joints": [
19
+ "gripper"
20
+ ],
21
+ "action_names": [
22
+ "motors"
23
+ ]
24
+ }
25
+ },
26
+ {
27
+ "registry_name": "normalizer_processor",
28
+ "config": {
29
+ "eps": 1e-08,
30
+ "features": {
31
+ "observation.images.cam_high": {
32
+ "type": "VISUAL",
33
+ "shape": [
34
+ 480,
35
+ 640,
36
+ 3
37
+ ]
38
+ },
39
+ "observation.images.cam_left_wrist": {
40
+ "type": "VISUAL",
41
+ "shape": [
42
+ 480,
43
+ 640,
44
+ 3
45
+ ]
46
+ },
47
+ "observation.images.cam_right_wrist": {
48
+ "type": "VISUAL",
49
+ "shape": [
50
+ 480,
51
+ 640,
52
+ 3
53
+ ]
54
+ },
55
+ "observation.state": {
56
+ "type": "STATE",
57
+ "shape": [
58
+ 14
59
+ ]
60
+ },
61
+ "action": {
62
+ "type": "ACTION",
63
+ "shape": [
64
+ 14
65
+ ]
66
+ }
67
+ },
68
+ "norm_map": {
69
+ "VISUAL": "IDENTITY",
70
+ "STATE": "QUANTILES",
71
+ "ACTION": "QUANTILES"
72
+ }
73
+ },
74
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
75
+ },
76
+ {
77
+ "registry_name": "action_evidence_prepare_anchor",
78
+ "config": {}
79
+ },
80
+ {
81
+ "registry_name": "action_evidence_prepare_pi05_prompt",
82
+ "config": {}
83
+ },
84
+ {
85
+ "registry_name": "tokenizer_processor",
86
+ "config": {
87
+ "max_length": 200,
88
+ "task_key": "task",
89
+ "padding_side": "right",
90
+ "padding": "max_length",
91
+ "truncation": true,
92
+ "tokenizer_name": "google/paligemma-3b-pt-224"
93
+ }
94
+ },
95
+ {
96
+ "registry_name": "device_processor",
97
+ "config": {
98
+ "device": "cuda",
99
+ "float_dtype": null
100
+ }
101
+ }
102
+ ]
103
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3590ff2a808dc009fed64697b14e573ac7b848b7e8164f37fa865fb6717bae93
3
+ size 6808
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/last/pretrained_model/train_config.json ADDED
@@ -0,0 +1,264 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/robotwin_adjust_bottle_aevla50",
4
+ "root": "/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50",
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "v3.0",
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "pyav",
80
+ "return_uint8": false,
81
+ "streaming": false
82
+ },
83
+ "env": null,
84
+ "policy": {
85
+ "type": "action_evidence_vla",
86
+ "n_obs_steps": 1,
87
+ "input_features": {
88
+ "observation.images.cam_high": {
89
+ "type": "VISUAL",
90
+ "shape": [
91
+ 480,
92
+ 640,
93
+ 3
94
+ ]
95
+ },
96
+ "observation.images.cam_left_wrist": {
97
+ "type": "VISUAL",
98
+ "shape": [
99
+ 480,
100
+ 640,
101
+ 3
102
+ ]
103
+ },
104
+ "observation.images.cam_right_wrist": {
105
+ "type": "VISUAL",
106
+ "shape": [
107
+ 480,
108
+ 640,
109
+ 3
110
+ ]
111
+ },
112
+ "observation.state": {
113
+ "type": "STATE",
114
+ "shape": [
115
+ 14
116
+ ]
117
+ }
118
+ },
119
+ "output_features": {
120
+ "action": {
121
+ "type": "ACTION",
122
+ "shape": [
123
+ 14
124
+ ]
125
+ }
126
+ },
127
+ "device": "cuda",
128
+ "use_amp": false,
129
+ "use_peft": false,
130
+ "push_to_hub": false,
131
+ "repo_id": null,
132
+ "private": null,
133
+ "tags": null,
134
+ "license": null,
135
+ "pretrained_path": null,
136
+ "pi05_init_path": "lerobot/pi05_base",
137
+ "paligemma_variant": "gemma_2b",
138
+ "action_expert_variant": "gemma_300m",
139
+ "pi05_precision": "bfloat16",
140
+ "chunk_size": 50,
141
+ "n_action_steps": 10,
142
+ "max_state_dim": 32,
143
+ "max_action_dim": 32,
144
+ "tokenizer_max_length": 200,
145
+ "image_resolution": [
146
+ 224,
147
+ 224
148
+ ],
149
+ "empty_cameras": 0,
150
+ "normalization_mapping": {
151
+ "VISUAL": "IDENTITY",
152
+ "STATE": "QUANTILES",
153
+ "ACTION": "QUANTILES"
154
+ },
155
+ "num_inference_steps": 10,
156
+ "time_sampling_beta_alpha": 1.5,
157
+ "time_sampling_beta_beta": 1.0,
158
+ "time_sampling_scale": 0.999,
159
+ "time_sampling_offset": 0.001,
160
+ "min_period": 0.004,
161
+ "max_period": 4.0,
162
+ "freeze_vlm": false,
163
+ "freeze_vision_encoder": false,
164
+ "gradient_checkpointing": true,
165
+ "offload_frozen_prefix_activations": true,
166
+ "phase_state_camera_index": 0,
167
+ "phase_state_frame_stride": 2,
168
+ "phase_state_vision_chunk_size": 32,
169
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
170
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
171
+ "phase_state_cache_episodes": 2,
172
+ "phase_state_max_steps": 1000,
173
+ "phase_state_checkpoint_strict": true,
174
+ "freeze_phase_state_encoder": true,
175
+ "phase_state_action_normalization": "QUANTILES",
176
+ "phase_state_action_dim": 14,
177
+ "phase_state_dim": 256,
178
+ "phase_state_layers": 4,
179
+ "phase_state_dropout": 0.1,
180
+ "phase_state_vision_pretrained": false,
181
+ "selector_heads": 8,
182
+ "evidence_dim": 256,
183
+ "use_relative_actions": false,
184
+ "relative_exclude_joints": [
185
+ "gripper"
186
+ ],
187
+ "action_feature_names": [
188
+ "motors"
189
+ ],
190
+ "optimizer_lr": 2.5e-05,
191
+ "evidence_optimizer_lr": 0.0001,
192
+ "action_optimizer_lr": 2.5e-05,
193
+ "vlm_optimizer_lr": 2.5e-05,
194
+ "vision_optimizer_lr": 2.5e-05,
195
+ "optimizer_betas": [
196
+ 0.9,
197
+ 0.95
198
+ ],
199
+ "optimizer_eps": 1e-08,
200
+ "optimizer_weight_decay": 0.01,
201
+ "optimizer_grad_clip_norm": 1.0,
202
+ "scheduler_warmup_steps": 33,
203
+ "scheduler_decay_steps": 1000,
204
+ "scheduler_decay_lr": 2.5e-06
205
+ },
206
+ "reward_model": null,
207
+ "output_dir": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
208
+ "job_name": "robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
209
+ "resume": false,
210
+ "seed": 42,
211
+ "cudnn_deterministic": false,
212
+ "num_workers": 4,
213
+ "batch_size": 16,
214
+ "fsdp": true,
215
+ "fsdp_save_training_state": false,
216
+ "prefetch_factor": 2,
217
+ "persistent_workers": true,
218
+ "steps": 1000,
219
+ "eval_freq": 0,
220
+ "log_freq": 100,
221
+ "tolerance_s": 0.0001,
222
+ "save_checkpoint": true,
223
+ "save_freq": 100,
224
+ "use_policy_training_preset": true,
225
+ "optimizer": {
226
+ "type": "adamw",
227
+ "lr": 2.5e-05,
228
+ "weight_decay": 0.01,
229
+ "grad_clip_norm": 1.0,
230
+ "betas": [
231
+ 0.9,
232
+ 0.95
233
+ ],
234
+ "eps": 1e-08
235
+ },
236
+ "scheduler": {
237
+ "type": "cosine_decay_with_warmup",
238
+ "num_warmup_steps": 33,
239
+ "num_decay_steps": 1000,
240
+ "peak_lr": 2.5e-05,
241
+ "decay_lr": 2.5e-06
242
+ },
243
+ "eval": {
244
+ "n_episodes": 50,
245
+ "batch_size": 50,
246
+ "use_async_envs": true,
247
+ "max_episodes_rendered": 10,
248
+ "reseed_policy_each_episode": false
249
+ },
250
+ "wandb": {
251
+ "enable": true,
252
+ "disable_artifact": true,
253
+ "project": "action-evidence-vla-pi05-phase",
254
+ "entity": null,
255
+ "notes": null,
256
+ "run_id": "i0fuukqo",
257
+ "mode": "online",
258
+ "add_tags": true
259
+ },
260
+ "peft": null,
261
+ "sample_weighting": null,
262
+ "rename_map": {},
263
+ "checkpoint_path": null
264
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-10-27T01:46:34.356935948+08:00","level":"INFO","msg":"stream: starting","core version":"0.24.2"}
2
+ {"time":"2026-10-27T01:46:35.077973577+08:00","level":"INFO","msg":"stream: created new stream","id":"i0fuukqo"}
3
+ {"time":"2026-10-27T01:46:35.07810336+08:00","level":"INFO","msg":"handler: started","stream_id":"i0fuukqo"}
4
+ {"time":"2026-10-27T01:46:35.078299082+08:00","level":"INFO","msg":"stream: started","id":"i0fuukqo"}
5
+ {"time":"2026-10-27T01:46:35.078327555+08:00","level":"INFO","msg":"writer: started","stream_id":"i0fuukqo"}
6
+ {"time":"2026-10-27T01:46:35.07833363+08:00","level":"INFO","msg":"sender: started","stream_id":"i0fuukqo"}
7
+ {"time":"2026-10-27T02:51:32.543676531+08:00","level":"INFO","msg":"stream: closing","id":"i0fuukqo"}
8
+ {"time":"2026-10-27T02:51:34.333141258+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
9
+ {"time":"2026-10-27T02:51:34.704560495+08:00","level":"INFO","msg":"handler: closed","stream_id":"i0fuukqo"}
10
+ {"time":"2026-10-27T02:51:34.704756589+08:00","level":"INFO","msg":"sender: closed","stream_id":"i0fuukqo"}
11
+ {"time":"2026-10-27T02:51:34.70479882+08:00","level":"INFO","msg":"stream: closed","id":"i0fuukqo"}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/debug.log ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Current SDK version is 0.24.2
2
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Configure stats pid to 103747
3
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():717] Logging user logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug.log
5
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-internal.log
6
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():844] calling init triggers
7
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
8
+ config: {'dataset': {'repo_id': 'local/robotwin_adjust_bottle_aevla50', 'root': '/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50', 'episodes': None, 'image_transforms': {'enable': False, 'max_num_transforms': 3, 'random_order': False, 'tfs': {'brightness': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'brightness': [0.8, 1.2]}}, 'contrast': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'contrast': [0.8, 1.2]}}, 'saturation': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'saturation': [0.5, 1.5]}}, 'hue': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'hue': [-0.05, 0.05]}}, 'sharpness': {'weight': 1.0, 'type': 'SharpnessJitter', 'kwargs': {'sharpness': [0.5, 1.5]}}, 'affine': {'weight': 1.0, 'type': 'RandomAffine', 'kwargs': {'degrees': [-5.0, 5.0], 'translate': [0.05, 0.05]}}}}, 'revision': 'v3.0', 'use_imagenet_stats': True, 'video_backend': 'pyav', 'return_uint8': False, 'streaming': False}, 'env': None, 'policy': {'type': 'action_evidence_vla', 'n_obs_steps': 1, 'input_features': {}, 'output_features': {}, 'device': 'cuda', 'use_amp': False, 'use_peft': False, 'push_to_hub': False, 'repo_id': None, 'private': None, 'tags': None, 'license': None, 'pretrained_path': None, 'pi05_init_path': 'lerobot/pi05_base', 'paligemma_variant': 'gemma_2b', 'action_expert_variant': 'gemma_300m', 'pi05_precision': 'bfloat16', 'chunk_size': 50, 'n_action_steps': 10, 'max_state_dim': 32, 'max_action_dim': 32, 'tokenizer_max_length': 200, 'image_resolution': [224, 224], 'empty_cameras': 0, 'normalization_mapping': {'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>, 'STATE': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'ACTION': <NormalizationMode.QUANTILES: 'QUANTILES'>}, 'num_inference_steps': 10, 'time_sampling_beta_alpha': 1.5, 'time_sampling_beta_beta': 1.0, 'time_sampling_scale': 0.999, 'time_sampling_offset': 0.001, 'min_period': 0.004, 'max_period': 4.0, 'freeze_vlm': False, 'freeze_vision_encoder': False, 'gradient_checkpointing': True, 'offload_frozen_prefix_activations': True, 'phase_state_camera_index': 0, 'phase_state_frame_stride': 2, 'phase_state_vision_chunk_size': 32, 'phase_state_checkpoint_path': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth', 'phase_state_cache_root': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14', 'phase_state_cache_episodes': 2, 'phase_state_max_steps': 1000, 'phase_state_checkpoint_strict': True, 'freeze_phase_state_encoder': True, 'phase_state_action_normalization': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'phase_state_action_dim': 14, 'phase_state_dim': 256, 'phase_state_layers': 4, 'phase_state_dropout': 0.1, 'phase_state_vision_pretrained': False, 'selector_heads': 8, 'evidence_dim': 256, 'use_relative_actions': False, 'relative_exclude_joints': ['gripper'], 'action_feature_names': None, 'optimizer_lr': 2.5e-05, 'evidence_optimizer_lr': 0.0001, 'action_optimizer_lr': 2.5e-05, 'vlm_optimizer_lr': 2.5e-05, 'vision_optimizer_lr': 2.5e-05, 'optimizer_betas': [0.9, 0.95], 'optimizer_eps': 1e-08, 'optimizer_weight_decay': 0.01, 'optimizer_grad_clip_norm': 1.0, 'scheduler_warmup_steps': 33, 'scheduler_decay_steps': 1000, 'scheduler_decay_lr': 2.5e-06}, 'reward_model': None, 'output_dir': '/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'job_name': 'robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'resume': False, 'seed': 42, 'cudnn_deterministic': False, 'num_workers': 4, 'batch_size': 16, 'fsdp': True, 'fsdp_save_training_state': False, 'prefetch_factor': 2, 'persistent_workers': True, 'steps': 1000, 'eval_freq': 0, 'log_freq': 100, 'tolerance_s': 0.0001, 'save_checkpoint': True, 'save_freq': 100, 'use_policy_training_preset': True, 'optimizer': {'type': 'adamw', 'lr': 2.5e-05, 'weight_decay': 0.01, 'grad_clip_norm': 1.0, 'betas': [0.9, 0.95], 'eps': 1e-08}, 'scheduler': {'type': 'cosine_decay_with_warmup', 'num_warmup_steps': 33, 'num_decay_steps': 1000, 'peak_lr': 2.5e-05, 'decay_lr': 2.5e-06}, 'eval': {'n_episodes': 50, 'batch_size': 50, 'use_async_envs': True, 'max_episodes_rendered': 10, 'reseed_policy_each_episode': False}, 'wandb': {'enable': True, 'disable_artifact': True, 'project': 'action-evidence-vla-pi05-phase', 'entity': None, 'notes': None, 'run_id': None, 'mode': 'online', 'add_tags': True}, 'peft': None, 'sample_weighting': None, 'rename_map': {}, 'checkpoint_path': None, '_wandb': {}}
9
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():892] starting backend
10
+ 2026-10-27 01:46:34,348 INFO MainThread:103747 [wandb_init.py:init():895] sending inform_init request
11
+ 2026-10-27 01:46:34,354 INFO MainThread:103747 [wandb_init.py:init():903] backend started and connected
12
+ 2026-10-27 01:46:34,355 INFO MainThread:103747 [wandb_init.py:init():973] updated telemetry
13
+ 2026-10-27 01:46:34,361 INFO MainThread:103747 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
14
+ 2026-10-27 01:46:35,830 INFO MainThread:103747 [wandb_init.py:init():1042] starting run threads in backend
15
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_console_start():2529] atexit reg
16
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2377] redirect: wrap_raw
17
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2446] Wrapping output streams.
