Whalswp commited on
Commit
0488647
·
verified ·
1 Parent(s): 45bb642

Back up BS128 baseline and three GuideVLA headgate 30K checkpoints

Browse files

Add the four selected complete checkpoints, including experiment configs and optimizer, RNG, scheduler and trainer state. Preserve the existing experiment hierarchy.

Files changed (40) hide show
  1. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/config.json +65 -0
  2. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml +197 -0
  3. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json +433 -0
  4. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors +3 -0
  5. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors +3 -0
  6. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json +0 -0
  7. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/optimizer.pt +3 -0
  8. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/rng_state.pth +3 -0
  9. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/scheduler.pt +3 -0
  10. GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/trainer_state.json +0 -0
  11. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/config.json +65 -0
  12. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml +214 -0
  13. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json +433 -0
  14. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors +3 -0
  15. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors +3 -0
  16. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json +0 -0
  17. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/optimizer.pt +3 -0
  18. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/rng_state.pth +3 -0
  19. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/scheduler.pt +3 -0
  20. GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/trainer_state.json +0 -0
  21. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/config.json +65 -0
  22. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml +201 -0
  23. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json +433 -0
  24. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors +3 -0
  25. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors +3 -0
  26. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json +0 -0
  27. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/optimizer.pt +3 -0
  28. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/rng_state.pth +3 -0
  29. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/scheduler.pt +3 -0
  30. GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/trainer_state.json +0 -0
  31. baseline_bs128_30k/checkpoint-30000/config.json +64 -0
  32. baseline_bs128_30k/checkpoint-30000/experiment_cfg/clvla_contract.yaml +64 -0
  33. baseline_bs128_30k/checkpoint-30000/experiment_cfg/metadata.json +433 -0
  34. baseline_bs128_30k/checkpoint-30000/model-00001-of-00002.safetensors +3 -0
  35. baseline_bs128_30k/checkpoint-30000/model-00002-of-00002.safetensors +3 -0
  36. baseline_bs128_30k/checkpoint-30000/model.safetensors.index.json +0 -0
  37. baseline_bs128_30k/checkpoint-30000/optimizer.pt +3 -0
  38. baseline_bs128_30k/checkpoint-30000/rng_state.pth +3 -0
  39. baseline_bs128_30k/checkpoint-30000/scheduler.pt +3 -0
  40. baseline_bs128_30k/checkpoint-30000/trainer_state.json +0 -0
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "clvla_lm_logits_to_keep": 1,
59
+ "compute_dtype": "bfloat16",
60
+ "hidden_size": 2048,
61
+ "model_dtype": "float32",
62
+ "model_type": "gr00t_n1_5",
63
+ "torch_dtype": "bfloat16",
64
+ "transformers_version": "4.51.3"
65
+ }
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml ADDED
@@ -0,0 +1,197 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: clvla_n15_guidevla_region_mass_v1
2
+ variant: guidevla_region_mass
3
+ relation_mode: null
4
+ future_token_mode: checkpoint_weight
5
+ vision_tuning: {}
6
+ auxiliary: {}
7
+ task_loss: {}
8
+ state_rkd: {}
9
+ state_subspace: {}
10
+ action_rkd: {}
11
+ relation_mismatch: {}
12
+ objective: {}
13
+ input_cache: {}
14
+ episode_shard: {}
15
+ n17_official_shard: {}
16
+ training:
17
+ dataset_path:
18
+ - /home/work/workspace/test/seonho/libero
19
+ output_dir: /home/work/workspace/test/seonho/groot_libero/train_ckpt/GuideVLA/headgate_gradient_bs128_30k_20261023
20
+ experiment_name: GuideVLA_HeadGate_gradient_BS128_30K_20261023
21
+ wandb_project: finetune-gr00t-n1d5-libero
22
+ data_config: examples.Libero.custom_data_config:LiberoDataConfig
23
+ base_model_path: nvidia/GR00T-N1.5-3B
24
+ embodiment_tag: new_embodiment
25
+ video_backend: torchcodec
26
+ batch_size: 128
27
+ max_steps: 30000
28
+ num_gpus: 1
29
+ save_steps: 10000
30
+ save_total_limit: 1
31
+ tune_llm: false
32
+ tune_visual: false
33
+ tune_projector: true
34
+ tune_diffusion_model: true
35
+ learning_rate: 0.0001
36
+ gradient_accumulation_steps: 1
37
+ weight_decay: 1.0e-05
38
+ warmup_ratio: 0.05
39
+ lora_rank: 0
40
+ lora_alpha: 16
41
+ lora_dropout: 0.1
42
+ lora_full_model: false
43
+ dataloader_num_workers: 16
44
+ dataloader_prefetch_factor: 4
45
+ report_to: wandb
46
+ resume: false
47
+ balance_dataset_weights: true
48
+ balance_trajectory_weights: true
49
+ state_dropout_prob: 0.0
50
+ dataset_backend: libero_pi_v2
51
+ n15_commit: 48581302b549cb22c455436cbbbe17dc4ddfde75
52
+ config_path: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/configs/gradient.yaml
53
+ config_sha256: acb984cecc0c4bc96d64a722bfa843232fc52bdeccdb97bea8dfc760ffc161d9
54
+ data_contract: libero_pi_v2
55
+ source_dataset:
56
+ dataset_backend: libero_pi_v2
57
+ dataset_path: /home/work/workspace/test/seonho/libero
58
+ episodes: 1693
59
+ frames: 273465
60
+ tasks: 40
61
+ fps: 10
62