18
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2469] Redirects installed.
19
+ 2026-10-27 01:46:36,002 INFO MainThread:103747 [wandb_init.py:init():1082] run started, returning control to user process
20
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [service_client.py:_forward_responses():94] Reached EOF.
21
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [mailbox.py:close():154] Closing mailbox, abandoning 1 handles.
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/config.yaml ADDED
@@ -0,0 +1,351 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.24.2
4
+ e:
5
+ ck9llrkydvf1se7fpd5kty8afbhldpo1:
6
+ args:
7
+ - --fsdp=true
8
+ - --fsdp_save_training_state=false
9
+ - --policy.discover_packages_path=lerobot.policies.action_evidence_vla
10
+ - --policy.type=action_evidence_vla
11
+ - --policy.device=cuda
12
+ - --policy.pi05_precision=bfloat16
13
+ - --policy.normalization_mapping={"VISUAL":"IDENTITY","STATE":"QUANTILES","ACTION":"QUANTILES"}
14
+ - --policy.phase_state_action_normalization=QUANTILES
15
+ - --policy.pi05_init_path=lerobot/pi05_base
16
+ - --policy.phase_state_checkpoint_path=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth
17
+ - --policy.phase_state_cache_root=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
18
+ - --policy.phase_state_checkpoint_strict=true
19
+ - --policy.freeze_phase_state_encoder=true
20
+ - --policy.freeze_vlm=false
21
+ - --policy.freeze_vision_encoder=false
22
+ - --policy.phase_state_camera_index=0
23
+ - --policy.phase_state_frame_stride=2
24
+ - --policy.phase_state_vision_chunk_size=32
25
+ - --policy.n_obs_steps=1
26
+ - --policy.chunk_size=50
27
+ - --policy.n_action_steps=10
28
+ - --policy.max_state_dim=32
29
+ - --policy.max_action_dim=32
30
+ - --policy.phase_state_action_dim=14
31
+ - --policy.empty_cameras=0
32
+ - --policy.tokenizer_max_length=200
33
+ - --policy.gradient_checkpointing=true
34
+ - --policy.evidence_optimizer_lr=0.0001
35
+ - --policy.action_optimizer_lr=0.000025
36
+ - --policy.vlm_optimizer_lr=0.000025
37
+ - --policy.vision_optimizer_lr=0.000025
38
+ - --policy.optimizer_lr=0.000025
39
+ - --policy.optimizer_weight_decay=0.01
40
+ - --policy.optimizer_grad_clip_norm=1.0
41
+ - --policy.scheduler_warmup_steps=33
42
+ - --policy.scheduler_decay_steps=1000
43
+ - --policy.scheduler_decay_lr=0.0000025
44
+ - --policy.push_to_hub=false
45
+ - --dataset.repo_id=local/robotwin_adjust_bottle_aevla50
46
+ - --dataset.root=/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50
47
+ - --dataset.revision=v3.0
48
+ - --dataset.video_backend=pyav
49
+ - --output_dir=/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
50
+ - --job_name=robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
51
+ - --steps=1000
52
+ - --batch_size=16
53
+ - --num_workers=4
54
+ - --persistent_workers=true
55
+ - --prefetch_factor=2
56
+ - --log_freq=100
57
+ - --save_freq=100
58
+ - --eval_freq=0
59
+ - --seed=42
60
+ - --wandb.enable=true
61
+ - --wandb.mode=online
62
+ - --wandb.project=action-evidence-vla-pi05-phase
63
+ - --wandb.disable_artifact=true
64
+ cpu_count: 72
65
+ cpu_count_logical: 144
66
+ cudaVersion: "13.0"
67
+ disk:
68
+ /:
69
+ total: "32212254720"
70
+ used: "9985990656"
71
+ executable: /root/autodl-tmp/projections/SGVLA/.venv/bin/python3
72
+ git:
73
+ commit: ca8da87e9e816c71691333960e7bbc3da699832e
74
+ remote: git@github.com:LliuTtao/SGVLA.git
75
+ gpu: NVIDIA A800-SXM4-80GB
76
+ gpu_count: 4
77
+ gpu_nvidia:
78
+ - architecture: Ampere
79
+ cudaCores: 6912
80
+ memoryTotal: "85899345920"
81
+ name: NVIDIA A800-SXM4-80GB
82
+ uuid: GPU-56601ec2-dc48-6741-6bdd-91c49df6f015
83
+ - architecture: Ampere
84
+ cudaCores: 6912
85
+ memoryTotal: "85899345920"
86
+ name: NVIDIA A800-SXM4-80GB
87
+ uuid: GPU-dfe6761a-23f4-c34c-9c03-01da02020e5f
88
+ - architecture: Ampere
89
+ cudaCores: 6912
90
+ memoryTotal: "85899345920"
91
+ name: NVIDIA A800-SXM4-80GB
92
+ uuid: GPU-8c94bc98-aa6a-d757-b3d0-b74b6f286705
93
+ - architecture: Ampere
94
+ cudaCores: 6912
95
+ memoryTotal: "85899345920"
96
+ name: NVIDIA A800-SXM4-80GB
97
+ uuid: GPU-24625a85-1603-30c0-36d9-c3a207715c1f
98
+ host: autodl-container-sq1jeg5hau-f8a4dbe8
99
+ memory:
100
+ total: "1081807933440"
101
+ os: Linux-5.15.0-78-generic-x86_64-with-glibc2.35
102
+ program: -m lerobot.scripts.lerobot_train
103
+ python: CPython 3.12.3
104
+ root: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
105
+ startedAt: "2026-10-26T17:46:34.131322Z"
106
+ writerId: ck9llrkydvf1se7fpd5kty8afbhldpo1
107
+ m: []
108
+ python_version: 3.12.3
109
+ t:
110
+ "1":
111
+ - 1
112
+ - 11
113
+ - 41
114
+ - 49
115
+ - 51
116
+ - 71
117
+ "2":
118
+ - 1
119
+ - 11
120
+ - 41
121
+ - 49
122
+ - 51
123
+ - 71
124
+ "3":
125
+ - 13
126
+ - 15
127
+ - 16
128
+ - 61
129
+ "4": 3.12.3
130
+ "5": 0.24.2
131
+ "6": 5.5.4
132
+ "10":
133
+ - 21
134
+ "12": 0.24.2
135
+ "13": linux-x86_64
136
+ batch_size:
137
+ value: 16
138
+ checkpoint_path:
139
+ value: null
140
+ cudnn_deterministic:
141
+ value: false
142
+ dataset:
143
+ value:
144
+ episodes: null
145
+ image_transforms:
146
+ enable: false
147
+ max_num_transforms: 3
148
+ random_order: false
149
+ tfs:
150
+ affine:
151
+ kwargs:
152
+ degrees:
153
+ - -5
154
+ - 5
155
+ translate:
156
+ - 0.05
157
+ - 0.05
158
+ type: RandomAffine
159
+ weight: 1
160
+ brightness:
161
+ kwargs:
162
+ brightness:
163
+ - 0.8
164
+ - 1.2
165
+ type: ColorJitter
166
+ weight: 1
167
+ contrast:
168
+ kwargs:
169
+ contrast:
170
+ - 0.8
171
+ - 1.2
172
+ type: ColorJitter
173
+ weight: 1
174
+ hue:
175
+ kwargs:
176
+ hue:
177
+ - -0.05
178
+ - 0.05
179
+ type: ColorJitter
180
+ weight: 1
181
+ saturation:
182
+ kwargs:
183
+ saturation:
184
+ - 0.5
185
+ - 1.5
186
+ type: ColorJitter
187
+ weight: 1
188
+ sharpness:
189
+ kwargs:
190
+ sharpness:
191
+ - 0.5
192
+ - 1.5
193
+ type: SharpnessJitter
194
+ weight: 1
195
+ repo_id: local/robotwin_adjust_bottle_aevla50
196
+ return_uint8: false
197
+ revision: v3.0
198
+ root: /root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50
199
+ streaming: false
200
+ use_imagenet_stats: true
201
+ video_backend: pyav
202
+ env:
203
+ value: null
204
+ eval:
205
+ value:
206
+ batch_size: 50
207
+ max_episodes_rendered: 10
208
+ n_episodes: 50
209
+ reseed_policy_each_episode: false
210
+ use_async_envs: true
211
+ eval_freq:
212
+ value: 0
213
+ fsdp:
214
+ value: true
215
+ fsdp_save_training_state:
216
+ value: false
217
+ job_name:
218
+ value: robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
219
+ log_freq:
220
+ value: 100
221
+ num_workers:
222
+ value: 4
223
+ optimizer:
224
+ value:
225
+ betas:
226
+ - 0.9
227
+ - 0.95
228
+ eps: 1e-08
229
+ grad_clip_norm: 1
230
+ lr: 2.5e-05
231
+ type: adamw
232
+ weight_decay: 0.01
233
+ output_dir:
234
+ value: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
235
+ peft:
236
+ value: null
237
+ persistent_workers:
238
+ value: true
239
+ policy:
240
+ value:
241
+ action_expert_variant: gemma_300m
242
+ action_feature_names: null
243
+ action_optimizer_lr: 2.5e-05
244
+ chunk_size: 50
245
+ device: cuda
246
+ empty_cameras: 0
247
+ evidence_dim: 256
248
+ evidence_optimizer_lr: 0.0001
249
+ freeze_phase_state_encoder: true
250
+ freeze_vision_encoder: false
251
+ freeze_vlm: false
252
+ gradient_checkpointing: true
253
+ image_resolution:
254
+ - 224
255
+ - 224
256
+ license: null
257
+ max_action_dim: 32
258
+ max_period: 4
259
+ max_state_dim: 32
260
+ min_period: 0.004
261
+ n_action_steps: 10
262
+ n_obs_steps: 1
263
+ normalization_mapping:
264
+ ACTION: QUANTILES
265
+ STATE: QUANTILES
266
+ VISUAL: IDENTITY
267
+ num_inference_steps: 10
268
+ offload_frozen_prefix_activations: true
269
+ optimizer_betas:
270
+ - 0.9
271
+ - 0.95
272
+ optimizer_eps: 1e-08
273
+ optimizer_grad_clip_norm: 1
274
+ optimizer_lr: 2.5e-05
275
+ optimizer_weight_decay: 0.01
276
+ paligemma_variant: gemma_2b
277
+ phase_state_action_dim: 14
278
+ phase_state_action_normalization: QUANTILES
279
+ phase_state_cache_episodes: 2
280
+ phase_state_cache_root: /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
281
+ phase_state_camera_index: 0
282
+ phase_state_checkpoint_path: /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth
283
+ phase_state_checkpoint_strict: true
284
+ phase_state_dim: 256
285
+ phase_state_dropout: 0.1
286
+ phase_state_frame_stride: 2
287
+ phase_state_layers: 4
288
+ phase_state_max_steps: 1000
289
+ phase_state_vision_chunk_size: 32
290
+ phase_state_vision_pretrained: false
291
+ pi05_init_path: lerobot/pi05_base
292
+ pi05_precision: bfloat16
293
+ pretrained_path: null
294
+ private: null
295
+ push_to_hub: false
296
+ relative_exclude_joints:
297
+ - gripper
298
+ repo_id: null
299
+ scheduler_decay_lr: 2.5e-06
300
+ scheduler_decay_steps: 1000
301
+ scheduler_warmup_steps: 33
302
+ selector_heads: 8
303
+ tags: null
304
+ time_sampling_beta_alpha: 1.5
305
+ time_sampling_beta_beta: 1
306
+ time_sampling_offset: 0.001
307
+ time_sampling_scale: 0.999
308
+ tokenizer_max_length: 200
309
+ type: action_evidence_vla
310
+ use_amp: false
311
+ use_peft: false
312
+ use_relative_actions: false
313
+ vision_optimizer_lr: 2.5e-05
314
+ vlm_optimizer_lr: 2.5e-05
315
+ prefetch_factor:
316
+ value: 2
317
+ resume:
318
+ value: false
319
+ reward_model:
320
+ value: null
321
+ sample_weighting:
322
+ value: null
323
+ save_checkpoint:
324
+ value: true
325
+ save_freq:
326
+ value: 100
327
+ scheduler:
328
+ value:
329
+ decay_lr: 2.5e-06
330
+ num_decay_steps: 1000
331
+ num_warmup_steps: 33
332
+ peak_lr: 2.5e-05
333
+ type: cosine_decay_with_warmup
334
+ seed:
335
+ value: 42
336
+ steps:
337
+ value: 1000
338
+ tolerance_s:
339
+ value: 0.0001
340
+ use_policy_training_preset:
341
+ value: true
342
+ wandb:
343
+ value:
344
+ add_tags: true
345
+ disable_artifact: true
346
+ enable: true
347
+ entity: null
348
+ mode: online
349
+ notes: null
350
+ project: action-evidence-vla-pi05-phase
351
+ run_id: null
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/output.log ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ INFO 2026-10-27 01:46:36 db_utils.py:121 Logs will be synced with wandb.
2
+ INFO 2026-10-27 01:46:36 db_utils.py:122 Track this run --> https://wandb.ai/3cteam/action-evidence-vla-pi05-phase/runs/i0fuukqo
3
+ INFO 2026-10-27 01:46:36 ot_train.py:338 Creating dataset
4
+ INFO 2026-10-27 01:46:36 ot_train.py:372 Creating policy
5
+ INFO 2026-10-27 01:50:26 ence_vla.py:584 Initialized Action-Evidence VLA from PI0.5 checkpoint lerobot/pi05_base (bare model keys)
6
+ INFO 2026-10-27 01:50:27 ence_vla.py:613 Validated offline causal phase states at /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
7
+ INFO 2026-10-27 01:50:36 ot_train.py:449 Creating optimizer and scheduler
8
+ INFO 2026-10-27 01:50:37 ot_train.py:489 Output dir: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
9
+ INFO 2026-10-27 01:50:37 ot_train.py:496 cfg.steps=1000 (1K)
10
+ INFO 2026-10-27 01:50:37 ot_train.py:497 dataset.num_frames=7188 (7K)
11
+ INFO 2026-10-27 01:50:37 ot_train.py:498 dataset.num_episodes=50
12
+ INFO 2026-10-27 01:50:37 ot_train.py:501 Effective batch size: 16 x 4 = 64 samples
13
+ INFO 2026-10-27 01:50:37 ot_train.py:504 num_learnable_params=3385530640 (3B)
14
+ INFO 2026-10-27 01:50:37 ot_train.py:505 num_total_params=4175503632 (4B)
15
+ Training: 0%| | 0/1000 [00:00<?, ?step/s]INFO 2026-10-27 01:50:38 ot_train.py:582 Start offline training on a fixed dataset, with effective batch size: 64 samples
16
+ Training: 10%|█ | 100/1000 [05:55<53:07, 3.54s/step]INFO 2026-10-27 01:56:34 ot_train.py:618 step:100 smpl:6K ep:45 epch:0.89 loss:0.091 grdn:0.985 lr:8.4e-05 updt_s:3.495 data_s:0.056
17
+ INFO 2026-10-27 01:56:34 ot_train.py:634 Checkpoint FSDP2 policy after step 100
18
+ INFO 2026-10-27 01:56:48 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000100/pretrained_model/pytorch_model.bin
19
+ Training: 20%|██ | 200/1000 [12:00<46:03, 3.45s/step]INFO 2026-10-27 02:02:38 ot_train.py:618 step:200 smpl:13K ep:89 epch:1.78 loss:0.016 grdn:0.259 lr:9.5e-05 updt_s:3.450 data_s:0.047
20
+ INFO 2026-10-27 02:02:38 ot_train.py:634 Checkpoint FSDP2 policy after step 200
21
+ INFO 2026-10-27 02:02:50 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000200/pretrained_model/pytorch_model.bin
22
+ Training: 30%|███ | 300/1000 [18:04<40:11, 3.45s/step]INFO 2026-10-27 02:08:43 ot_train.py:618 step:300 smpl:19K ep:134 epch:2.67 loss:0.011 grdn:0.210 lr:8.7e-05 updt_s:3.465 data_s:0.046
23
+ INFO 2026-10-27 02:08:43 ot_train.py:634 Checkpoint FSDP2 policy after step 300
24
+ INFO 2026-10-27 02:08:54 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000300/pretrained_model/pytorch_model.bin
25
+ Training: 40%|████ | 400/1000 [24:09<40:57, 4.10s/step]INFO 2026-10-27 02:14:47 ot_train.py:618 step:400 smpl:26K ep:178 epch:3.56 loss:0.009 grdn:0.190 lr:7.5e-05 updt_s:3.467 data_s:0.049
26
+ INFO 2026-10-27 02:14:47 ot_train.py:634 Checkpoint FSDP2 policy after step 400
27
+ INFO 2026-10-27 02:14:59 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000400/pretrained_model/pytorch_model.bin
28
+ Training: 50%|█████ | 500/1000 [30:11<29:04, 3.49s/step]INFO 2026-10-27 02:20:50 ot_train.py:618 step:500 smpl:32K ep:223 epch:4.45 loss:0.007 grdn:0.170 lr:6.2e-05 updt_s:3.467 data_s:0.028
29
+ INFO 2026-10-27 02:20:50 ot_train.py:634 Checkpoint FSDP2 policy after step 500
30
+ INFO 2026-10-27 02:21:02 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000500/pretrained_model/pytorch_model.bin
31
+ Training: 60%|██████ | 600/1000 [36:17<23:33, 3.53s/step]INFO 2026-10-27 02:26:56 ot_train.py:618 step:600 smpl:38K ep:267 epch:5.34 loss:0.006 grdn:0.148 lr:4.8e-05 updt_s:3.475 data_s:0.048
32
+ INFO 2026-10-27 02:26:56 ot_train.py:634 Checkpoint FSDP2 policy after step 600
33
+ INFO 2026-10-27 02:27:08 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000600/pretrained_model/pytorch_model.bin
34
+ Training: 70%|███████ | 700/1000 [42:23<17:21, 3.47s/step]INFO 2026-10-27 02:33:02 ot_train.py:618 step:700 smpl:45K ep:312 epch:6.23 loss:0.005 grdn:0.149 lr:3.5e-05 updt_s:3.476 data_s:0.047
35
+ INFO 2026-10-27 02:33:02 ot_train.py:634 Checkpoint FSDP2 policy after step 700
36
+ INFO 2026-10-27 02:33:14 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000700/pretrained_model/pytorch_model.bin
37
+ Training: 80%|████████ | 800/1000 [48:29<13:13, 3.97s/step]INFO 2026-10-27 02:39:07 ot_train.py:618 step:800 smpl:51K ep:356 epch:7.12 loss:0.004 grdn:0.128 lr:2.3e-05 updt_s:3.473 data_s:0.049
38
+ INFO 2026-10-27 02:39:07 ot_train.py:634 Checkpoint FSDP2 policy after step 800
39
+ INFO 2026-10-27 02:39:19 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000800/pretrained_model/pytorch_model.bin
40
+ Training: 90%|█████████ | 900/1000 [54:33<05:49, 3.49s/step]INFO 2026-10-27 02:45:11 ot_train.py:618 step:900 smpl:58K ep:401 epch:8.01 loss:0.004 grdn:0.122 lr:1.5e-05 updt_s:3.481 data_s:0.029
41
+ INFO 2026-10-27 02:45:11 ot_train.py:634 Checkpoint FSDP2 policy after step 900
42
+ INFO 2026-10-27 02:45:24 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/pytorch_model.bin
43
+ Training: 100%|██████████| 1000/1000 [1:00:40<00:00, 3.49s/step]INFO 2026-10-27 02:51:19 ot_train.py:618 step:1K smpl:64K ep:445 epch:8.90 loss:0.004 grdn:0.122 lr:1.1e-05 updt_s:3.491 data_s:0.048
44
+ INFO 2026-10-27 02:51:19 ot_train.py:634 Checkpoint FSDP2 policy after step 1000
45
+ INFO 2026-10-27 02:51:31 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/pytorch_model.bin
46
+ Training: 100%|██████████| 1000/1000 [1:00:52<00:00, 3.65s/step]
47
+ INFO 2026-10-27 02:51:31 ot_train.py:723 End of training
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/requirements.txt ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ lerobot==0.5.2
2
+ Wand==0.6.13
3
+ absl-py==2.5.0
4
+ accelerate==1.15.0
5
+ aiohappyeyeballs==2.7.1
6
+ aiohttp==3.14.3
7
+ aiosignal==1.4.0
8
+ annotated-doc==0.0.5
9
+ annotated-types==0.8.0
10
+ antlr4-python3-runtime==4.9.3
11
+ anyio==4.15.1
12
+ attrs==26.1.0
13
+ av==15.1.0
14
+ bddl==1.0.1
15
+ certifi==2026.7.22
16
+ charset-normalizer==3.5.1
17
+ click==8.5.0
18
+ cloudpickle==3.1.2
19
+ cmake==3.31.10
20
+ contourpy==1.3.3
21
+ cuda-bindings==12.9.7
22
+ cuda-pathfinder==1.8.1
23
+ cuda-toolkit==12.8.1
24
+ cycler==0.12.1
25
+ datasets==4.8.5
26
+ dill==0.4.1
27
+ draccus==0.10.0
28
+ easydict==1.13
29
+ egl_probe==1.0.2
30
+ einops==0.8.2
31
+ etils==1.14.0
32
+ evdev==2.0.0
33
+ Farama-Notifications==0.0.6
34
+ fastjsonschema==2.22.2
35
+ filelock==3.32.6
36
+ fonttools==4.64.0
37
+ frozenlist==1.8.0
38
+ fsspec==2026.2.0
39
+ future==1.0.0
40
+ gitdb==4.0.12
41
+ GitPython==3.1.62
42
+ glfw==2.10.2
43
+ grpcio==1.73.1
44
+ gym==0.26.2
45
+ gym-notices==0.1.0
46
+ gymnasium==1.3.0
47
+ h11==0.16.0
48
+ h5py==3.16.0
49
+ hf_egl_probe==1.0.2
50
+ hf-xet==1.6.0
51
+ httpcore==1.0.9
52
+ httpx==0.28.1
53
+ huggingface_hub==1.30.0
54
+ hydra-core==1.3.6
55
+ idna==3.19
56
+ ImageIO==2.37.4
57
+ imageio-ffmpeg==0.6.0
58
+ iniconfig==2.3.0
59
+ Jinja2==3.1.6
60
+ jsonlines==4.0.0
61
+ jsonschema==4.26.0
62
+ jsonschema-specifications==2025.9.1
63
+ jupyter_core==5.9.1
64
+ jupytext==1.19.5
65
+ kiwisolver==1.5.1
66
+ lazy-loader==0.6
67
+ lerobot==0.5.2
68
+ libero==0.1.0
69
+ llvmlite==0.49.0
70
+ Markdown==3.10.3
71
+ markdown-it-py==4.2.0
72
+ MarkupSafe==3.0.3
73
+ matplotlib==3.11.1
74
+ mdit-py-plugins==0.6.1
75
+ mdurl==0.1.2
76
+ mergedeep==1.3.4
77
+ mpmath==1.3.0
78
+ mujoco==3.7.0
79
+ multidict==6.8.0
80
+ multiprocess==0.70.19
81
+ mypy_extensions==1.1.0
82
+ nbformat==5.11.1
83
+ networkx==3.6.1
84
+ numba==0.67.0
85
+ numpy==2.2.6
86
+ nvidia-cublas-cu12==12.8.4.1
87
+ nvidia-cuda-cupti-cu12==12.8.90
88
+ nvidia-cuda-nvrtc-cu12==12.8.93
89
+ nvidia-cuda-runtime-cu12==12.8.90
90
+ nvidia-cudnn-cu12==9.19.0.56
91
+ nvidia-cufft-cu12==11.3.3.83
92
+ nvidia-cufile-cu12==1.13.1.3
93
+ nvidia-curand-cu12==10.3.9.90
94
+ nvidia-cusolver-cu12==11.7.3.90
95
+ nvidia-cusparse-cu12==12.5.8.93
96
+ nvidia-cusparselt-cu12==0.7.1
97
+ nvidia-nccl-cu12==2.28.9
98
+ nvidia-nvjitlink-cu12==12.8.93
99
+ nvidia-nvshmem-cu12==3.4.5
100
+ nvidia-nvtx-cu12==12.8.90
101
+ omegaconf==2.3.1
102
+ opencv-python==4.14.0.94
103
+ opencv-python-headless==4.13.0.92
104
+ packaging==25.0
105
+ pandas==2.3.3
106
+ pillow==12.3.0
107
+ platformdirs==4.11.8
108
+ pluggy==1.6.0
109
+ propcache==0.5.2
110
+ protobuf==6.31.1
111
+ psutil==7.2.2
112
+ pyarrow==25.0.1
113
+ pydantic==2.13.5
114
+ pydantic_core==2.46.5
115
+ Pygments==2.21.0
116
+ pynput==1.8.2
117
+ PyOpenGL==3.1.10
118
+ pyparsing==3.3.2
119
+ pytest==8.4.2
120
+ python-dateutil==2.9.0.post0
121
+ python-xlib==0.33
122
+ pytz==2026.3.post1
123
+ PyYAML==6.0.3
124
+ pyyaml-include==1.4.1
125
+ referencing==0.37.0
126
+ regex==2026.9.3
127
+ requests==2.34.2
128
+ rich==15.0.0
129
+ robomimic==0.2.0
130
+ robosuite==1.4.1
131
+ rpds-py==2026.6.3
132
+ safetensors==0.8.0
133
+ scikit-image==0.25.2
134
+ scipy==1.18.1
135
+ sentry-sdk==2.69.1
136
+ setuptools==80.10.2
137
+ shellingham==1.5.4
138
+ six==1.17.0
139
+ smmap==5.0.3
140
+ sympy==1.14.0
141
+ tensorboard==2.20.0
142
+ tensorboard-data-server==0.7.2
143
+ tensorboardX==2.6.5
144
+ termcolor==3.3.0
145
+ thop==0.1.1-2209072238
146
+ tifffile==2026.9.20
147
+ timm==1.0.29
148
+ tokenizers==0.22.2
149
+ toml==0.10.2
150
+ torch==2.11.0+cu128
151
+ torchcodec==0.11.1
152
+ torchvision==0.26.0+cu128
153
+ tqdm==4.70.0
154
+ traitlets==5.16.1
155
+ transformers==5.5.4
156
+ triton==3.6.0
157
+ typer==0.27.2
158
+ typing_extensions==4.16.0
159
+ typing-inspect==0.9.0
160
+ typing-inspection==0.4.4
161
+ tzdata==2026.3
162
+ urllib3==2.7.0
163
+ wandb==0.24.2
164
+ Werkzeug==3.1.8
165
+ xxhash==4.0.1
166
+ yarl==1.24.5
167
+ zipp==4.1.0
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/wandb-metadata.json ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-5.15.0-78-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.12.3",
4
+ "startedAt": "2026-10-26T17:46:34.131322Z",
5
+ "args": [
6
+ "--fsdp=true",
7
+ "--fsdp_save_training_state=false",
8
+ "--policy.discover_packages_path=lerobot.policies.action_evidence_vla",
9
+ "--policy.type=action_evidence_vla",
10
+ "--policy.device=cuda",
11
+ "--policy.pi05_precision=bfloat16",
12
+ "--policy.normalization_mapping={\"VISUAL\":\"IDENTITY\",\"STATE\":\"QUANTILES\",\"ACTION\":\"QUANTILES\"}",
13
+ "--policy.phase_state_action_normalization=QUANTILES",
14
+ "--policy.pi05_init_path=lerobot/pi05_base",
15
+ "--policy.phase_state_checkpoint_path=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
16
+ "--policy.phase_state_cache_root=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
17
+ "--policy.phase_state_checkpoint_strict=true",
18
+ "--policy.freeze_phase_state_encoder=true",
19
+ "--policy.freeze_vlm=false",
20
+ "--policy.freeze_vision_encoder=false",
21
+ "--policy.phase_state_camera_index=0",
22
+ "--policy.phase_state_frame_stride=2",
23
+ "--policy.phase_state_vision_chunk_size=32",
24
+ "--policy.n_obs_steps=1",
25
+ "--policy.chunk_size=50",
26
+ "--policy.n_action_steps=10",
27
+ "--policy.max_state_dim=32",
28
+ "--policy.max_action_dim=32",
29
+ "--policy.phase_state_action_dim=14",
30
+ "--policy.empty_cameras=0",
31
+ "--policy.tokenizer_max_length=200",
32
+ "--policy.gradient_checkpointing=true",
33
+ "--policy.evidence_optimizer_lr=0.0001",
34
+ "--policy.action_optimizer_lr=0.000025",
35
+ "--policy.vlm_optimizer_lr=0.000025",
36
+ "--policy.vision_optimizer_lr=0.000025",
37
+ "--policy.optimizer_lr=0.000025",
38
+ "--policy.optimizer_weight_decay=0.01",
39
+ "--policy.optimizer_grad_clip_norm=1.0",
40
+ "--policy.scheduler_warmup_steps=33",
41
+ "--policy.scheduler_decay_steps=1000",
42
+ "--policy.scheduler_decay_lr=0.0000025",
43
+ "--policy.push_to_hub=false",
44
+ "--dataset.repo_id=local/robotwin_adjust_bottle_aevla50",
45
+ "--dataset.root=/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50",
46
+ "--dataset.revision=v3.0",
47
+ "--dataset.video_backend=pyav",
48
+ "--output_dir=/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
49
+ "--job_name=robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
50
+ "--steps=1000",