+ info_sha256: 77c0da93c67960bac803a8ea25dabf380d752c26c1d91c54151121dd10e7d82f
63
+ stats_sha256: 1fe6c2d548a7998394488ae68bc2be46a51c8cd3798af6ab74fff79f8b20e293
64
+ action_horizon: 16
65
+ config_status: ready
66
+ guidevla:
67
+ enabled: true
68
+ teacher_source: public_stage_object_mask
69
+ annotation_root: /home/work/workspace/test/seonho/clvla/Isaac-GR00T-N1.5/clvla/cfg_libero/GuideVLA/annotation
70
+ frame_policy: same_frame_index
71
+ missing_annotation_policy: FM_only
72
+ extra_alignment_filter: false
73
+ views:
74
+ - front
75
+ - wrist
76
+ grid:
77
+ - 16
78
+ - 16
79
+ mask_crop_replay: true
80
+ mask_coverage_threshold: 0.5
81
+ student:
82
+ module: action_head.model.transformer_blocks
83
+ attention_module: attn1
84
+ block_indices_0based:
85
+ - 6
86
+ - 8
87
+ - 10
88
+ expected_block_count: 16
89
+ expected_interleave_self_attention: true
90
+ expected_cross_attention_block_indices_0based:
91
+ - 0
92
+ - 2
93
+ - 4
94
+ - 6
95
+ - 8
96
+ - 10
97
+ - 12
98
+ - 14
99
+ head_indices_0based:
100
+ - 0
101
+ - 1
102
+ - 2
103
+ - 3
104
+ - 4
105
+ - 5
106
+ - 6
107
+ - 7
108
+ expected_head_count: 32
109
+ expected_head_dimension: 48
110
+ head_selection: fixed_prefix_no_samplewise_selection
111
+ query_selection: last_action_horizon_tokens
112
+ expected_action_query_count: 16
113
+ exclude_state_query: true
114
+ exclude_future_queries: true
115
+ ignore_padded_action_queries: true
116
+ query_validity_source: action_mask_any_valid_dimension
117
+ image_key_indices: backbone_image_context_token_positions_preserved_after_processing
118
+ softmax_scope: native_encoder_context_keys
119
+ renormalize_image_attention_for_loss: false
120
+ attention_probability_stage: before_attention_dropout
121
+ stop_gradient: false
122
+ loss:
123
+ name: negative_log_retained_region_mass
124
+ target: external_binary_stage_object_mask
125
+ use_teacher_probability_amplitude: false
126
+ normalize_binary_region_mask: false
127
+ non_image_key_target_weight: 0.0
128
+ selected_head_reduction: mean_before_region_mass
129
+ query_reduction: mean_log_loss_over_valid_action_queries
130
+ layer_reduction: mean
131
+ formula: mean_layer_valid_query(-log(clamp_min(sum_patch(mean_selected_head(P_image)*keep_mask),epsilon)))
132
+ epsilon: 1.0e-06
133
+ empty_or_zero_mass_teacher_policy: exclude_aux_keep_native_fm
134
+ flow_matching_weight: 1.0
135
+ lambda_start: 0.001
136
+ lambda_end: 0.001
137
+ lambda_schedule: constant
138
+ lambda_warmup_ratio: 0.0
139
+ lambda_warmup_optimizer_steps: 0
140
+ schedule_step_source: optimizer_global_step
141
+ gradient_surgery: none
142
+ guidevla_inference_behavior: stock_architecture_no_annotation_or_teacher_hooks
143
+ source_module_sha256:
144
+ clvla/data/guidevla_annotation.py: 8886d09bcd7bf5e8978372403c7e4a63d141aa4337e1e430afbe5eb442229467
145
+ clvla/patches/guidevla_region_mass_patch.py: 60d1a163eca3c23f226bd592263da357880d3132a0ef1fab03a78654cd2002f3
146
+ clvla/patches/aif_region_mass_patch.py: 177c7bbfb3c3135510b3fff9ccdea895e93f604c6e6f89d2d6ef245ab7a31a5d
147
+ clvla/ops/guidevla_20261018/bootstrap.py: c4f7f798156b3b821f4b2a6eb48f78072f3e6e0e6e6d751c1dfd99f20ba950ef
148
+ clvla/ops/guidevla_20261018/launch.py: 6ea6d33bed21141db8d05b2940106094b14cbad760e2ef22078400045f08649f
149
+ annotation_manifest_sha256: a0c2810a0bbb321acd5a43e7d5dee3d50a1f10460bee71a4446b0af55769d948
150
+ annotation_mapping_sha256: 715d209e1d331f95c4b94c431a40374c12c09e6d680b2c4fd4c042bae9111b3f
151
+ distribution:
152
+ method: gradient
153
+ auxiliary_start_step: 0
154
+ auxiliary_ramp_steps: 0
155
+ weight_policy: constant
156
+ prior_floor: 0.9
157
+ target_kl_budget: 0.01
158
+ canonical_frames_per_annotated_demo: 16
159
+ original_FM_region_on_all_frames: true
160
+ inference: unchanged; no teacher, labels or additional head
161
+ weight: 0.1
162
+ allocator_limit_gib: 72
163
+ importance_source: current_training_policy_online
164
+ supervision_prior: detached_current_native_attention_same_forward
165
+ intervention: supervised_head_action_query_post_softmax_removal
166
+ raw_provenance_root: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/raw/gradient
167
+ gate_scope:
168
+ layers:
169
+ - 6
170
+ - 8
171
+ - 10
172
+ heads:
173
+ - 0
174
+ - 1
175
+ - 2
176
+ - 3
177
+ - 4
178
+ - 5
179
+ - 6
180
+ - 7
181
+ queries: last16_valid_action
182
+ keys: existing_ROI_image_keys
183
+ online_chunk_size: 4
184
+ online_fm_noise_draws: 2
185
+ online_fm_times:
186
+ - 0.3
187
+ - 0.7
188
+ online_target_stop_gradient: true
189
+ raw_snapshot_steps: 200
190
+ online_selection: all selected items; no sample cap
191
+ additional_VLM_forwards: 0
192
+ distribution_source_sha256:
193
+ /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/online_runtime.py: c1514aa243ee54014159ff6d86ea4e77e69a93a8058b897a13ab059a761c214c
194
+ /home/work/workspace/test/seonho/analysis/guidevla_head_gate_20261023/head_gate.py: 45084629aea378610a917f31cf777c9b2436b192fd8cde3c34c6dfe78cc83a45
195
+ /home/work/workspace/test/seonho/analysis/guidevla_head_gate_20261023/image_reference.py: 890c278cc885f4e5200cc24ed44033eb11330e2b43655bf59fd54c7465ca52fb
196
+ distribution_inference: unchanged native model; head gates are training-only importance
197
+ interventions
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "x": {
6
+ "max": [
7
+ 0.21031762659549713
8
+ ],
9
+ "min": [
10
+ -0.4828203022480011
11
+ ],
12
+ "mean": [
13
+ -0.04651879519224167
14
+ ],
15
+ "std": [
16
+ 0.10494378954172134
17
+ ],
18
+ "q01": [
19
+ -0.39912487387657164
20
+ ],
21
+ "q99": [
22
+ 0.13556506335735302
23
+ ]
24
+ },
25
+ "y": {
26
+ "max": [
27
+ 0.39128610491752625
28
+ ],
29
+ "min": [