51
+ "--batch_size=16",
52
+ "--num_workers=4",
53
+ "--persistent_workers=true",
54
+ "--prefetch_factor=2",
55
+ "--log_freq=100",
56
+ "--save_freq=100",
57
+ "--eval_freq=0",
58
+ "--seed=42",
59
+ "--wandb.enable=true",
60
+ "--wandb.mode=online",
61
+ "--wandb.project=action-evidence-vla-pi05-phase",
62
+ "--wandb.disable_artifact=true"
63
+ ],
64
+ "program": "-m lerobot.scripts.lerobot_train",
65
+ "git": {
66
+ "remote": "git@github.com:LliuTtao/SGVLA.git",
67
+ "commit": "ca8da87e9e816c71691333960e7bbc3da699832e"
68
+ },
69
+ "root": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
70
+ "host": "autodl-container-sq1jeg5hau-f8a4dbe8",
71
+ "executable": "/root/autodl-tmp/projections/SGVLA/.venv/bin/python3",
72
+ "cpu_count": 72,
73
+ "cpu_count_logical": 144,
74
+ "gpu": "NVIDIA A800-SXM4-80GB",
75
+ "gpu_count": 4,
76
+ "disk": {
77
+ "/": {
78
+ "total": "32212254720",
79
+ "used": "9985990656"
80
+ }
81
+ },
82
+ "memory": {
83
+ "total": "1081807933440"
84
+ },
85
+ "gpu_nvidia": [
86
+ {
87
+ "name": "NVIDIA A800-SXM4-80GB",
88
+ "memoryTotal": "85899345920",
89
+ "cudaCores": 6912,
90
+ "architecture": "Ampere",
91
+ "uuid": "GPU-56601ec2-dc48-6741-6bdd-91c49df6f015"
92
+ },
93
+ {
94
+ "name": "NVIDIA A800-SXM4-80GB",
95
+ "memoryTotal": "85899345920",
96
+ "cudaCores": 6912,
97
+ "architecture": "Ampere",
98
+ "uuid": "GPU-dfe6761a-23f4-c34c-9c03-01da02020e5f"
99
+ },
100
+ {
101
+ "name": "NVIDIA A800-SXM4-80GB",
102
+ "memoryTotal": "85899345920",
103
+ "cudaCores": 6912,
104
+ "architecture": "Ampere",
105
+ "uuid": "GPU-8c94bc98-aa6a-d757-b3d0-b74b6f286705"
106
+ },
107
+ {
108
+ "name": "NVIDIA A800-SXM4-80GB",
109
+ "memoryTotal": "85899345920",
110
+ "cudaCores": 6912,
111
+ "architecture": "Ampere",
112
+ "uuid": "GPU-24625a85-1603-30c0-36d9-c3a207715c1f"
113
+ }
114
+ ],
115
+ "cudaVersion": "13.0",
116
+ "writerId": "ck9llrkydvf1se7fpd5kty8afbhldpo1"
117
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/lr":1.0725600182623405e-05,"_wandb":{"runtime":3896},"train/lr_evidence":1e-05,"train/steps":1000,"train/update_s":3.491381091848016,"train/epochs":8.903728436282693,"train/lr_vision":2.5e-06,"train/phase_source_cached":1,"train/samples":64000,"_timestamp":1.793040679184209e+09,"train/flow_loss":0.003335193730890751,"train/phase_norm":16.032630920410156,"train/dataloading_s":0.04767226405441761,"train/lr_vlm":2.5e-06,"train/lr_action":2.5e-06,"train/loss":0.003335193730890751,"_step":1000,"train/grad_norm":0.12221777759492397,"train/episodes":445.1864218141347,"_runtime":3896.710829306}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug-core.log ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-10-27T01:46:34.161700452+08:00","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpy9lxv8xs/port-103747.txt","pid":103747,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
2
+ {"time":"2026-10-27T01:46:34.16222388+08:00","level":"INFO","msg":"server: will exit if parent process dies","ppid":103747}
3
+ {"time":"2026-10-27T01:46:34.162219051+08:00","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-103747-104318-2466334627/socket","Net":"unix"}}
4
+ {"time":"2026-10-27T01:46:34.348800092+08:00","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
5
+ {"time":"2026-10-27T01:46:34.356770825+08:00","level":"INFO","msg":"handleInformInit: received","streamId":"i0fuukqo","id":"1(@)"}
6
+ {"time":"2026-10-27T01:46:35.078317996+08:00","level":"INFO","msg":"handleInformInit: stream started","streamId":"i0fuukqo","id":"1(@)"}
7
+ {"time":"2026-10-27T01:46:41.004619656+08:00","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"dr4c0na8qdst"}
8
+ {"time":"2026-10-27T02:51:32.543511218+08:00","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
9
+ {"time":"2026-10-27T02:51:32.543709995+08:00","level":"INFO","msg":"server is shutting down"}
10
+ {"time":"2026-10-27T02:51:32.543682199+08:00","level":"INFO","msg":"connection: closing","id":"1(@)"}
11
+ {"time":"2026-10-27T02:51:32.543838853+08:00","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
12
+ {"time":"2026-10-27T02:51:32.543885249+08:00","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-103747-104318-2466334627/socket","Net":"unix"}}
13
+ {"time":"2026-10-27T02:51:34.73429679+08:00","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
14
+ {"time":"2026-10-27T02:51:34.734355139+08:00","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
15
+ {"time":"2026-10-27T02:51:34.734373526+08:00","level":"INFO","msg":"server is closed"}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-10-27T01:46:34.356935948+08:00","level":"INFO","msg":"stream: starting","core version":"0.24.2"}
2
+ {"time":"2026-10-27T01:46:35.077973577+08:00","level":"INFO","msg":"stream: created new stream","id":"i0fuukqo"}
3
+ {"time":"2026-10-27T01:46:35.07810336+08:00","level":"INFO","msg":"handler: started","stream_id":"i0fuukqo"}
4
+ {"time":"2026-10-27T01:46:35.078299082+08:00","level":"INFO","msg":"stream: started","id":"i0fuukqo"}
5
+ {"time":"2026-10-27T01:46:35.078327555+08:00","level":"INFO","msg":"writer: started","stream_id":"i0fuukqo"}
6
+ {"time":"2026-10-27T01:46:35.07833363+08:00","level":"INFO","msg":"sender: started","stream_id":"i0fuukqo"}
7
+ {"time":"2026-10-27T02:51:32.543676531+08:00","level":"INFO","msg":"stream: closing","id":"i0fuukqo"}
8
+ {"time":"2026-10-27T02:51:34.333141258+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
9
+ {"time":"2026-10-27T02:51:34.704560495+08:00","level":"INFO","msg":"handler: closed","stream_id":"i0fuukqo"}
10
+ {"time":"2026-10-27T02:51:34.704756589+08:00","level":"INFO","msg":"sender: closed","stream_id":"i0fuukqo"}
11
+ {"time":"2026-10-27T02:51:34.70479882+08:00","level":"INFO","msg":"stream: closed","id":"i0fuukqo"}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/logs/debug.log ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Current SDK version is 0.24.2
2
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Configure stats pid to 103747
3
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():717] Logging user logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug.log
5
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-internal.log
6
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():844] calling init triggers
7
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
8
+ config: {'dataset': {'repo_id': 'local/robotwin_adjust_bottle_aevla50', 'root': '/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50', 'episodes': None, 'image_transforms': {'enable': False, 'max_num_transforms': 3, 'random_order': False, 'tfs': {'brightness': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'brightness': [0.8, 1.2]}}, 'contrast': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'contrast': [0.8, 1.2]}}, 'saturation': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'saturation': [0.5, 1.5]}}, 'hue': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'hue': [-0.05, 0.05]}}, 'sharpness': {'weight': 1.0, 'type': 'SharpnessJitter', 'kwargs': {'sharpness': [0.5, 1.5]}}, 'affine': {'weight': 1.0, 'type': 'RandomAffine', 'kwargs': {'degrees': [-5.0, 5.0], 'translate': [0.05, 0.05]}}}}, 'revision': 'v3.0', 'use_imagenet_stats': True, 'video_backend': 'pyav', 'return_uint8': False, 'streaming': False}, 'env': None, 'policy': {'type': 'action_evidence_vla', 'n_obs_steps': 1, 'input_features': {}, 'output_features': {}, 'device': 'cuda', 'use_amp': False, 'use_peft': False, 'push_to_hub': False, 'repo_id': None, 'private': None, 'tags': None, 'license': None, 'pretrained_path': None, 'pi05_init_path': 'lerobot/pi05_base', 'paligemma_variant': 'gemma_2b', 'action_expert_variant': 'gemma_300m', 'pi05_precision': 'bfloat16', 'chunk_size': 50, 'n_action_steps': 10, 'max_state_dim': 32, 'max_action_dim': 32, 'tokenizer_max_length': 200, 'image_resolution': [224, 224], 'empty_cameras': 0, 'normalization_mapping': {'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>, 'STATE': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'ACTION': <NormalizationMode.QUANTILES: 'QUANTILES'>}, 'num_inference_steps': 10, 'time_sampling_beta_alpha': 1.5, 'time_sampling_beta_beta': 1.0, 'time_sampling_scale': 0.999, 'time_sampling_offset': 0.001, 'min_period': 0.004, 'max_period': 4.0, 'freeze_vlm': False, 'freeze_vision_encoder': False, 'gradient_checkpointing': True, 'offload_frozen_prefix_activations': True, 'phase_state_camera_index': 0, 'phase_state_frame_stride': 2, 'phase_state_vision_chunk_size': 32, 'phase_state_checkpoint_path': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth', 'phase_state_cache_root': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14', 'phase_state_cache_episodes': 2, 'phase_state_max_steps': 1000, 'phase_state_checkpoint_strict': True, 'freeze_phase_state_encoder': True, 'phase_state_action_normalization': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'phase_state_action_dim': 14, 'phase_state_dim': 256, 'phase_state_layers': 4, 'phase_state_dropout': 0.1, 'phase_state_vision_pretrained': False, 'selector_heads': 8, 'evidence_dim': 256, 'use_relative_actions': False, 'relative_exclude_joints': ['gripper'], 'action_feature_names': None, 'optimizer_lr': 2.5e-05, 'evidence_optimizer_lr': 0.0001, 'action_optimizer_lr': 2.5e-05, 'vlm_optimizer_lr': 2.5e-05, 'vision_optimizer_lr': 2.5e-05, 'optimizer_betas': [0.9, 0.95], 'optimizer_eps': 1e-08, 'optimizer_weight_decay': 0.01, 'optimizer_grad_clip_norm': 1.0, 'scheduler_warmup_steps': 33, 'scheduler_decay_steps': 1000, 'scheduler_decay_lr': 2.5e-06}, 'reward_model': None, 'output_dir': '/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'job_name': 'robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'resume': False, 'seed': 42, 'cudnn_deterministic': False, 'num_workers': 4, 'batch_size': 16, 'fsdp': True, 'fsdp_save_training_state': False, 'prefetch_factor': 2, 'persistent_workers': True, 'steps': 1000, 'eval_freq': 0, 'log_freq': 100, 'tolerance_s': 0.0001, 'save_checkpoint': True, 'save_freq': 100, 'use_policy_training_preset': True, 'optimizer': {'type': 'adamw', 'lr': 2.5e-05, 'weight_decay': 0.01, 'grad_clip_norm': 1.0, 'betas': [0.9, 0.95], 'eps': 1e-08}, 'scheduler': {'type': 'cosine_decay_with_warmup', 'num_warmup_steps': 33, 'num_decay_steps': 1000, 'peak_lr': 2.5e-05, 'decay_lr': 2.5e-06}, 'eval': {'n_episodes': 50, 'batch_size': 50, 'use_async_envs': True, 'max_episodes_rendered': 10, 'reseed_policy_each_episode': False}, 'wandb': {'enable': True, 'disable_artifact': True, 'project': 'action-evidence-vla-pi05-phase', 'entity': None, 'notes': None, 'run_id': None, 'mode': 'online', 'add_tags': True}, 'peft': None, 'sample_weighting': None, 'rename_map': {}, 'checkpoint_path': None, '_wandb': {}}
9
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():892] starting backend
10
+ 2026-10-27 01:46:34,348 INFO MainThread:103747 [wandb_init.py:init():895] sending inform_init request
11
+ 2026-10-27 01:46:34,354 INFO MainThread:103747 [wandb_init.py:init():903] backend started and connected
12
+ 2026-10-27 01:46:34,355 INFO MainThread:103747 [wandb_init.py:init():973] updated telemetry
13
+ 2026-10-27 01:46:34,361 INFO MainThread:103747 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
14
+ 2026-10-27 01:46:35,830 INFO MainThread:103747 [wandb_init.py:init():1042] starting run threads in backend
15
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_console_start():2529] atexit reg
16
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2377] redirect: wrap_raw
17
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2446] Wrapping output streams.
18
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2469] Redirects installed.
19
+ 2026-10-27 01:46:36,002 INFO MainThread:103747 [wandb_init.py:init():1082] run started, returning control to user process
20
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [service_client.py:_forward_responses():94] Reached EOF.
21
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [mailbox.py:close():154] Closing mailbox, abandoning 1 handles.
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/latest-run/run-i0fuukqo.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:614d586f8a04e4be4df955accb4757c918412cf396dc1f1748d433eb81962c78
3
+ size 617073
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/config.yaml ADDED
@@ -0,0 +1,351 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.24.2
4
+ e:
5
+ ck9llrkydvf1se7fpd5kty8afbhldpo1:
6
+ args:
7
+ - --fsdp=true
8
+ - --fsdp_save_training_state=false
9
+ - --policy.discover_packages_path=lerobot.policies.action_evidence_vla
10
+ - --policy.type=action_evidence_vla
11
+ - --policy.device=cuda
12
+ - --policy.pi05_precision=bfloat16
13
+ - --policy.normalization_mapping={"VISUAL":"IDENTITY","STATE":"QUANTILES","ACTION":"QUANTILES"}
14
+ - --policy.phase_state_action_normalization=QUANTILES
15
+ - --policy.pi05_init_path=lerobot/pi05_base
16
+ - --policy.phase_state_checkpoint_path=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth
17
+ - --policy.phase_state_cache_root=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
18
+ - --policy.phase_state_checkpoint_strict=true
19
+ - --policy.freeze_phase_state_encoder=true
20
+ - --policy.freeze_vlm=false
21
+ - --policy.freeze_vision_encoder=false
22
+ - --policy.phase_state_camera_index=0
23
+ - --policy.phase_state_frame_stride=2
24
+ - --policy.phase_state_vision_chunk_size=32
25
+ - --policy.n_obs_steps=1
26
+ - --policy.chunk_size=50
27
+ - --policy.n_action_steps=10
28
+ - --policy.max_state_dim=32
29
+ - --policy.max_action_dim=32
30
+ - --policy.phase_state_action_dim=14
31
+ - --policy.empty_cameras=0
32
+ - --policy.tokenizer_max_length=200
33
+ - --policy.gradient_checkpointing=true
34
+ - --policy.evidence_optimizer_lr=0.0001
35
+ - --policy.action_optimizer_lr=0.000025
36
+ - --policy.vlm_optimizer_lr=0.000025
37
+ - --policy.vision_optimizer_lr=0.000025
38
+ - --policy.optimizer_lr=0.000025
39
+ - --policy.optimizer_weight_decay=0.01
40
+ - --policy.optimizer_grad_clip_norm=1.0
41
+ - --policy.scheduler_warmup_steps=33
42
+ - --policy.scheduler_decay_steps=1000
43
+ - --policy.scheduler_decay_lr=0.0000025
44
+ - --policy.push_to_hub=false
45
+ - --dataset.repo_id=local/robotwin_adjust_bottle_aevla50
46
+ - --dataset.root=/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50
47
+ - --dataset.revision=v3.0
48
+ - --dataset.video_backend=pyav
49
+ - --output_dir=/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
50
+ - --job_name=robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
51
+ - --steps=1000
52
+ - --batch_size=16
53
+ - --num_workers=4
54
+ - --persistent_workers=true
55
+ - --prefetch_factor=2
56
+ - --log_freq=100
57
+ - --save_freq=100
58
+ - --eval_freq=0
59
+ - --seed=42
60
+ - --wandb.enable=true
61
+ - --wandb.mode=online
62
+ - --wandb.project=action-evidence-vla-pi05-phase
63
+ - --wandb.disable_artifact=true
64
+ cpu_count: 72
65
+ cpu_count_logical: 144
66
+ cudaVersion: "13.0"
67
+ disk:
68
+ /:
69
+ total: "32212254720"
70
+ used: "9985990656"
71
+ executable: /root/autodl-tmp/projections/SGVLA/.venv/bin/python3
72
+ git:
73
+ commit: ca8da87e9e816c71691333960e7bbc3da699832e
74
+ remote: git@github.com:LliuTtao/SGVLA.git
75
+ gpu: NVIDIA A800-SXM4-80GB
76
+ gpu_count: 4
77
+ gpu_nvidia:
78
+ - architecture: Ampere
79
+ cudaCores: 6912
80
+ memoryTotal: "85899345920"
81
+ name: NVIDIA A800-SXM4-80GB
82
+ uuid: GPU-56601ec2-dc48-6741-6bdd-91c49df6f015
83
+ - architecture: Ampere
84
+ cudaCores: 6912
85
+ memoryTotal: "85899345920"
86
+ name: NVIDIA A800-SXM4-80GB
87
+ uuid: GPU-dfe6761a-23f4-c34c-9c03-01da02020e5f
88
+ - architecture: Ampere
89
+ cudaCores: 6912
90
+ memoryTotal: "85899345920"
91
+ name: NVIDIA A800-SXM4-80GB
92
+ uuid: GPU-8c94bc98-aa6a-d757-b3d0-b74b6f286705
93
+ - architecture: Ampere
94
+ cudaCores: 6912
95
+ memoryTotal: "85899345920"
96
+ name: NVIDIA A800-SXM4-80GB
97
+ uuid: GPU-24625a85-1603-30c0-36d9-c3a207715c1f
98
+ host: autodl-container-sq1jeg5hau-f8a4dbe8
99
+ memory:
100
+ total: "1081807933440"
101
+ os: Linux-5.15.0-78-generic-x86_64-with-glibc2.35
102
+ program: -m lerobot.scripts.lerobot_train
103
+ python: CPython 3.12.3
104
+ root: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
105
+ startedAt: "2026-10-26T17:46:34.131322Z"
106
+ writerId: ck9llrkydvf1se7fpd5kty8afbhldpo1
107
+ m: []
108
+ python_version: 3.12.3
109
+ t:
110
+ "1":
111
+ - 1
112
+ - 11
113
+ - 41
114
+ - 49
115
+ - 51
116
+ - 71
117
+ "2":
118
+ - 1
119
+ - 11
120
+ - 41
121
+ - 49
122
+ - 51
123
+ - 71
124
+ "3":
125
+ - 13
126
+ - 15
127
+ - 16
128
+ - 61
129
+ "4": 3.12.3
130
+ "5": 0.24.2
131
+ "6": 5.5.4
132
+ "10":
133
+ - 21
134
+ "12": 0.24.2
135
+ "13": linux-x86_64
136
+ batch_size:
137
+ value: 16
138
+ checkpoint_path:
139
+ value: null
140
+ cudnn_deterministic:
141
+ value: false
142
+ dataset:
143
+ value:
144
+ episodes: null
145
+ image_transforms:
146
+ enable: false
147
+ max_num_transforms: 3
148
+ random_order: false
149
+ tfs:
150
+ affine:
151
+ kwargs:
152
+ degrees:
153
+ - -5
154
+ - 5
155
+ translate:
156
+ - 0.05
157
+ - 0.05
158
+ type: RandomAffine
159
+ weight: 1
160
+ brightness:
161
+ kwargs:
162
+ brightness:
163
+ - 0.8
164
+ - 1.2
165
+ type: ColorJitter
166
+ weight: 1
167
+ contrast:
168
+ kwargs:
169
+ contrast:
170
+ - 0.8
171
+ - 1.2
172
+ type: ColorJitter
173
+ weight: 1
174
+ hue:
175
+ kwargs:
176
+ hue:
177
+ - -0.05
178
+ - 0.05
179
+ type: ColorJitter
180
+ weight: 1
181
+ saturation:
182
+ kwargs:
183
+ saturation:
184
+ - 0.5
185
+ - 1.5
186
+ type: ColorJitter
187
+ weight: 1
188
+ sharpness:
189
+ kwargs:
190
+ sharpness:
191
+ - 0.5
192
+ - 1.5
193
+ type: SharpnessJitter
194
+ weight: 1
195
+ repo_id: local/robotwin_adjust_bottle_aevla50
196
+ return_uint8: false
197
+ revision: v3.0
198
+ root: /root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50
199
+ streaming: false
200
+ use_imagenet_stats: true
201
+ video_backend: pyav
202
+ env:
203
+ value: null
204
+ eval:
205
+ value:
206
+ batch_size: 50
207
+ max_episodes_rendered: 10
208
+ n_episodes: 50
209
+ reseed_policy_each_episode: false
210
+ use_async_envs: true
211
+ eval_freq:
212
+ value: 0
213
+ fsdp:
214
+ value: true
215
+ fsdp_save_training_state:
216
+ value: false
217
+ job_name:
218
+ value: robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
219
+ log_freq:
220
+ value: 100
221
+ num_workers:
222
+ value: 4
223
+ optimizer:
224
+ value:
225
+ betas:
226
+ - 0.9
227
+ - 0.95
228
+ eps: 1e-08
229
+ grad_clip_norm: 1
230
+ lr: 2.5e-05
231
+ type: adamw
232
+ weight_decay: 0.01
233
+ output_dir:
234
+ value: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
235
+ peft:
236
+ value: null
237
+ persistent_workers:
238
+ value: true
239
+ policy:
240
+ value:
241
+ action_expert_variant: gemma_300m
242
+ action_feature_names: null
243
+ action_optimizer_lr: 2.5e-05
244
+ chunk_size: 50
245
+ device: cuda
246
+ empty_cameras: 0
247
+ evidence_dim: 256
248
+ evidence_optimizer_lr: 0.0001
249
+ freeze_phase_state_encoder: true
250
+ freeze_vision_encoder: false
251
+ freeze_vlm: false
252
+ gradient_checkpointing: true
253
+ image_resolution:
254
+ - 224
255
+ - 224
256
+ license: null
257
+ max_action_dim: 32
258
+ max_period: 4
259
+ max_state_dim: 32
260
+ min_period: 0.004
261
+ n_action_steps: 10
262
+ n_obs_steps: 1
263
+ normalization_mapping:
264
+ ACTION: QUANTILES
265
+ STATE: QUANTILES
266
+ VISUAL: IDENTITY
267
+ num_inference_steps: 10
268
+ offload_frozen_prefix_activations: true
269
+ optimizer_betas:
270
+ - 0.9
271
+ - 0.95
272
+ optimizer_eps: 1e-08
273
+ optimizer_grad_clip_norm: 1
274
+ optimizer_lr: 2.5e-05
275
+ optimizer_weight_decay: 0.01
276
+ paligemma_variant: gemma_2b
277
+ phase_state_action_dim: 14
278
+ phase_state_action_normalization: QUANTILES
279
+ phase_state_cache_episodes: 2
280
+ phase_state_cache_root: /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
281
+ phase_state_camera_index: 0
282
+ phase_state_checkpoint_path: /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth
283
+ phase_state_checkpoint_strict: true
284
+ phase_state_dim: 256
285
+ phase_state_dropout: 0.1
286
+ phase_state_frame_stride: 2
287
+ phase_state_layers: 4
288
+ phase_state_max_steps: 1000
289
+ phase_state_vision_chunk_size: 32
290
+ phase_state_vision_pretrained: false
291
+ pi05_init_path: lerobot/pi05_base
292
+ pi05_precision: bfloat16
293
+ pretrained_path: null
294
+ private: null
295
+ push_to_hub: false
296
+ relative_exclude_joints:
297
+ - gripper
298
+ repo_id: null
299
+ scheduler_decay_lr: 2.5e-06
300
+ scheduler_decay_steps: 1000
301
+ scheduler_warmup_steps: 33
302
+ selector_heads: 8
303
+ tags: null
304
+ time_sampling_beta_alpha: 1.5
305
+ time_sampling_beta_beta: 1
306
+ time_sampling_offset: 0.001
307
+ time_sampling_scale: 0.999
308
+ tokenizer_max_length: 200
309
+ type: action_evidence_vla
310
+ use_amp: false
311
+ use_peft: false
312
+ use_relative_actions: false
313
+ vision_optimizer_lr: 2.5e-05
314
+ vlm_optimizer_lr: 2.5e-05
315
+ prefetch_factor:
316
+ value: 2
317
+ resume:
318
+ value: false
319
+ reward_model:
320
+ value: null
321
+ sample_weighting:
322
+ value: null
323
+ save_checkpoint:
324
+ value: true
325
+ save_freq:
326
+ value: 100
327
+ scheduler:
328
+ value:
329
+ decay_lr: 2.5e-06
330
+ num_decay_steps: 1000
331
+ num_warmup_steps: 33
332
+ peak_lr: 2.5e-05
333
+ type: cosine_decay_with_warmup
334
+ seed:
335
+ value: 42
336
+ steps:
337
+ value: 1000
338
+ tolerance_s:
339
+ value: 0.0001
340
+ use_policy_training_preset:
341
+ value: true
342
+ wandb:
343
+ value:
344
+ add_tags: true
345
+ disable_artifact: true
346
+ enable: true
347
+ entity: null
348
+ mode: online
349
+ notes: null
350
+ project: action-evidence-vla-pi05-phase
351
+ run_id: null
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/output.log ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ INFO 2026-10-27 01:46:36 db_utils.py:121 Logs will be synced with wandb.