30
+ -0.3255046010017395
31
+ ],
32
+ "mean": [
33
+ 0.03440921753644943
34
+ ],
35
+ "std": [
36
+ 0.15176637470722198
37
+ ],
38
+ "q01": [
39
+ -0.26883512616157534
40
+ ],
41
+ "q99": [
42
+ 0.33566484451293943
43
+ ]
44
+ },
45
+ "z": {
46
+ "max": [
47
+ 1.3660105466842651
48
+ ],
49
+ "min": [
50
+ 0.008128180168569088
51
+ ],
52
+ "mean": [
53
+ 0.7645525336265564
54
+ ],
55
+ "std": [
56
+ 0.3785160183906555
57
+ ],
58
+ "q01": [
59
+ 0.03826696485280991
60
+ ],
61
+ "q99": [
62
+ 1.2706659078598022
63
+ ]
64
+ },
65
+ "roll": {
66
+ "max": [
67
+ 3.6714255809783936
68
+ ],
69
+ "min": [
70
+ 0.35277295112609863
71
+ ],
72
+ "mean": [
73
+ 2.972202777862549
74
+ ],
75
+ "std": [
76
+ 0.34427398443222046
77
+ ],
78
+ "q01": [
79
+ 1.5089580535888671
80
+ ],
81
+ "q99": [
82
+ 3.2773460674285886
83
+ ]
84
+ },
85
+ "pitch": {
86
+ "max": [
87
+ 3.560650587081909
88
+ ],
89
+ "min": [
90
+ -3.641430377960205
91
+ ],
92
+ "mean": [
93
+ -0.22047005593776703
94
+ ],
95
+ "std": [
96
+ 0.9069469571113586
97
+ ],
98
+ "q01": [
99
+ -2.719791135787964
100
+ ],
101
+ "q99": [
102
+ 2.406109695434569
103
+ ]
104
+ },
105
+ "yaw": {
106
+ "max": [
107
+ 1.386339545249939
108
+ ],
109
+ "min": [
110
+ -1.842738389968872
111
+ ],
112
+ "mean": [
113
+ -0.1255796253681183
114
+ ],
115
+ "std": [
116
+ 0.3253920078277588
117
+ ],
118
+ "q01": [
119
+ -1.08050856590271
120
+ ],
121
+ "q99": [
122
+ 0.59776972055435
123
+ ]
124
+ },
125
+ "gripper": {
126
+ "max": [
127
+ 0.04233968257904053,
128
+ 0.0013633022317662835
129
+ ],
130
+ "min": [
131
+ -0.0013586411951109767,
132
+ -0.042040832340717316
133
+ ],
134
+ "mean": [
135
+ 0.026914266869425774,
136
+ -0.02719070389866829
137
+ ],
138
+ "std": [
139
+ 0.014175914227962494,
140
+ 0.014058894477784634
141
+ ],
142
+ "q01": [
143
+ 0.0017423711018636821,
144
+ -0.04002561450004578
145
+ ],
146
+ "q99": [
147
+ 0.04031316146254539,
148
+ -0.0017781084170565016
149
+ ]
150
+ }
151
+ },
152
+ "action": {
153
+ "x": {
154
+ "max": [
155
+ 0.9375
156
+ ],
157
+ "min": [
158
+ -0.9375
159
+ ],
160
+ "mean": [
161
+ 0.06278137117624283
162
+ ],
163
+ "std": [
164
+ 0.3355240225791931
165
+ ],
166
+ "q01": [
167
+ -0.7044642567634583
168
+ ],
169
+ "q99": [
170
+ 0.9375
171
+ ]
172
+ },
173
+ "y": {
174
+ "max": [
175
+ 0.9375
176
+ ],
177
+ "min": [
178
+ -0.9375
179
+ ],
180
+ "mean": [
181
+ 0.0868409126996994
182
+ ],
183
+ "std": [
184
+ 0.3784470558166504
185
+ ],
186
+ "q01": [
187
+ -0.8008928298950195
188
+ ],
189
+ "q99": [
190
+ 0.8678571581840515
191
+ ]
192
+ },
193
+ "z": {
194
+ "max": [
195
+ 0.9375
196
+ ],
197
+ "min": [
198
+ -0.9375
199
+ ],
200
+ "mean": [
201
+ -0.09037282317876816
202
+ ],
203
+ "std": [
204
+ 0.44472837448120117
205
+ ],
206
+ "q01": [
207
+ -0.9375
208
+ ],
209
+ "q99": [
210
+ 0.9375
211
+ ]
212
+ },
213
+ "roll": {
214
+ "max": [
215
+ 0.3557142913341522
216
+ ],
217
+ "min": [
218
+ -0.2582142949104309
219
+ ],
220
+ "mean": [
221
+ 0.0005407406715676188
222
+ ],
223
+ "std": [
224
+ 0.03924351558089256
225
+ ],
226
+ "q01": [
227
+ -0.11464285850524902
228
+ ],
229
+ "q99": [
230
+ 0.13178572058677673
231
+ ]
232
+ },
233
+ "pitch": {
234
+ "max": [
235
+ 0.375
236
+ ],
237
+ "min": [
238
+ -0.375
239
+ ],
240
+ "mean": [
241
+ 0.005643361248075962
242
+ ],
243
+ "std": [
244
+ 0.06339313089847565
245
+ ],
246
+ "q01": [
247
+ -0.1639285683631897
248
+ ],
249
+ "q99": [
250
+ 0.19285714626312256
251
+ ]
252
+ },
253
+ "yaw": {
254
+ "max": [
255
+ 0.375
256
+ ],
257
+ "min": [
258
+ -0.3675000071525574
259
+ ],
260
+ "mean": [
261
+ -0.005229088477790356
262
+ ],
263
+ "std": [
264
+ 0.07797032594680786
265
+ ],
266
+ "q01": [
267
+ -0.2239285707473755
268
+ ],
269
+ "q99": [
270
+ 0.335357129573822
271
+ ]
272
+ },
273
+ "gripper": {
274
+ "max": [
275
+ 1.0
276
+ ],
277
+ "min": [
278
+ -1.0
279
+ ],
280
+ "mean": [
281
+ -0.04964079707860947
282
+ ],
283
+ "std": [
284
+ 0.9987710118293762
285
+ ],
286
+ "q01": [
287
+ -1.0
288
+ ],
289
+ "q99": [
290
+ 1.0
291
+ ]
292
+ }
293
+ }
294
+ },
295
+ "modalities": {
296
+ "video": {
297
+ "image": {
298
+ "resolution": [
299
+ 256,
300
+ 256
301
+ ],
302
+ "channels": 3,
303
+ "fps": 10.0
304
+ },
305
+ "wrist_image": {
306
+ "resolution": [
307
+ 256,
308
+ 256
309
+ ],
310
+ "channels": 3,
311
+ "fps": 10.0
312
+ }
313
+ },
314
+ "state": {
315
+ "x": {
316
+ "absolute": true,
317
+ "rotation_type": null,
318
+ "shape": [
319
+ 1
320
+ ],
321
+ "continuous": true
322
+ },
323
+ "y": {
324
+ "absolute": true,
325
+ "rotation_type": null,
326
+ "shape": [
327
+ 1
328
+ ],
329
+ "continuous": true
330
+ },
331
+ "z": {
332
+ "absolute": true,
333
+ "rotation_type": null,
334
+ "shape": [
335
+ 1
336
+ ],
337
+ "continuous": true
338
+ },
339
+ "roll": {
340
+ "absolute": true,
341
+ "rotation_type": null,
342
+ "shape": [
343
+ 1
344
+ ],
345
+ "continuous": true
346
+ },
347
+ "pitch": {
348
+ "absolute": true,
349
+ "rotation_type": null,
350
+ "shape": [
351
+ 1
352
+ ],
353
+ "continuous": true
354
+ },
355
+ "yaw": {
356
+ "absolute": true,
357
+ "rotation_type": null,
358
+ "shape": [
359
+ 1
360
+ ],
361
+ "continuous": true
362
+ },
363
+ "gripper": {
364
+ "absolute": true,
365
+ "rotation_type": null,
366
+ "shape": [
367
+ 2
368
+ ],
369
+ "continuous": true
370
+ }
371
+ },
372
+ "action": {
373
+ "x": {
374
+ "absolute": true,
375
+ "rotation_type": null,
376
+ "shape": [
377
+ 1
378
+ ],
379
+ "continuous": true
380
+ },
381
+ "y": {
382
+ "absolute": true,
383
+ "rotation_type": null,
384
+ "shape": [
385
+ 1
386
+ ],
387
+ "continuous": true
388
+ },
389
+ "z": {
390
+ "absolute": true,
391
+ "rotation_type": null,
392
+ "shape": [
393
+ 1
394
+ ],
395
+ "continuous": true
396
+ },
397
+ "roll": {
398
+ "absolute": true,
399
+ "rotation_type": null,
400
+ "shape": [
401
+ 1
402
+ ],
403
+ "continuous": true
404
+ },
405
+ "pitch": {
406
+ "absolute": true,
407
+ "rotation_type": null,
408
+ "shape": [
409
+ 1
410
+ ],
411
+ "continuous": true
412
+ },
413
+ "yaw": {
414
+ "absolute": true,
415
+ "rotation_type": null,
416
+ "shape": [
417
+ 1
418
+ ],
419
+ "continuous": true
420
+ },
421
+ "gripper": {
422
+ "absolute": true,
423
+ "rotation_type": null,
424
+ "shape": [
425
+ 1
426
+ ],
427
+ "continuous": true
428
+ }
429
+ }
430
+ },
431
+ "embodiment_tag": "new_embodiment"
432
+ }
433
+ }
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d17e089ed25d59bd9896352f18262b1af67365b333828c3b98e9258736c8a6da
3
+ size 4999367032
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2165f23f618773f0b0ff76459843ae9fb78b4bef94a8ac954997a35b4462e37e
3
+ size 2586705312
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d92c733179379574621324bfa12f24c83c162121364b0f5497f52317e71d3d4
3
+ size 8550720062
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb56d629866c90996018f1f830acd404ef91d919d6f113d46f65a9fdb983c449
3
+ size 14244
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe8c620e9a7808e10f2bc64fa81fe5b1efb6dcc2acb3596ed32f58119a5368c1
3
+ size 1064
GuideVLA/headgate_gradient_bs128_30k_20261023/checkpoint-30000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "clvla_lm_logits_to_keep": 1,
59
+ "compute_dtype": "bfloat16",
60
+ "hidden_size": 2048,
61
+ "model_dtype": "float32",
62
+ "model_type": "gr00t_n1_5",
63
+ "torch_dtype": "bfloat16",
64
+ "transformers_version": "4.51.3"
65
+ }
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml ADDED
@@ -0,0 +1,214 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: clvla_n15_guidevla_region_mass_v1
2
+ variant: guidevla_region_mass
3
+ relation_mode: null
4
+ future_token_mode: checkpoint_weight
5
+ vision_tuning: {}
6
+ auxiliary: {}
7
+ task_loss: {}
8
+ state_rkd: {}
9
+ state_subspace: {}
10
+ action_rkd: {}
11
+ relation_mismatch: {}
12
+ objective: {}
13
+ input_cache: {}
14
+ episode_shard: {}
15
+ n17_official_shard: {}
16
+ training:
17
+ dataset_path:
18
+ - /home/work/workspace/test/seonho/libero
19
+ output_dir: /home/work/workspace/test/seonho/groot_libero/train_ckpt/GuideVLA/headgate_rise16_bs128_30k_20261023
20
+ experiment_name: GuideVLA_HeadGate_RISE16_BS128_30K_20261023
21
+ wandb_project: finetune-gr00t-n1d5-libero
22
+ data_config: examples.Libero.custom_data_config:LiberoDataConfig
23
+ base_model_path: nvidia/GR00T-N1.5-3B
24
+ embodiment_tag: new_embodiment
25
+ video_backend: torchcodec
26
+ batch_size: 128
27
+ max_steps: 30000
28
+ num_gpus: 1
29
+ save_steps: 10000
30
+ save_total_limit: 1
31
+ tune_llm: false
32
+ tune_visual: false
33
+ tune_projector: true
34
+ tune_diffusion_model: true
35
+ learning_rate: 0.0001
36
+ gradient_accumulation_steps: 1
37
+ weight_decay: 1.0e-05
38
+ warmup_ratio: 0.05
39
+ lora_rank: 0
40
+ lora_alpha: 16
41
+ lora_dropout: 0.1
42
+ lora_full_model: false
43
+ dataloader_num_workers: 16
44
+ dataloader_prefetch_factor: 4
45
+ report_to: wandb
46
+ resume: false
47
+ balance_dataset_weights: true
48
+ balance_trajectory_weights: true
49
+ state_dropout_prob: 0.0
50
+ dataset_backend: libero_pi_v2
51
+ n15_commit: 48581302b549cb22c455436cbbbe17dc4ddfde75
52
+ config_path: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/configs/rise.yaml
53
+ config_sha256: ce38b6552ba6f59fbc1f1ea303aae1d0c7d709ccf6549d6b9894cb8a66ff7eb1
54
+ data_contract: libero_pi_v2
55
+ source_dataset:
56
+ dataset_backend: libero_pi_v2
57
+ dataset_path: /home/work/workspace/test/seonho/libero
58
+ episodes: 1693
59
+ frames: 273465
60
+ tasks: 40
61
+ fps: 10
62
+ info_sha256: 77c0da93c67960bac803a8ea25dabf380d752c26c1d91c54151121dd10e7d82f
63
+ stats_sha256: 1fe6c2d548a7998394488ae68bc2be46a51c8cd3798af6ab74fff79f8b20e293
64
+ action_horizon: 16
65
+ config_status: ready
66
+ guidevla:
67
+ enabled: true
68
+ teacher_source: public_stage_object_mask
69
+ annotation_root: /home/work/workspace/test/seonho/clvla/Isaac-GR00T-N1.5/clvla/cfg_libero/GuideVLA/annotation
70
+ frame_policy: same_frame_index
71
+ missing_annotation_policy: FM_only
72
+ extra_alignment_filter: false
73
+ views:
74
+ - front
75
+ - wrist
76
+ grid:
77
+ - 16
78
+ - 16
79
+ mask_crop_replay: true
80
+ mask_coverage_threshold: 0.5
81
+ student:
82
+ module: action_head.model.transformer_blocks
83
+ attention_module: attn1
84
+ block_indices_0based:
85
+ - 6
86
+ - 8
87
+ - 10
88
+ expected_block_count: 16
89
+ expected_interleave_self_attention: true
90
+ expected_cross_attention_block_indices_0based:
91
+ - 0
92
+ - 2
93
+ - 4
94
+ - 6
95
+ - 8
96
+ - 10
97
+ - 12
98
+ - 14
99
+ head_indices_0based:
100
+ - 0
101
+ - 1
102
+ - 2
103
+ - 3
104
+ - 4
105
+ - 5
106
+ - 6
107
+ - 7
108
+ expected_head_count: 32
109
+ expected_head_dimension: 48
110
+ head_selection: fixed_prefix_no_samplewise_selection
111
+ query_selection: last_action_horizon_tokens
112
+ expected_action_query_count: 16
113
+ exclude_state_query: true
114
+ exclude_future_queries: true
115
+ ignore_padded_action_queries: true
116
+ query_validity_source: action_mask_any_valid_dimension
117
+ image_key_indices: backbone_image_context_token_positions_preserved_after_processing
118
+ softmax_scope: native_encoder_context_keys
119
+ renormalize_image_attention_for_loss: false
120
+ attention_probability_stage: before_attention_dropout
121
+ stop_gradient: false
122
+ loss:
123
+ name: negative_log_retained_region_mass
124
+ target: external_binary_stage_object_mask