2
+ INFO 2026-10-27 01:46:36 db_utils.py:122 Track this run --> https://wandb.ai/3cteam/action-evidence-vla-pi05-phase/runs/i0fuukqo
3
+ INFO 2026-10-27 01:46:36 ot_train.py:338 Creating dataset
4
+ INFO 2026-10-27 01:46:36 ot_train.py:372 Creating policy
5
+ INFO 2026-10-27 01:50:26 ence_vla.py:584 Initialized Action-Evidence VLA from PI0.5 checkpoint lerobot/pi05_base (bare model keys)
6
+ INFO 2026-10-27 01:50:27 ence_vla.py:613 Validated offline causal phase states at /root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14
7
+ INFO 2026-10-27 01:50:36 ot_train.py:449 Creating optimizer and scheduler
8
+ INFO 2026-10-27 01:50:37 ot_train.py:489 Output dir: /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only
9
+ INFO 2026-10-27 01:50:37 ot_train.py:496 cfg.steps=1000 (1K)
10
+ INFO 2026-10-27 01:50:37 ot_train.py:497 dataset.num_frames=7188 (7K)
11
+ INFO 2026-10-27 01:50:37 ot_train.py:498 dataset.num_episodes=50
12
+ INFO 2026-10-27 01:50:37 ot_train.py:501 Effective batch size: 16 x 4 = 64 samples
13
+ INFO 2026-10-27 01:50:37 ot_train.py:504 num_learnable_params=3385530640 (3B)
14
+ INFO 2026-10-27 01:50:37 ot_train.py:505 num_total_params=4175503632 (4B)
15
+ Training: 0%| | 0/1000 [00:00<?, ?step/s]INFO 2026-10-27 01:50:38 ot_train.py:582 Start offline training on a fixed dataset, with effective batch size: 64 samples
16
+ Training: 10%|█ | 100/1000 [05:55<53:07, 3.54s/step]INFO 2026-10-27 01:56:34 ot_train.py:618 step:100 smpl:6K ep:45 epch:0.89 loss:0.091 grdn:0.985 lr:8.4e-05 updt_s:3.495 data_s:0.056
17
+ INFO 2026-10-27 01:56:34 ot_train.py:634 Checkpoint FSDP2 policy after step 100
18
+ INFO 2026-10-27 01:56:48 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000100/pretrained_model/pytorch_model.bin
19
+ Training: 20%|██ | 200/1000 [12:00<46:03, 3.45s/step]INFO 2026-10-27 02:02:38 ot_train.py:618 step:200 smpl:13K ep:89 epch:1.78 loss:0.016 grdn:0.259 lr:9.5e-05 updt_s:3.450 data_s:0.047
20
+ INFO 2026-10-27 02:02:38 ot_train.py:634 Checkpoint FSDP2 policy after step 200
21
+ INFO 2026-10-27 02:02:50 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000200/pretrained_model/pytorch_model.bin
22
+ Training: 30%|███ | 300/1000 [18:04<40:11, 3.45s/step]INFO 2026-10-27 02:08:43 ot_train.py:618 step:300 smpl:19K ep:134 epch:2.67 loss:0.011 grdn:0.210 lr:8.7e-05 updt_s:3.465 data_s:0.046
23
+ INFO 2026-10-27 02:08:43 ot_train.py:634 Checkpoint FSDP2 policy after step 300
24
+ INFO 2026-10-27 02:08:54 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000300/pretrained_model/pytorch_model.bin
25
+ Training: 40%|████ | 400/1000 [24:09<40:57, 4.10s/step]INFO 2026-10-27 02:14:47 ot_train.py:618 step:400 smpl:26K ep:178 epch:3.56 loss:0.009 grdn:0.190 lr:7.5e-05 updt_s:3.467 data_s:0.049
26
+ INFO 2026-10-27 02:14:47 ot_train.py:634 Checkpoint FSDP2 policy after step 400
27
+ INFO 2026-10-27 02:14:59 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000400/pretrained_model/pytorch_model.bin
28
+ Training: 50%|█████ | 500/1000 [30:11<29:04, 3.49s/step]INFO 2026-10-27 02:20:50 ot_train.py:618 step:500 smpl:32K ep:223 epch:4.45 loss:0.007 grdn:0.170 lr:6.2e-05 updt_s:3.467 data_s:0.028
29
+ INFO 2026-10-27 02:20:50 ot_train.py:634 Checkpoint FSDP2 policy after step 500
30
+ INFO 2026-10-27 02:21:02 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000500/pretrained_model/pytorch_model.bin
31
+ Training: 60%|██████ | 600/1000 [36:17<23:33, 3.53s/step]INFO 2026-10-27 02:26:56 ot_train.py:618 step:600 smpl:38K ep:267 epch:5.34 loss:0.006 grdn:0.148 lr:4.8e-05 updt_s:3.475 data_s:0.048
32
+ INFO 2026-10-27 02:26:56 ot_train.py:634 Checkpoint FSDP2 policy after step 600
33
+ INFO 2026-10-27 02:27:08 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000600/pretrained_model/pytorch_model.bin
34
+ Training: 70%|███████ | 700/1000 [42:23<17:21, 3.47s/step]INFO 2026-10-27 02:33:02 ot_train.py:618 step:700 smpl:45K ep:312 epch:6.23 loss:0.005 grdn:0.149 lr:3.5e-05 updt_s:3.476 data_s:0.047
35
+ INFO 2026-10-27 02:33:02 ot_train.py:634 Checkpoint FSDP2 policy after step 700
36
+ INFO 2026-10-27 02:33:14 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000700/pretrained_model/pytorch_model.bin
37
+ Training: 80%|████████ | 800/1000 [48:29<13:13, 3.97s/step]INFO 2026-10-27 02:39:07 ot_train.py:618 step:800 smpl:51K ep:356 epch:7.12 loss:0.004 grdn:0.128 lr:2.3e-05 updt_s:3.473 data_s:0.049
38
+ INFO 2026-10-27 02:39:07 ot_train.py:634 Checkpoint FSDP2 policy after step 800
39
+ INFO 2026-10-27 02:39:19 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000800/pretrained_model/pytorch_model.bin
40
+ Training: 90%|█████████ | 900/1000 [54:33<05:49, 3.49s/step]INFO 2026-10-27 02:45:11 ot_train.py:618 step:900 smpl:58K ep:401 epch:8.01 loss:0.004 grdn:0.122 lr:1.5e-05 updt_s:3.481 data_s:0.029
41
+ INFO 2026-10-27 02:45:11 ot_train.py:634 Checkpoint FSDP2 policy after step 900
42
+ INFO 2026-10-27 02:45:24 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/000900/pretrained_model/pytorch_model.bin
43
+ Training: 100%|██████████| 1000/1000 [1:00:40<00:00, 3.49s/step]INFO 2026-10-27 02:51:19 ot_train.py:618 step:1K smpl:64K ep:445 epch:8.90 loss:0.004 grdn:0.122 lr:1.1e-05 updt_s:3.491 data_s:0.048
44
+ INFO 2026-10-27 02:51:19 ot_train.py:634 Checkpoint FSDP2 policy after step 1000
45
+ INFO 2026-10-27 02:51:31 lerator.py:3591 [RANK 0] Model weights saved in /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/checkpoints/001000/pretrained_model/pytorch_model.bin
46
+ Training: 100%|██████████| 1000/1000 [1:00:52<00:00, 3.65s/step]
47
+ INFO 2026-10-27 02:51:31 ot_train.py:723 End of training
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/requirements.txt ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ lerobot==0.5.2
2
+ Wand==0.6.13
3
+ absl-py==2.5.0
4
+ accelerate==1.15.0
5
+ aiohappyeyeballs==2.7.1
6
+ aiohttp==3.14.3
7
+ aiosignal==1.4.0
8
+ annotated-doc==0.0.5
9
+ annotated-types==0.8.0
10
+ antlr4-python3-runtime==4.9.3
11
+ anyio==4.15.1
12
+ attrs==26.1.0
13
+ av==15.1.0
14
+ bddl==1.0.1
15
+ certifi==2026.7.22
16
+ charset-normalizer==3.5.1
17
+ click==8.5.0
18
+ cloudpickle==3.1.2
19
+ cmake==3.31.10
20
+ contourpy==1.3.3
21
+ cuda-bindings==12.9.7
22
+ cuda-pathfinder==1.8.1
23
+ cuda-toolkit==12.8.1
24
+ cycler==0.12.1
25
+ datasets==4.8.5
26
+ dill==0.4.1
27
+ draccus==0.10.0
28
+ easydict==1.13
29
+ egl_probe==1.0.2
30
+ einops==0.8.2
31
+ etils==1.14.0
32
+ evdev==2.0.0
33
+ Farama-Notifications==0.0.6
34
+ fastjsonschema==2.22.2
35
+ filelock==3.32.6
36
+ fonttools==4.64.0
37
+ frozenlist==1.8.0
38
+ fsspec==2026.2.0
39
+ future==1.0.0
40
+ gitdb==4.0.12
41
+ GitPython==3.1.62
42
+ glfw==2.10.2
43
+ grpcio==1.73.1
44
+ gym==0.26.2
45
+ gym-notices==0.1.0
46
+ gymnasium==1.3.0
47
+ h11==0.16.0
48
+ h5py==3.16.0
49
+ hf_egl_probe==1.0.2
50
+ hf-xet==1.6.0
51
+ httpcore==1.0.9
52
+ httpx==0.28.1
53
+ huggingface_hub==1.30.0
54
+ hydra-core==1.3.6
55
+ idna==3.19
56
+ ImageIO==2.37.4
57
+ imageio-ffmpeg==0.6.0
58
+ iniconfig==2.3.0
59
+ Jinja2==3.1.6
60
+ jsonlines==4.0.0
61
+ jsonschema==4.26.0
62
+ jsonschema-specifications==2025.9.1
63
+ jupyter_core==5.9.1
64
+ jupytext==1.19.5
65
+ kiwisolver==1.5.1
66
+ lazy-loader==0.6
67
+ lerobot==0.5.2
68
+ libero==0.1.0
69
+ llvmlite==0.49.0
70
+ Markdown==3.10.3
71
+ markdown-it-py==4.2.0
72
+ MarkupSafe==3.0.3
73
+ matplotlib==3.11.1
74
+ mdit-py-plugins==0.6.1
75
+ mdurl==0.1.2
76
+ mergedeep==1.3.4
77
+ mpmath==1.3.0
78
+ mujoco==3.7.0
79
+ multidict==6.8.0
80
+ multiprocess==0.70.19
81
+ mypy_extensions==1.1.0
82
+ nbformat==5.11.1
83
+ networkx==3.6.1
84
+ numba==0.67.0
85
+ numpy==2.2.6
86
+ nvidia-cublas-cu12==12.8.4.1
87
+ nvidia-cuda-cupti-cu12==12.8.90
88
+ nvidia-cuda-nvrtc-cu12==12.8.93
89
+ nvidia-cuda-runtime-cu12==12.8.90
90
+ nvidia-cudnn-cu12==9.19.0.56
91
+ nvidia-cufft-cu12==11.3.3.83
92
+ nvidia-cufile-cu12==1.13.1.3
93
+ nvidia-curand-cu12==10.3.9.90
94
+ nvidia-cusolver-cu12==11.7.3.90
95
+ nvidia-cusparse-cu12==12.5.8.93
96
+ nvidia-cusparselt-cu12==0.7.1
97
+ nvidia-nccl-cu12==2.28.9
98
+ nvidia-nvjitlink-cu12==12.8.93
99
+ nvidia-nvshmem-cu12==3.4.5
100
+ nvidia-nvtx-cu12==12.8.90
101
+ omegaconf==2.3.1
102
+ opencv-python==4.14.0.94
103
+ opencv-python-headless==4.13.0.92
104
+ packaging==25.0
105
+ pandas==2.3.3
106
+ pillow==12.3.0
107
+ platformdirs==4.11.8
108
+ pluggy==1.6.0
109
+ propcache==0.5.2
110
+ protobuf==6.31.1
111
+ psutil==7.2.2
112
+ pyarrow==25.0.1
113
+ pydantic==2.13.5
114
+ pydantic_core==2.46.5
115
+ Pygments==2.21.0
116
+ pynput==1.8.2
117
+ PyOpenGL==3.1.10
118
+ pyparsing==3.3.2
119
+ pytest==8.4.2
120
+ python-dateutil==2.9.0.post0
121
+ python-xlib==0.33
122
+ pytz==2026.3.post1
123
+ PyYAML==6.0.3
124
+ pyyaml-include==1.4.1
125
+ referencing==0.37.0
126
+ regex==2026.9.3
127
+ requests==2.34.2
128
+ rich==15.0.0
129
+ robomimic==0.2.0
130
+ robosuite==1.4.1
131
+ rpds-py==2026.6.3
132
+ safetensors==0.8.0
133
+ scikit-image==0.25.2
134
+ scipy==1.18.1
135
+ sentry-sdk==2.69.1
136
+ setuptools==80.10.2
137
+ shellingham==1.5.4
138
+ six==1.17.0
139
+ smmap==5.0.3
140
+ sympy==1.14.0
141
+ tensorboard==2.20.0
142
+ tensorboard-data-server==0.7.2
143
+ tensorboardX==2.6.5
144
+ termcolor==3.3.0
145
+ thop==0.1.1-2209072238
146
+ tifffile==2026.9.20
147
+ timm==1.0.29
148
+ tokenizers==0.22.2
149
+ toml==0.10.2
150
+ torch==2.11.0+cu128
151
+ torchcodec==0.11.1
152
+ torchvision==0.26.0+cu128
153
+ tqdm==4.70.0
154
+ traitlets==5.16.1
155
+ transformers==5.5.4
156
+ triton==3.6.0
157
+ typer==0.27.2
158
+ typing_extensions==4.16.0
159
+ typing-inspect==0.9.0
160
+ typing-inspection==0.4.4
161
+ tzdata==2026.3
162
+ urllib3==2.7.0
163
+ wandb==0.24.2
164
+ Werkzeug==3.1.8
165
+ xxhash==4.0.1
166
+ yarl==1.24.5
167
+ zipp==4.1.0
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/wandb-metadata.json ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-5.15.0-78-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.12.3",
4
+ "startedAt": "2026-10-26T17:46:34.131322Z",
5
+ "args": [
6
+ "--fsdp=true",
7
+ "--fsdp_save_training_state=false",
8
+ "--policy.discover_packages_path=lerobot.policies.action_evidence_vla",
9
+ "--policy.type=action_evidence_vla",
10
+ "--policy.device=cuda",
11
+ "--policy.pi05_precision=bfloat16",
12
+ "--policy.normalization_mapping={\"VISUAL\":\"IDENTITY\",\"STATE\":\"QUANTILES\",\"ACTION\":\"QUANTILES\"}",
13
+ "--policy.phase_state_action_normalization=QUANTILES",
14
+ "--policy.pi05_init_path=lerobot/pi05_base",
15
+ "--policy.phase_state_checkpoint_path=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth",
16
+ "--policy.phase_state_cache_root=/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14",
17
+ "--policy.phase_state_checkpoint_strict=true",
18
+ "--policy.freeze_phase_state_encoder=true",
19
+ "--policy.freeze_vlm=false",
20
+ "--policy.freeze_vision_encoder=false",
21
+ "--policy.phase_state_camera_index=0",
22
+ "--policy.phase_state_frame_stride=2",
23
+ "--policy.phase_state_vision_chunk_size=32",
24
+ "--policy.n_obs_steps=1",
25
+ "--policy.chunk_size=50",
26
+ "--policy.n_action_steps=10",
27
+ "--policy.max_state_dim=32",
28
+ "--policy.max_action_dim=32",
29
+ "--policy.phase_state_action_dim=14",
30
+ "--policy.empty_cameras=0",
31
+ "--policy.tokenizer_max_length=200",
32
+ "--policy.gradient_checkpointing=true",
33
+ "--policy.evidence_optimizer_lr=0.0001",
34
+ "--policy.action_optimizer_lr=0.000025",
35
+ "--policy.vlm_optimizer_lr=0.000025",
36
+ "--policy.vision_optimizer_lr=0.000025",
37
+ "--policy.optimizer_lr=0.000025",
38
+ "--policy.optimizer_weight_decay=0.01",
39
+ "--policy.optimizer_grad_clip_norm=1.0",
40
+ "--policy.scheduler_warmup_steps=33",
41
+ "--policy.scheduler_decay_steps=1000",
42
+ "--policy.scheduler_decay_lr=0.0000025",
43
+ "--policy.push_to_hub=false",
44
+ "--dataset.repo_id=local/robotwin_adjust_bottle_aevla50",
45
+ "--dataset.root=/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50",
46
+ "--dataset.revision=v3.0",
47
+ "--dataset.video_backend=pyav",
48
+ "--output_dir=/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
49
+ "--job_name=robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
50
+ "--steps=1000",
51
+ "--batch_size=16",