125
+ use_teacher_probability_amplitude: false
126
+ normalize_binary_region_mask: false
127
+ non_image_key_target_weight: 0.0
128
+ selected_head_reduction: mean_before_region_mass
129
+ query_reduction: mean_log_loss_over_valid_action_queries
130
+ layer_reduction: mean
131
+ formula: mean_layer_valid_query(-log(clamp_min(sum_patch(mean_selected_head(P_image)*keep_mask),epsilon)))
132
+ epsilon: 1.0e-06
133
+ empty_or_zero_mass_teacher_policy: exclude_aux_keep_native_fm
134
+ flow_matching_weight: 1.0
135
+ lambda_start: 0.001
136
+ lambda_end: 0.001
137
+ lambda_schedule: constant
138
+ lambda_warmup_ratio: 0.0
139
+ lambda_warmup_optimizer_steps: 0
140
+ schedule_step_source: optimizer_global_step
141
+ gradient_surgery: none
142
+ guidevla_inference_behavior: stock_architecture_no_annotation_or_teacher_hooks
143
+ source_module_sha256:
144
+ clvla/data/guidevla_annotation.py: 8886d09bcd7bf5e8978372403c7e4a63d141aa4337e1e430afbe5eb442229467
145
+ clvla/patches/guidevla_region_mass_patch.py: 60d1a163eca3c23f226bd592263da357880d3132a0ef1fab03a78654cd2002f3
146
+ clvla/patches/aif_region_mass_patch.py: 177c7bbfb3c3135510b3fff9ccdea895e93f604c6e6f89d2d6ef245ab7a31a5d
147
+ clvla/ops/guidevla_20261018/bootstrap.py: c4f7f798156b3b821f4b2a6eb48f78072f3e6e0e6e6d751c1dfd99f20ba950ef
148
+ clvla/ops/guidevla_20261018/launch.py: 6ea6d33bed21141db8d05b2940106094b14cbad760e2ef22078400045f08649f
149
+ annotation_manifest_sha256: a0c2810a0bbb321acd5a43e7d5dee3d50a1f10460bee71a4446b0af55769d948
150
+ annotation_mapping_sha256: 715d209e1d331f95c4b94c431a40374c12c09e6d680b2c4fd4c042bae9111b3f
151
+ distribution:
152
+ method: rise
153
+ cache_root: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/labels/rise
154
+ teacher_checkpoint: /home/work/workspace/test/seonho/groot_libero/train_ckpt/GuideVLA/guidevla_object_bs128_30k/checkpoint-30000
155
+ auxiliary_start_step: 0
156
+ auxiliary_ramp_steps: 0
157
+ weight_policy: constant
158
+ prior_floor: 0.9
159
+ target_kl_budget: 0.01
160
+ rise_conditions: 16
161
+ sparse_regions_per_view: 2
162
+ sparse_block_shape:
163
+ - 2
164
+ - 2
165
+ sparse_pair_banks: 2
166
+ canonical_frames_per_annotated_demo: 16
167
+ original_FM_region_on_all_frames: true
168
+ cache_miss_policy: keep_original_FM_and_region
169
+ cache_geometry: same VideoCrop RNG replay
170
+ inference: unchanged; no teacher, labels or additional head
171
+ weight: 0.1
172
+ allocator_limit_gib: 68
173
+ importance_source: fixed_guidevla_teacher_head_path_cache
174
+ supervision_prior: cached_fixed_teacher_attention
175
+ intervention: supervised_head_action_query_post_softmax_removal
176
+ raw_provenance_root: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/raw/rise
177
+ gate_scope:
178
+ layers:
179
+ - 6
180
+ - 8
181
+ - 10
182
+ heads:
183
+ - 0
184
+ - 1
185
+ - 2
186
+ - 3
187
+ - 4
188
+ - 5
189
+ - 6
190
+ - 7
191
+ queries: last16_valid_action
192
+ keys: existing_ROI_image_keys
193
+ distribution_inference: unchanged native model; distribution supervision is training-only
194
+ distribution_source_sha256:
195
+ /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/cached_runtime.py: e95878147733fe41d64c8df9e208f565e7d2b0b2b41c64cbd075170866a1ade6
196
+ distribution_cache_ready:
197
+ complete: true
198
+ mode: rise
199
+ frames: 26032
200
+ demos: 1627
201
+ plan_sha256: 3f8cefbc5e5e4cb3fb7b79ce21f12e5a178e9c7decca450c311bbeab453b604f
202
+ teacher_checkpoint: /home/work/workspace/test/seonho/groot_libero/train_ckpt/GuideVLA/guidevla_object_bs128_30k/checkpoint-30000
203
+ intervention: supervised_head_action_query_post_softmax_removal
204
+ rise_conditions: 16
205
+ native_VLM_batches: 3254
206
+ raw_root: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/raw/rise
207
+ raw_attention_dtype: float32
208
+ source_teacher_unchanged: true
209
+ headgate_training_source_sha256:
210
+ prepare_cache.py: a98b211c69f713549f460ae64e05fcc193244e953b66c5f8e4335a4b5347700a
211
+ data_support.py: 5e83e0595a2eff294d6e4cb852ede2496a9fb9479730987ca686c85e24070c41
212
+ cached_runtime.py: e95878147733fe41d64c8df9e208f565e7d2b0b2b41c64cbd075170866a1ade6
213
+ launch_rise.py: b4efe52490d895a6cc16c3de16dd9c5faff88d48ea68f04ef13c545c0217fbb9
214
+ headgate_source_sha256: 45084629aea378610a917f31cf777c9b2436b192fd8cde3c34c6dfe78cc83a45
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "x": {
6
+ "max": [
7
+ 0.21031762659549713
8
+ ],
9
+ "min": [
10
+ -0.4828203022480011
11
+ ],
12
+ "mean": [
13
+ -0.04651879519224167
14
+ ],
15
+ "std": [
16
+ 0.10494378954172134
17
+ ],
18
+ "q01": [
19
+ -0.39912487387657164
20
+ ],
21
+ "q99": [
22
+ 0.13556506335735302
23
+ ]
24
+ },
25
+ "y": {
26
+ "max": [
27
+ 0.39128610491752625
28
+ ],
29
+ "min": [
30
+ -0.3255046010017395
31
+ ],
32
+ "mean": [
33
+ 0.03440921753644943
34
+ ],
35
+ "std": [
36
+ 0.15176637470722198
37
+ ],
38
+ "q01": [
39
+ -0.26883512616157534
40
+ ],
41
+ "q99": [
42
+ 0.33566484451293943
43
+ ]
44
+ },
45
+ "z": {
46
+ "max": [
47
+ 1.3660105466842651
48
+ ],
49
+ "min": [
50
+ 0.008128180168569088
51
+ ],
52
+ "mean": [
53
+ 0.7645525336265564
54
+ ],
55
+ "std": [
56
+ 0.3785160183906555
57
+ ],
58
+ "q01": [
59
+ 0.03826696485280991
60
+ ],
61
+ "q99": [
62
+ 1.2706659078598022
63
+ ]
64
+ },
65
+ "roll": {
66
+ "max": [
67
+ 3.6714255809783936
68
+ ],
69
+ "min": [
70
+ 0.35277295112609863
71
+ ],
72
+ "mean": [
73
+ 2.972202777862549
74
+ ],
75
+ "std": [
76
+ 0.34427398443222046
77
+ ],
78
+ "q01": [
79
+ 1.5089580535888671
80
+ ],
81
+ "q99": [
82
+ 3.2773460674285886
83
+ ]
84
+ },
85
+ "pitch": {
86
+ "max": [
87
+ 3.560650587081909
88
+ ],
89
+ "min": [
90
+ -3.641430377960205
91
+ ],
92