52
+ "--num_workers=4",
53
+ "--persistent_workers=true",
54
+ "--prefetch_factor=2",
55
+ "--log_freq=100",
56
+ "--save_freq=100",
57
+ "--eval_freq=0",
58
+ "--seed=42",
59
+ "--wandb.enable=true",
60
+ "--wandb.mode=online",
61
+ "--wandb.project=action-evidence-vla-pi05-phase",
62
+ "--wandb.disable_artifact=true"
63
+ ],
64
+ "program": "-m lerobot.scripts.lerobot_train",
65
+ "git": {
66
+ "remote": "git@github.com:LliuTtao/SGVLA.git",
67
+ "commit": "ca8da87e9e816c71691333960e7bbc3da699832e"
68
+ },
69
+ "root": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only",
70
+ "host": "autodl-container-sq1jeg5hau-f8a4dbe8",
71
+ "executable": "/root/autodl-tmp/projections/SGVLA/.venv/bin/python3",
72
+ "cpu_count": 72,
73
+ "cpu_count_logical": 144,
74
+ "gpu": "NVIDIA A800-SXM4-80GB",
75
+ "gpu_count": 4,
76
+ "disk": {
77
+ "/": {
78
+ "total": "32212254720",
79
+ "used": "9985990656"
80
+ }
81
+ },
82
+ "memory": {
83
+ "total": "1081807933440"
84
+ },
85
+ "gpu_nvidia": [
86
+ {
87
+ "name": "NVIDIA A800-SXM4-80GB",
88
+ "memoryTotal": "85899345920",
89
+ "cudaCores": 6912,
90
+ "architecture": "Ampere",
91
+ "uuid": "GPU-56601ec2-dc48-6741-6bdd-91c49df6f015"
92
+ },
93
+ {
94
+ "name": "NVIDIA A800-SXM4-80GB",
95
+ "memoryTotal": "85899345920",
96
+ "cudaCores": 6912,
97
+ "architecture": "Ampere",
98
+ "uuid": "GPU-dfe6761a-23f4-c34c-9c03-01da02020e5f"
99
+ },
100
+ {
101
+ "name": "NVIDIA A800-SXM4-80GB",
102
+ "memoryTotal": "85899345920",
103
+ "cudaCores": 6912,
104
+ "architecture": "Ampere",
105
+ "uuid": "GPU-8c94bc98-aa6a-d757-b3d0-b74b6f286705"
106
+ },
107
+ {
108
+ "name": "NVIDIA A800-SXM4-80GB",
109
+ "memoryTotal": "85899345920",
110
+ "cudaCores": 6912,
111
+ "architecture": "Ampere",
112
+ "uuid": "GPU-24625a85-1603-30c0-36d9-c3a207715c1f"
113
+ }
114
+ ],
115
+ "cudaVersion": "13.0",
116
+ "writerId": "ck9llrkydvf1se7fpd5kty8afbhldpo1"
117
+ }
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/lr":1.0725600182623405e-05,"_wandb":{"runtime":3896},"train/lr_evidence":1e-05,"train/steps":1000,"train/update_s":3.491381091848016,"train/epochs":8.903728436282693,"train/lr_vision":2.5e-06,"train/phase_source_cached":1,"train/samples":64000,"_timestamp":1.793040679184209e+09,"train/flow_loss":0.003335193730890751,"train/phase_norm":16.032630920410156,"train/dataloading_s":0.04767226405441761,"train/lr_vlm":2.5e-06,"train/lr_action":2.5e-06,"train/loss":0.003335193730890751,"_step":1000,"train/grad_norm":0.12221777759492397,"train/episodes":445.1864218141347,"_runtime":3896.710829306}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-core.log ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-10-27T01:46:34.161700452+08:00","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpy9lxv8xs/port-103747.txt","pid":103747,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
2
+ {"time":"2026-10-27T01:46:34.16222388+08:00","level":"INFO","msg":"server: will exit if parent process dies","ppid":103747}
3
+ {"time":"2026-10-27T01:46:34.162219051+08:00","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-103747-104318-2466334627/socket","Net":"unix"}}
4
+ {"time":"2026-10-27T01:46:34.348800092+08:00","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
5
+ {"time":"2026-10-27T01:46:34.356770825+08:00","level":"INFO","msg":"handleInformInit: received","streamId":"i0fuukqo","id":"1(@)"}
6
+ {"time":"2026-10-27T01:46:35.078317996+08:00","level":"INFO","msg":"handleInformInit: stream started","streamId":"i0fuukqo","id":"1(@)"}
7
+ {"time":"2026-10-27T01:46:41.004619656+08:00","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"dr4c0na8qdst"}
8
+ {"time":"2026-10-27T02:51:32.543511218+08:00","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
9
+ {"time":"2026-10-27T02:51:32.543709995+08:00","level":"INFO","msg":"server is shutting down"}
10
+ {"time":"2026-10-27T02:51:32.543682199+08:00","level":"INFO","msg":"connection: closing","id":"1(@)"}
11
+ {"time":"2026-10-27T02:51:32.543838853+08:00","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
12
+ {"time":"2026-10-27T02:51:32.543885249+08:00","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-103747-104318-2466334627/socket","Net":"unix"}}
13
+ {"time":"2026-10-27T02:51:34.73429679+08:00","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
14
+ {"time":"2026-10-27T02:51:34.734355139+08:00","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
15
+ {"time":"2026-10-27T02:51:34.734373526+08:00","level":"INFO","msg":"server is closed"}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-10-27T01:46:34.356935948+08:00","level":"INFO","msg":"stream: starting","core version":"0.24.2"}
2
+ {"time":"2026-10-27T01:46:35.077973577+08:00","level":"INFO","msg":"stream: created new stream","id":"i0fuukqo"}
3
+ {"time":"2026-10-27T01:46:35.07810336+08:00","level":"INFO","msg":"handler: started","stream_id":"i0fuukqo"}
4
+ {"time":"2026-10-27T01:46:35.078299082+08:00","level":"INFO","msg":"stream: started","id":"i0fuukqo"}
5
+ {"time":"2026-10-27T01:46:35.078327555+08:00","level":"INFO","msg":"writer: started","stream_id":"i0fuukqo"}
6
+ {"time":"2026-10-27T01:46:35.07833363+08:00","level":"INFO","msg":"sender: started","stream_id":"i0fuukqo"}
7
+ {"time":"2026-10-27T02:51:32.543676531+08:00","level":"INFO","msg":"stream: closing","id":"i0fuukqo"}
8
+ {"time":"2026-10-27T02:51:34.333141258+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
9
+ {"time":"2026-10-27T02:51:34.704560495+08:00","level":"INFO","msg":"handler: closed","stream_id":"i0fuukqo"}
10
+ {"time":"2026-10-27T02:51:34.704756589+08:00","level":"INFO","msg":"sender: closed","stream_id":"i0fuukqo"}
11
+ {"time":"2026-10-27T02:51:34.70479882+08:00","level":"INFO","msg":"stream: closed","id":"i0fuukqo"}
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug.log ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Current SDK version is 0.24.2
2
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Configure stats pid to 103747
3
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():717] Logging user logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug.log
5
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to /root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/logs/debug-internal.log
6
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():844] calling init triggers
7
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
8
+ config: {'dataset': {'repo_id': 'local/robotwin_adjust_bottle_aevla50', 'root': '/root/autodl-fs/datasets/robotwin_adjust_bottle_aevla50', 'episodes': None, 'image_transforms': {'enable': False, 'max_num_transforms': 3, 'random_order': False, 'tfs': {'brightness': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'brightness': [0.8, 1.2]}}, 'contrast': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'contrast': [0.8, 1.2]}}, 'saturation': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'saturation': [0.5, 1.5]}}, 'hue': {'weight': 1.0, 'type': 'ColorJitter', 'kwargs': {'hue': [-0.05, 0.05]}}, 'sharpness': {'weight': 1.0, 'type': 'SharpnessJitter', 'kwargs': {'sharpness': [0.5, 1.5]}}, 'affine': {'weight': 1.0, 'type': 'RandomAffine', 'kwargs': {'degrees': [-5.0, 5.0], 'translate': [0.05, 0.05]}}}}, 'revision': 'v3.0', 'use_imagenet_stats': True, 'video_backend': 'pyav', 'return_uint8': False, 'streaming': False}, 'env': None, 'policy': {'type': 'action_evidence_vla', 'n_obs_steps': 1, 'input_features': {}, 'output_features': {}, 'device': 'cuda', 'use_amp': False, 'use_peft': False, 'push_to_hub': False, 'repo_id': None, 'private': None, 'tags': None, 'license': None, 'pretrained_path': None, 'pi05_init_path': 'lerobot/pi05_base', 'paligemma_variant': 'gemma_2b', 'action_expert_variant': 'gemma_300m', 'pi05_precision': 'bfloat16', 'chunk_size': 50, 'n_action_steps': 10, 'max_state_dim': 32, 'max_action_dim': 32, 'tokenizer_max_length': 200, 'image_resolution': [224, 224], 'empty_cameras': 0, 'normalization_mapping': {'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>, 'STATE': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'ACTION': <NormalizationMode.QUANTILES: 'QUANTILES'>}, 'num_inference_steps': 10, 'time_sampling_beta_alpha': 1.5, 'time_sampling_beta_beta': 1.0, 'time_sampling_scale': 0.999, 'time_sampling_offset': 0.001, 'min_period': 0.004, 'max_period': 4.0, 'freeze_vlm': False, 'freeze_vision_encoder': False, 'gradient_checkpointing': True, 'offload_frozen_prefix_activations': True, 'phase_state_camera_index': 0, 'phase_state_frame_stride': 2, 'phase_state_vision_chunk_size': 32, 'phase_state_checkpoint_path': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_adjust_bottle_vbe14/best_val_loss.pth', 'phase_state_cache_root': '/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_adjust_bottle_vbe14', 'phase_state_cache_episodes': 2, 'phase_state_max_steps': 1000, 'phase_state_checkpoint_strict': True, 'freeze_phase_state_encoder': True, 'phase_state_action_normalization': <NormalizationMode.QUANTILES: 'QUANTILES'>, 'phase_state_action_dim': 14, 'phase_state_dim': 256, 'phase_state_layers': 4, 'phase_state_dropout': 0.1, 'phase_state_vision_pretrained': False, 'selector_heads': 8, 'evidence_dim': 256, 'use_relative_actions': False, 'relative_exclude_joints': ['gripper'], 'action_feature_names': None, 'optimizer_lr': 2.5e-05, 'evidence_optimizer_lr': 0.0001, 'action_optimizer_lr': 2.5e-05, 'vlm_optimizer_lr': 2.5e-05, 'vision_optimizer_lr': 2.5e-05, 'optimizer_betas': [0.9, 0.95], 'optimizer_eps': 1e-08, 'optimizer_weight_decay': 0.01, 'optimizer_grad_clip_norm': 1.0, 'scheduler_warmup_steps': 33, 'scheduler_decay_steps': 1000, 'scheduler_decay_lr': 2.5e-06}, 'reward_model': None, 'output_dir': '/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'job_name': 'robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only', 'resume': False, 'seed': 42, 'cudnn_deterministic': False, 'num_workers': 4, 'batch_size': 16, 'fsdp': True, 'fsdp_save_training_state': False, 'prefetch_factor': 2, 'persistent_workers': True, 'steps': 1000, 'eval_freq': 0, 'log_freq': 100, 'tolerance_s': 0.0001, 'save_checkpoint': True, 'save_freq': 100, 'use_policy_training_preset': True, 'optimizer': {'type': 'adamw', 'lr': 2.5e-05, 'weight_decay': 0.01, 'grad_clip_norm': 1.0, 'betas': [0.9, 0.95], 'eps': 1e-08}, 'scheduler': {'type': 'cosine_decay_with_warmup', 'num_warmup_steps': 33, 'num_decay_steps': 1000, 'peak_lr': 2.5e-05, 'decay_lr': 2.5e-06}, 'eval': {'n_episodes': 50, 'batch_size': 50, 'use_async_envs': True, 'max_episodes_rendered': 10, 'reseed_policy_each_episode': False}, 'wandb': {'enable': True, 'disable_artifact': True, 'project': 'action-evidence-vla-pi05-phase', 'entity': None, 'notes': None, 'run_id': None, 'mode': 'online', 'add_tags': True}, 'peft': None, 'sample_weighting': None, 'rename_map': {}, 'checkpoint_path': None, '_wandb': {}}
9
+ 2026-10-27 01:46:34,132 INFO MainThread:103747 [wandb_init.py:init():892] starting backend
10
+ 2026-10-27 01:46:34,348 INFO MainThread:103747 [wandb_init.py:init():895] sending inform_init request
11
+ 2026-10-27 01:46:34,354 INFO MainThread:103747 [wandb_init.py:init():903] backend started and connected
12
+ 2026-10-27 01:46:34,355 INFO MainThread:103747 [wandb_init.py:init():973] updated telemetry
13
+ 2026-10-27 01:46:34,361 INFO MainThread:103747 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
14
+ 2026-10-27 01:46:35,830 INFO MainThread:103747 [wandb_init.py:init():1042] starting run threads in backend
15
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_console_start():2529] atexit reg
16
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2377] redirect: wrap_raw
17
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2446] Wrapping output streams.
18
+ 2026-10-27 01:46:35,998 INFO MainThread:103747 [wandb_run.py:_redirect():2469] Redirects installed.
19
+ 2026-10-27 01:46:36,002 INFO MainThread:103747 [wandb_init.py:init():1082] run started, returning control to user process
20
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [service_client.py:_forward_responses():94] Reached EOF.
21
+ 2026-10-27 02:51:32,543 INFO wandb-AsyncioManager-main:103747 [mailbox.py:close():154] Closing mailbox, abandoning 1 handles.
robotwin_adjust_bottle_stage2_fsdp2_gb64_1k_model_only/wandb/run-20261027_014634-i0fuukqo/run-i0fuukqo.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:614d586f8a04e4be4df955accb4757c918412cf396dc1f1748d433eb81962c78
3
+ size 617073
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "action_evidence_vla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.cam_high": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 480,
9
+ 640,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.cam_left_wrist": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 480,
17
+ 640,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.images.cam_right_wrist": {
22
+ "type": "VISUAL",
23
+ "shape": [
24
+ 480,
25
+ 640,
26
+ 3
27
+ ]
28
+ },
29
+ "observation.state": {
30
+ "type": "STATE",
31
+ "shape": [
32
+ 14
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": false,
48
+ "repo_id": null,
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": null,
53
+ "pi05_init_path": "lerobot/pi05_base",
54
+ "paligemma_variant": "gemma_2b",
55
+ "action_expert_variant": "gemma_300m",
56
+ "pi05_precision": "bfloat16",
57
+ "chunk_size": 50,
58
+ "n_action_steps": 10,
59
+ "max_state_dim": 32,
60
+ "max_action_dim": 32,
61
+ "tokenizer_max_length": 200,
62
+ "image_resolution": [
63
+ 224,
64
+ 224
65
+ ],
66
+ "empty_cameras": 0,
67
+ "normalization_mapping": {
68
+ "VISUAL": "IDENTITY",
69
+ "STATE": "QUANTILES",
70
+ "ACTION": "QUANTILES"
71
+ },
72
+ "num_inference_steps": 10,
73
+ "time_sampling_beta_alpha": 1.5,
74
+ "time_sampling_beta_beta": 1.0,
75
+ "time_sampling_scale": 0.999,
76
+ "time_sampling_offset": 0.001,
77
+ "min_period": 0.004,
78
+ "max_period": 4.0,
79
+ "freeze_vlm": false,
80
+ "freeze_vision_encoder": false,
81
+ "gradient_checkpointing": true,
82
+ "offload_frozen_prefix_activations": true,
83
+ "phase_state_camera_index": 0,
84
+ "phase_state_frame_stride": 2,
85
+ "phase_state_vision_chunk_size": 32,
86
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_click_bell_vbe14/best_val_loss.pth",
87
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_click_bell_vbe14",
88
+ "phase_state_cache_episodes": 2,
89
+ "phase_state_max_steps": 1000,
90
+ "phase_state_checkpoint_strict": true,
91
+ "freeze_phase_state_encoder": true,
92
+ "phase_state_action_normalization": "QUANTILES",
93
+ "phase_state_action_dim": 14,
94
+ "phase_state_dim": 256,
95
+ "phase_state_layers": 4,
96
+ "phase_state_dropout": 0.1,
97
+ "phase_state_vision_pretrained": false,
98
+ "selector_heads": 8,
99
+ "evidence_dim": 256,
100
+ "use_relative_actions": false,
101
+ "relative_exclude_joints": [
102
+ "gripper"
103
+ ],
104
+ "action_feature_names": [
105
+ "motors"
106
+ ],
107
+ "optimizer_lr": 2.5e-05,
108
+ "evidence_optimizer_lr": 0.0001,
109
+ "action_optimizer_lr": 2.5e-05,
110
+ "vlm_optimizer_lr": 2.5e-05,
111
+ "vision_optimizer_lr": 2.5e-05,
112
+ "optimizer_betas": [
113
+ 0.9,
114
+ 0.95
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 0.01,
118
+ "optimizer_grad_clip_norm": 1.0,
119
+ "scheduler_warmup_steps": 33,
120
+ "scheduler_decay_steps": 1000,
121
+ "scheduler_decay_lr": 2.5e-06
122
+ }
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7067d00abf2572104fdf0df0c64757416851c1d78c9347f77734c11fa4bf5b55
3
+ size 14437807336
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 14
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "IDENTITY",
18
+ "STATE": "QUANTILES",
19
+ "ACTION": "QUANTILES"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "absolute_actions_processor",
26
+ "config": {
27
+ "enabled": false
28
+ }
29
+ },
30
+ {
31
+ "registry_name": "device_processor",
32
+ "config": {
33
+ "device": "cpu",
34
+ "float_dtype": null
35
+ }
36
+ }
37
+ ]
38
+ }
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2414b7c506df464d6f8bdb3973e45e04a9b0f8648b90b74be8cf1847a80e238
3
+ size 6792
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "delta_actions_processor",
16
+ "config": {
17
+ "enabled": false,
18
+ "exclude_joints": [
19
+ "gripper"
20
+ ],
21
+ "action_names": [
22
+ "motors"
23
+ ]
24
+ }
25
+ },
26
+ {
27
+ "registry_name": "normalizer_processor",
28
+ "config": {
29
+ "eps": 1e-08,
30
+ "features": {
31
+ "observation.images.cam_high": {
32
+ "type": "VISUAL",
33
+ "shape": [
34
+ 480,
35
+ 640,
36
+ 3
37
+ ]
38
+ },
39
+ "observation.images.cam_left_wrist": {
40
+ "type": "VISUAL",
41
+ "shape": [
42
+ 480,
43
+ 640,
44
+ 3
45
+ ]
46
+ },
47
+ "observation.images.cam_right_wrist": {
48
+ "type": "VISUAL",
49
+ "shape": [
50
+ 480,
51
+ 640,
52
+ 3
53
+ ]
54
+ },
55
+ "observation.state": {
56
+ "type": "STATE",
57
+ "shape": [
58
+ 14
59
+ ]
60
+ },
61
+ "action": {
62
+ "type": "ACTION",
63
+ "shape": [
64
+ 14
65
+ ]
66
+ }
67
+ },
68
+ "norm_map": {
69
+ "VISUAL": "IDENTITY",
70
+ "STATE": "QUANTILES",
71
+ "ACTION": "QUANTILES"
72
+ }
73
+ },
74
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
75
+ },
76
+ {
77
+ "registry_name": "action_evidence_prepare_anchor",
78
+ "config": {}
79
+ },
80
+ {
81
+ "registry_name": "action_evidence_prepare_pi05_prompt",
82
+ "config": {}
83
+ },
84
+ {
85
+ "registry_name": "tokenizer_processor",
86
+ "config": {
87
+ "max_length": 200,
88
+ "task_key": "task",
89
+ "padding_side": "right",
90
+ "padding": "max_length",
91
+ "truncation": true,
92
+ "tokenizer_name": "google/paligemma-3b-pt-224"
93
+ }
94
+ },
95
+ {
96
+ "registry_name": "device_processor",
97
+ "config": {
98
+ "device": "cuda",
99
+ "float_dtype": null
100
+ }
101
+ }
102
+ ]
103
+ }
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f89cc8c6973d0d96fa65c60b61f2fd1f2e9ea359e8fe603fc300ec3dc2d8af0e
3
+ size 6808
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/000800/pretrained_model/train_config.json ADDED
@@ -0,0 +1,264 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/robotwin_click_bell_aevla50",
4
+ "root": "/root/autodl-fs/datasets/robotwin_click_bell_aevla50",
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "v3.0",
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "pyav",
80
+ "return_uint8": false,
81
+ "streaming": false
82
+ },
83
+ "env": null,
84
+ "policy": {
85
+ "type": "action_evidence_vla",
86
+ "n_obs_steps": 1,
87
+ "input_features": {
88
+ "observation.images.cam_high": {
89
+ "type": "VISUAL",
90
+ "shape": [
91
+ 480,
92
+ 640,
93
+ 3
94
+ ]
95
+ },
96
+ "observation.images.cam_left_wrist": {
97
+ "type": "VISUAL",
98
+ "shape": [
99
+ 480,
100
+ 640,
101
+ 3
102
+ ]
103
+ },
104
+ "observation.images.cam_right_wrist": {
105
+ "type": "VISUAL",
106
+ "shape": [
107
+ 480,
108
+ 640,
109
+ 3
110
+ ]
111
+ },
112
+ "observation.state": {
113
+ "type": "STATE",
114
+ "shape": [
115
+ 14
116
+ ]
117
+ }
118
+ },
119
+ "output_features": {
120
+ "action": {
121
+ "type": "ACTION",
122
+ "shape": [
123
+ 14
124
+ ]
125
+ }
126
+ },
127
+ "device": "cuda",
128
+ "use_amp": false,
129
+ "use_peft": false,
130
+ "push_to_hub": false,
131
+ "repo_id": null,
132
+ "private": null,
133
+ "tags": null,
134
+ "license": null,
135
+ "pretrained_path": null,
136
+ "pi05_init_path": "lerobot/pi05_base",
137
+ "paligemma_variant": "gemma_2b",
138
+ "action_expert_variant": "gemma_300m",
139
+ "pi05_precision": "bfloat16",
140
+ "chunk_size": 50,
141
+ "n_action_steps": 10,
142
+ "max_state_dim": 32,
143
+ "max_action_dim": 32,
144
+ "tokenizer_max_length": 200,
145
+ "image_resolution": [
146
+ 224,
147
+ 224
148
+ ],
149
+ "empty_cameras": 0,
150
+ "normalization_mapping": {
151
+ "VISUAL": "IDENTITY",
152
+ "STATE": "QUANTILES",
153
+ "ACTION": "QUANTILES"
154
+ },
155
+ "num_inference_steps": 10,
156
+ "time_sampling_beta_alpha": 1.5,
157
+ "time_sampling_beta_beta": 1.0,
158
+ "time_sampling_scale": 0.999,
159
+ "time_sampling_offset": 0.001,
160
+ "min_period": 0.004,
161
+ "max_period": 4.0,
162
+ "freeze_vlm": false,
163
+ "freeze_vision_encoder": false,
164
+ "gradient_checkpointing": true,
165
+ "offload_frozen_prefix_activations": true,
166
+ "phase_state_camera_index": 0,
167
+ "phase_state_frame_stride": 2,
168
+ "phase_state_vision_chunk_size": 32,
169
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_click_bell_vbe14/best_val_loss.pth",
170
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_click_bell_vbe14",
171
+ "phase_state_cache_episodes": 2,
172
+ "phase_state_max_steps": 1000,
173
+ "phase_state_checkpoint_strict": true,
174
+ "freeze_phase_state_encoder": true,
175
+ "phase_state_action_normalization": "QUANTILES",
176
+ "phase_state_action_dim": 14,
177
+ "phase_state_dim": 256,
178
+ "phase_state_layers": 4,
179
+ "phase_state_dropout": 0.1,
180
+ "phase_state_vision_pretrained": false,
181
+ "selector_heads": 8,
182
+ "evidence_dim": 256,
183
+ "use_relative_actions": false,
184
+ "relative_exclude_joints": [
185
+ "gripper"
186
+ ],
187
+ "action_feature_names": [
188
+ "motors"
189
+ ],
190
+ "optimizer_lr": 2.5e-05,
191
+ "evidence_optimizer_lr": 0.0001,
192
+ "action_optimizer_lr": 2.5e-05,
193
+ "vlm_optimizer_lr": 2.5e-05,
194
+ "vision_optimizer_lr": 2.5e-05,
195
+ "optimizer_betas": [
196
+ 0.9,
197
+ 0.95
198
+ ],
199
+ "optimizer_eps": 1e-08,
200
+ "optimizer_weight_decay": 0.01,
201
+ "optimizer_grad_clip_norm": 1.0,
202
+ "scheduler_warmup_steps": 33,
203
+ "scheduler_decay_steps": 1000,
204
+ "scheduler_decay_lr": 2.5e-06
205
+ },
206
+ "reward_model": null,
207
+ "output_dir": "/root/autodl-tmp/projections/SGVLA/outputs/train/ActionEvidenceVLA/robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20",
208
+ "job_name": "robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20",
209
+ "resume": false,
210
+ "seed": 42,
211
+ "cudnn_deterministic": false,
212
+ "num_workers": 4,
213
+ "batch_size": 16,
214
+ "fsdp": true,
215
+ "fsdp_save_training_state": false,
216
+ "prefetch_factor": 2,
217
+ "persistent_workers": true,
218
+ "steps": 1000,
219
+ "eval_freq": 0,
220
+ "log_freq": 100,
221
+ "tolerance_s": 0.0001,
222
+ "save_checkpoint": true,
223
+ "save_freq": 100,
224
+ "use_policy_training_preset": true,
225
+ "optimizer": {
226
+ "type": "adamw",
227
+ "lr": 2.5e-05,
228
+ "weight_decay": 0.01,
229
+ "grad_clip_norm": 1.0,
230
+ "betas": [
231
+ 0.9,
232
+ 0.95
233
+ ],
234
+ "eps": 1e-08
235
+ },
236
+ "scheduler": {
237
+ "type": "cosine_decay_with_warmup",
238
+ "num_warmup_steps": 33,
239
+ "num_decay_steps": 1000,
240
+ "peak_lr": 2.5e-05,
241
+ "decay_lr": 2.5e-06
242
+ },
243
+ "eval": {
244
+ "n_episodes": 50,
245
+ "batch_size": 50,
246
+ "use_async_envs": true,
247
+ "max_episodes_rendered": 10,
248
+ "reseed_policy_each_episode": false
249
+ },
250
+ "wandb": {
251
+ "enable": true,
252
+ "disable_artifact": true,
253
+ "project": "action-evidence-vla-pi05-phase",
254
+ "entity": null,
255
+ "notes": null,
256
+ "run_id": "vmkt7kml",
257
+ "mode": "online",
258
+ "add_tags": true
259
+ },
260
+ "peft": null,
261
+ "sample_weighting": null,
262
+ "rename_map": {},
263
+ "checkpoint_path": null
264
+ }
robotwin_click_bell_stage2_fsdp2_gb64_1k_model_only_official20/checkpoints/last/pretrained_model/config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "action_evidence_vla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.cam_high": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 480,
9
+ 640,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.cam_left_wrist": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 480,
17
+ 640,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.images.cam_right_wrist": {
22
+ "type": "VISUAL",
23
+ "shape": [
24
+ 480,
25
+ 640,
26
+ 3
27
+ ]
28
+ },
29
+ "observation.state": {
30
+ "type": "STATE",
31
+ "shape": [
32
+ 14
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": false,
48
+ "repo_id": null,
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": null,
53
+ "pi05_init_path": "lerobot/pi05_base",
54
+ "paligemma_variant": "gemma_2b",
55
+ "action_expert_variant": "gemma_300m",
56
+ "pi05_precision": "bfloat16",
57
+ "chunk_size": 50,
58
+ "n_action_steps": 10,
59
+ "max_state_dim": 32,
60
+ "max_action_dim": 32,
61
+ "tokenizer_max_length": 200,
62
+ "image_resolution": [
63
+ 224,
64
+ 224
65
+ ],
66
+ "empty_cameras": 0,
67
+ "normalization_mapping": {
68
+ "VISUAL": "IDENTITY",
69
+ "STATE": "QUANTILES",
70
+ "ACTION": "QUANTILES"
71
+ },
72
+ "num_inference_steps": 10,
73
+ "time_sampling_beta_alpha": 1.5,
74
+ "time_sampling_beta_beta": 1.0,
75
+ "time_sampling_scale": 0.999,
76
+ "time_sampling_offset": 0.001,
77
+ "min_period": 0.004,
78
+ "max_period": 4.0,
79
+ "freeze_vlm": false,
80
+ "freeze_vision_encoder": false,
81
+ "gradient_checkpointing": true,
82
+ "offload_frozen_prefix_activations": true,
83
+ "phase_state_camera_index": 0,
84
+ "phase_state_frame_stride": 2,
85
+ "phase_state_vision_chunk_size": 32,
86
+ "phase_state_checkpoint_path": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/robotwin_click_bell_vbe14/best_val_loss.pth",
87
+ "phase_state_cache_root": "/root/autodl-tmp/projections/SGVLA/outputs/ActionEvidenceVLA/phase_state_cache/robotwin_click_bell_vbe14",
88
+ "phase_state_cache_episodes": 2,
89
+ "phase_state_max_steps": 1000,
90
+ "phase_state_checkpoint_strict": true,
91
+ "freeze_phase_state_encoder": true,
92
+ "phase_state_action_normalization": "QUANTILES",
93
+ "phase_state_action_dim": 14,
94
+ "phase_state_dim": 256,
95
+ "phase_state_layers": 4,
96
+ "phase_state_dropout": 0.1,
97
+ "phase_state_vision_pretrained": false,
98
+ "selector_heads": 8,
99
+ "evidence_dim": 256,
100
+ "use_relative_actions": false,
101
+ "relative_exclude_joints": [
102
+ "gripper"
103
+ ],
104
+ "action_feature_names": [
105
+ "motors"
106
+ ],
107
+ "optimizer_lr": 2.5e-05,
108
+ "evidence_optimizer_lr": 0.0001,
109
+ "action_optimizer_lr": 2.5e-05,
110
+ "vlm_optimizer_lr": 2.5e-05,
111
+ "vision_optimizer_lr": 2.5e-05,
112
+ "optimizer_betas": [
113
+ 0.9,
114
+ 0.95
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 0.01,
118
+ "optimizer_grad_clip_norm": 1.0,
119
+ "scheduler_warmup_steps": 33,
120
+ "scheduler_decay_steps": 1000,
121
+ "scheduler_decay_lr": 2.5e-06
122
+ }