+ "mean": [
93
+ -0.22047005593776703
94
+ ],
95
+ "std": [
96
+ 0.9069469571113586
97
+ ],
98
+ "q01": [
99
+ -2.719791135787964
100
+ ],
101
+ "q99": [
102
+ 2.406109695434569
103
+ ]
104
+ },
105
+ "yaw": {
106
+ "max": [
107
+ 1.386339545249939
108
+ ],
109
+ "min": [
110
+ -1.842738389968872
111
+ ],
112
+ "mean": [
113
+ -0.1255796253681183
114
+ ],
115
+ "std": [
116
+ 0.3253920078277588
117
+ ],
118
+ "q01": [
119
+ -1.08050856590271
120
+ ],
121
+ "q99": [
122
+ 0.59776972055435
123
+ ]
124
+ },
125
+ "gripper": {
126
+ "max": [
127
+ 0.04233968257904053,
128
+ 0.0013633022317662835
129
+ ],
130
+ "min": [
131
+ -0.0013586411951109767,
132
+ -0.042040832340717316
133
+ ],
134
+ "mean": [
135
+ 0.026914266869425774,
136
+ -0.02719070389866829
137
+ ],
138
+ "std": [
139
+ 0.014175914227962494,
140
+ 0.014058894477784634
141
+ ],
142
+ "q01": [
143
+ 0.0017423711018636821,
144
+ -0.04002561450004578
145
+ ],
146
+ "q99": [
147
+ 0.04031316146254539,
148
+ -0.0017781084170565016
149
+ ]
150
+ }
151
+ },
152
+ "action": {
153
+ "x": {
154
+ "max": [
155
+ 0.9375
156
+ ],
157
+ "min": [
158
+ -0.9375
159
+ ],
160
+ "mean": [
161
+ 0.06278137117624283
162
+ ],
163
+ "std": [
164
+ 0.3355240225791931
165
+ ],
166
+ "q01": [
167
+ -0.7044642567634583
168
+ ],
169
+ "q99": [
170
+ 0.9375
171
+ ]
172
+ },
173
+ "y": {
174
+ "max": [
175
+ 0.9375
176
+ ],
177
+ "min": [
178
+ -0.9375
179
+ ],
180
+ "mean": [
181
+ 0.0868409126996994
182
+ ],
183
+ "std": [
184
+ 0.3784470558166504
185
+ ],
186
+ "q01": [
187
+ -0.8008928298950195
188
+ ],
189
+ "q99": [
190
+ 0.8678571581840515
191
+ ]
192
+ },
193
+ "z": {
194
+ "max": [
195
+ 0.9375
196
+ ],
197
+ "min": [
198
+ -0.9375
199
+ ],
200
+ "mean": [
201
+ -0.09037282317876816
202
+ ],
203
+ "std": [
204
+ 0.44472837448120117
205
+ ],
206
+ "q01": [
207
+ -0.9375
208
+ ],
209
+ "q99": [
210
+ 0.9375
211
+ ]
212
+ },
213
+ "roll": {
214
+ "max": [
215
+ 0.3557142913341522
216
+ ],
217
+ "min": [
218
+ -0.2582142949104309
219
+ ],
220
+ "mean": [
221
+ 0.0005407406715676188
222
+ ],
223
+ "std": [
224
+ 0.03924351558089256
225
+ ],
226
+ "q01": [
227
+ -0.11464285850524902
228
+ ],
229
+ "q99": [
230
+ 0.13178572058677673
231
+ ]
232
+ },
233
+ "pitch": {
234
+ "max": [
235
+ 0.375
236
+ ],
237
+ "min": [
238
+ -0.375
239
+ ],
240
+ "mean": [
241
+ 0.005643361248075962
242
+ ],
243
+ "std": [
244
+ 0.06339313089847565
245
+ ],
246
+ "q01": [
247
+ -0.1639285683631897
248
+ ],
249
+ "q99": [
250
+ 0.19285714626312256
251
+ ]
252
+ },
253
+ "yaw": {
254
+ "max": [
255
+ 0.375
256
+ ],
257
+ "min": [
258
+ -0.3675000071525574
259
+ ],
260
+ "mean": [
261
+ -0.005229088477790356
262
+ ],
263
+ "std": [
264
+ 0.07797032594680786
265
+ ],
266
+ "q01": [
267
+ -0.2239285707473755
268
+ ],
269
+ "q99": [
270
+ 0.335357129573822
271
+ ]
272
+ },
273
+ "gripper": {
274
+ "max": [
275
+ 1.0
276
+ ],
277
+ "min": [
278
+ -1.0
279
+ ],
280
+ "mean": [
281
+ -0.04964079707860947
282
+ ],
283
+ "std": [
284
+ 0.9987710118293762
285
+ ],
286
+ "q01": [
287
+ -1.0
288
+ ],
289
+ "q99": [
290
+ 1.0
291
+ ]
292
+ }
293
+ }
294
+ },
295
+ "modalities": {
296
+ "video": {
297
+ "image": {
298
+ "resolution": [
299
+ 256,
300
+ 256
301
+ ],
302
+ "channels": 3,
303
+ "fps": 10.0
304
+ },
305
+ "wrist_image": {
306
+ "resolution": [
307
+ 256,
308
+ 256
309
+ ],
310
+ "channels": 3,
311
+ "fps": 10.0
312
+ }
313
+ },
314
+ "state": {
315
+ "x": {
316
+ "absolute": true,
317
+ "rotation_type": null,
318
+ "shape": [
319
+ 1
320
+ ],
321
+ "continuous": true
322
+ },
323
+ "y": {
324
+ "absolute": true,
325
+ "rotation_type": null,
326
+ "shape": [
327
+ 1
328
+ ],
329
+ "continuous": true
330
+ },
331
+ "z": {
332
+ "absolute": true,
333
+ "rotation_type": null,
334
+ "shape": [
335
+ 1
336
+ ],
337
+ "continuous": true
338
+ },
339
+ "roll": {
340
+ "absolute": true,
341
+ "rotation_type": null,
342
+ "shape": [
343
+ 1
344
+ ],
345
+ "continuous": true
346
+ },
347
+ "pitch": {
348
+ "absolute": true,
349
+ "rotation_type": null,
350
+ "shape": [
351
+ 1
352
+ ],
353
+ "continuous": true
354
+ },
355
+ "yaw": {
356
+ "absolute": true,
357
+ "rotation_type": null,
358
+ "shape": [
359
+ 1
360
+ ],
361
+ "continuous": true
362
+ },
363
+ "gripper": {
364
+ "absolute": true,
365
+ "rotation_type": null,
366
+ "shape": [
367
+ 2
368
+ ],
369
+ "continuous": true
370
+ }
371
+ },
372
+ "action": {
373
+ "x": {
374
+ "absolute": true,
375
+ "rotation_type": null,
376
+ "shape": [
377
+ 1
378
+ ],
379
+ "continuous": true
380
+ },
381
+ "y": {
382
+ "absolute": true,
383
+ "rotation_type": null,
384
+ "shape": [
385
+ 1
386
+ ],
387
+ "continuous": true
388
+ },
389
+ "z": {
390
+ "absolute": true,
391
+ "rotation_type": null,
392
+ "shape": [
393
+ 1
394
+ ],
395
+ "continuous": true
396
+ },
397
+ "roll": {
398
+ "absolute": true,
399
+ "rotation_type": null,
400
+ "shape": [
401
+ 1
402
+ ],
403
+ "continuous": true
404
+ },
405
+ "pitch": {
406
+ "absolute": true,
407
+ "rotation_type": null,
408
+ "shape": [
409
+ 1
410
+ ],
411
+ "continuous": true
412
+ },
413
+ "yaw": {
414
+ "absolute": true,
415
+ "rotation_type": null,
416
+ "shape": [
417
+ 1
418
+ ],
419
+ "continuous": true
420
+ },
421
+ "gripper": {
422
+ "absolute": true,
423
+ "rotation_type": null,
424
+ "shape": [
425
+ 1
426
+ ],
427
+ "continuous": true
428
+ }
429
+ }
430
+ },
431
+ "embodiment_tag": "new_embodiment"
432
+ }
433
+ }
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42f6ad456b5923aa3849000232dedc7c1653c26dfe327ab9fbf3b4c3d6b49b45
3
+ size 4999367032
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd37240782d968d52833f64d20281f64410c3094697038ab2af57d0e4b3b30ac
3
+ size 2586705312
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bf48e3365d5b510db19a60c9890b4d564a51fbf61f12341485e5fa17dac14b0
3
+ size 8550720062
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb56d629866c90996018f1f830acd404ef91d919d6f113d46f65a9fdb983c449
3
+ size 14244
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe8c620e9a7808e10f2bc64fa81fe5b1efb6dcc2acb3596ed32f58119a5368c1
3
+ size 1064
GuideVLA/headgate_rise16_bs128_30k_20261023/checkpoint-30000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "clvla_lm_logits_to_keep": 1,
59
+ "compute_dtype": "bfloat16",
60
+ "hidden_size": 2048,
61
+ "model_dtype": "float32",
62
+ "model_type": "gr00t_n1_5",
63
+ "torch_dtype": "bfloat16",
64
+ "transformers_version": "4.51.3"
65
+ }
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/experiment_cfg/clvla_contract.yaml ADDED
@@ -0,0 +1,201 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: clvla_n15_guidevla_region_mass_v1
2
+ variant: guidevla_region_mass
3
+ relation_mode: null
4
+ future_token_mode: checkpoint_weight
5
+ vision_tuning: {}
6
+ auxiliary: {}
7
+ task_loss: {}
8
+ state_rkd: {}
9
+ state_subspace: {}
10
+ action_rkd: {}
11
+ relation_mismatch: {}
12
+ objective: {}
13
+ input_cache: {}
14
+ episode_shard: {}
15
+ n17_official_shard: {}
16
+ training:
17
+ dataset_path:
18
+ - /home/work/workspace/test/seonho/libero
19
+ output_dir: /home/work/workspace/test/seonho/groot_libero/train_ckpt/GuideVLA/headgate_sparse_bs128_30k_20261023
20
+ experiment_name: GuideVLA_HeadGate_sparse_BS128_30K_20261023
21
+ wandb_project: finetune-gr00t-n1d5-libero
22
+ data_config: examples.Libero.custom_data_config:LiberoDataConfig
23
+ base_model_path: nvidia/GR00T-N1.5-3B
24
+ embodiment_tag: new_embodiment
25
+ video_backend: torchcodec
26
+ batch_size: 128
27
+ max_steps: 30000
28
+ num_gpus: 1
29
+ save_steps: 10000
30
+ save_total_limit: 1
31
+ tune_llm: false
32
+ tune_visual: false
33
+ tune_projector: true
34
+ tune_diffusion_model: true
35
+ learning_rate: 0.0001
36
+ gradient_accumulation_steps: 1
37
+ weight_decay: 1.0e-05
38
+ warmup_ratio: 0.05
39
+ lora_rank: 0
40
+ lora_alpha: 16
41
+ lora_dropout: 0.1
42
+ lora_full_model: false
43
+ dataloader_num_workers: 16
44
+ dataloader_prefetch_factor: 4
45
+ report_to: wandb
46
+ resume: false
47
+ balance_dataset_weights: true
48
+ balance_trajectory_weights: true
49
+ state_dropout_prob: 0.0
50
+ dataset_backend: libero_pi_v2
51
+ n15_commit: 48581302b549cb22c455436cbbbe17dc4ddfde75
52
+ config_path: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/configs/sparse.yaml
53
+ config_sha256: 2a9e8cbcc29faa855e8abd540f124f353edcb2cd374416922067ddf51a3939ce
54
+ data_contract: libero_pi_v2
55
+ source_dataset:
56
+ dataset_backend: libero_pi_v2
57
+ dataset_path: /home/work/workspace/test/seonho/libero
58
+ episodes: 1693
59
+ frames: 273465
60
+ tasks: 40
61
+ fps: 10
62
+ info_sha256: 77c0da93c67960bac803a8ea25dabf380d752c26c1d91c54151121dd10e7d82f
63
+ stats_sha256: 1fe6c2d548a7998394488ae68bc2be46a51c8cd3798af6ab74fff79f8b20e293
64
+ action_horizon: 16
65
+ config_status: ready
66
+ guidevla:
67
+ enabled: true
68
+ teacher_source: public_stage_object_mask
69
+ annotation_root: /home/work/workspace/test/seonho/clvla/Isaac-GR00T-N1.5/clvla/cfg_libero/GuideVLA/annotation
70
+ frame_policy: same_frame_index
71
+ missing_annotation_policy: FM_only
72
+ extra_alignment_filter: false
73
+ views:
74
+ - front
75
+ - wrist
76
+ grid:
77
+ - 16
78
+ - 16
79
+ mask_crop_replay: true
80
+ mask_coverage_threshold: 0.5
81
+ student:
82
+ module: action_head.model.transformer_blocks
83
+ attention_module: attn1
84
+ block_indices_0based:
85
+ - 6
86
+ - 8
87
+ - 10
88
+ expected_block_count: 16
89
+ expected_interleave_self_attention: true
90
+ expected_cross_attention_block_indices_0based:
91
+ - 0
92
+ - 2
93
+ - 4
94
+ - 6
95
+ - 8
96
+ - 10
97
+ - 12
98
+ - 14
99
+ head_indices_0based:
100
+ - 0
101
+ - 1
102
+ - 2
103
+ - 3
104
+ - 4
105
+ - 5
106
+ - 6
107
+ - 7
108
+ expected_head_count: 32
109
+ expected_head_dimension: 48
110
+ head_selection: fixed_prefix_no_samplewise_selection
111
+ query_selection: last_action_horizon_tokens
112
+ expected_action_query_count: 16
113
+ exclude_state_query: true
114
+ exclude_future_queries: true
115
+ ignore_padded_action_queries: true
116
+ query_validity_source: action_mask_any_valid_dimension
117
+ image_key_indices: backbone_image_context_token_positions_preserved_after_processing
118
+ softmax_scope: native_encoder_context_keys
119
+ renormalize_image_attention_for_loss: false
120
+ attention_probability_stage: before_attention_dropout
121
+ stop_gradient: false
122
+ loss:
123
+ name: negative_log_retained_region_mass
124
+ target: external_binary_stage_object_mask
125
+ use_teacher_probability_amplitude: false
126
+ normalize_binary_region_mask: false
127
+ non_image_key_target_weight: 0.0
128
+ selected_head_reduction: mean_before_region_mass
129
+ query_reduction: mean_log_loss_over_valid_action_queries
130
+ layer_reduction: mean
131
+ formula: mean_layer_valid_query(-log(clamp_min(sum_patch(mean_selected_head(P_image)*keep_mask),epsilon)))
132
+ epsilon: 1.0e-06
133
+ empty_or_zero_mass_teacher_policy: exclude_aux_keep_native_fm
134
+ flow_matching_weight: 1.0
135
+ lambda_start: 0.001
136
+ lambda_end: 0.001
137
+ lambda_schedule: constant
138
+ lambda_warmup_ratio: 0.0
139
+ lambda_warmup_optimizer_steps: 0
140
+ schedule_step_source: optimizer_global_step
141
+ gradient_surgery: none
142
+ guidevla_inference_behavior: stock_architecture_no_annotation_or_teacher_hooks
143
+ source_module_sha256:
144
+ clvla/data/guidevla_annotation.py: 8886d09bcd7bf5e8978372403c7e4a63d141aa4337e1e430afbe5eb442229467
145
+ clvla/patches/guidevla_region_mass_patch.py: 60d1a163eca3c23f226bd592263da357880d3132a0ef1fab03a78654cd2002f3
146
+ clvla/patches/aif_region_mass_patch.py: 177c7bbfb3c3135510b3fff9ccdea895e93f604c6e6f89d2d6ef245ab7a31a5d
147
+ clvla/ops/guidevla_20261018/bootstrap.py: c4f7f798156b3b821f4b2a6eb48f78072f3e6e0e6e6d751c1dfd99f20ba950ef
148
+ clvla/ops/guidevla_20261018/launch.py: 6ea6d33bed21141db8d05b2940106094b14cbad760e2ef22078400045f08649f
149
+ annotation_manifest_sha256: a0c2810a0bbb321acd5a43e7d5dee3d50a1f10460bee71a4446b0af55769d948
150
+ annotation_mapping_sha256: 715d209e1d331f95c4b94c431a40374c12c09e6d680b2c4fd4c042bae9111b3f
151
+ distribution:
152
+ method: sparse
153
+ auxiliary_start_step: 0
154
+ auxiliary_ramp_steps: 0
155
+ weight_policy: constant
156
+ prior_floor: 0.9
157
+ target_kl_budget: 0.01
158
+ sparse_regions_per_view: 2
159
+ sparse_block_shape:
160
+ - 2
161
+ - 2
162
+ canonical_frames_per_annotated_demo: 16
163
+ original_FM_region_on_all_frames: true
164
+ inference: unchanged; no teacher, labels or additional head
165
+ weight: 0.1
166
+ allocator_limit_gib: 72
167
+ importance_source: current_training_policy_online
168
+ supervision_prior: detached_current_native_attention_same_forward
169
+ intervention: supervised_head_action_query_post_softmax_removal
170
+ raw_provenance_root: /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/raw/sparse
171
+ gate_scope:
172
+ layers:
173
+ - 6
174
+ - 8
175
+ - 10
176
+ heads:
177
+ - 0
178
+ - 1
179
+ - 2
180
+ - 3
181
+ - 4
182
+ - 5
183
+ - 6
184
+ - 7
185
+ queries: last16_valid_action
186
+ keys: existing_ROI_image_keys
187
+ online_chunk_size: 4
188
+ online_fm_noise_draws: 2
189
+ online_fm_times:
190
+ - 0.3
191
+ - 0.7
192
+ online_target_stop_gradient: true
193
+ raw_snapshot_steps: 200
194
+ online_selection: all selected items; no sample cap
195
+ additional_VLM_forwards: 0
196
+ distribution_source_sha256:
197
+ /home/work/workspace/test/seonho/analysis/guidevla_headgate_train_20261023/online_runtime.py: c1514aa243ee54014159ff6d86ea4e77e69a93a8058b897a13ab059a761c214c
198
+ /home/work/workspace/test/seonho/analysis/guidevla_head_gate_20261023/head_gate.py: 45084629aea378610a917f31cf777c9b2436b192fd8cde3c34c6dfe78cc83a45
199
+ /home/work/workspace/test/seonho/analysis/guidevla_head_gate_20261023/image_reference.py: 890c278cc885f4e5200cc24ed44033eb11330e2b43655bf59fd54c7465ca52fb
200
+ distribution_inference: unchanged native model; head gates are training-only importance
201
+ interventions
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "x": {
6
+ "max": [
7
+ 0.21031762659549713
8
+ ],
9
+ "min": [
10
+ -0.4828203022480011
11
+ ],
12
+ "mean": [
13
+ -0.04651879519224167
14
+ ],
15
+ "std": [
16
+ 0.10494378954172134
17
+ ],
18
+ "q01": [
19
+ -0.39912487387657164
20
+ ],
21
+ "q99": [
22
+ 0.13556506335735302
23
+ ]
24
+ },
25
+ "y": {
26
+ "max": [
27
+ 0.39128610491752625
28
+ ],
29
+ "min": [
30
+ -0.3255046010017395
31
+ ],
32
+ "mean": [
33
+ 0.03440921753644943
34
+ ],
35
+ "std": [
36
+ 0.15176637470722198
37
+ ],
38
+ "q01": [
39
+ -0.26883512616157534
40
+ ],
41
+ "q99": [
42
+ 0.33566484451293943
43
+ ]
44
+ },
45
+ "z": {
46
+ "max": [
47
+ 1.3660105466842651
48
+ ],
49
+ "min": [
50
+ 0.008128180168569088
51
+ ],
52
+ "mean": [
53
+ 0.7645525336265564
54
+ ],
55
+ "std": [
56
+ 0.3785160183906555
57
+ ],
58
+ "q01": [
59
+ 0.03826696485280991
60
+ ],
61
+ "q99": [
62
+ 1.2706659078598022
63
+ ]
64
+ },
65
+ "roll": {
66
+ "max": [
67
+ 3.6714255809783936
68
+ ],
69
+ "min": [
70
+ 0.35277295112609863
71
+ ],
72
+ "mean": [
73
+ 2.972202777862549
74
+ ],
75
+ "std": [
76
+ 0.34427398443222046
77
+ ],
78
+ "q01": [
79
+ 1.5089580535888671
80
+ ],
81
+ "q99": [
82
+ 3.2773460674285886
83
+ ]
84
+ },
85
+ "pitch": {
86
+ "max": [
87
+ 3.560650587081909
88
+ ],
89
+ "min": [
90
+ -3.641430377960205
91
+ ],
92
+ "mean": [
93
+ -0.22047005593776703
94
+ ],
95
+ "std": [
96
+ 0.9069469571113586
97
+ ],
98
+ "q01": [
99
+ -2.719791135787964
100
+ ],
101
+ "q99": [
102
+ 2.406109695434569
103
+ ]
104
+ },
105
+ "yaw": {
106
+ "max": [
107
+ 1.386339545249939
108
+ ],
109
+ "min": [
110
+ -1.842738389968872
111
+ ],
112
+ "mean": [
113
+ -0.1255796253681183
114
+ ],
115
+ "std": [
116
+ 0.3253920078277588
117
+ ],
118
+ "q01": [
119
+ -1.08050856590271
120
+ ],
121
+ "q99": [
122
+ 0.59776972055435
123
+ ]
124
+ },
125
+ "gripper": {
126
+ "max": [
127
+ 0.04233968257904053,
128
+ 0.0013633022317662835
129
+ ],
130
+ "min": [
131
+ -0.0013586411951109767,
132
+ -0.042040832340717316
133
+ ],
134
+ "mean": [
135
+ 0.026914266869425774,
136
+ -0.02719070389866829
137
+ ],
138
+ "std": [
139
+ 0.014175914227962494,
140
+ 0.014058894477784634
141
+ ],
142
+ "q01": [
143
+ 0.0017423711018636821,
144
+ -0.04002561450004578
145
+ ],
146
+ "q99": [
147
+ 0.04031316146254539,
148
+ -0.0017781084170565016
149
+ ]
150
+ }
151
+ },
152
+ "action": {
153
+ "x": {
154
+ "max": [
155
+ 0.9375
156
+ ],
157
+ "min": [
158
+ -0.9375
159
+ ],
160
+ "mean": [
161
+ 0.06278137117624283
162
+ ],
163
+ "std": [
164
+ 0.3355240225791931
165
+ ],
166
+ "q01": [
167
+ -0.7044642567634583
168
+ ],
169
+ "q99": [
170
+ 0.9375
171
+ ]
172
+ },
173
+ "y": {
174
+ "max": [
175
+ 0.9375
176
+ ],
177
+ "min": [
178
+ -0.9375
179
+ ],
180
+ "mean": [
181
+ 0.0868409126996994
182
+ ],
183
+ "std": [
184
+ 0.3784470558166504
185
+ ],
186
+ "q01": [
187
+ -0.8008928298950195
188
+ ],
189
+ "q99": [
190
+ 0.8678571581840515
191
+ ]
192
+ },
193
+ "z": {
194
+ "max": [
195
+ 0.9375
196
+ ],
197
+ "min": [
198
+ -0.9375
199
+ ],
200
+ "mean": [
201
+ -0.09037282317876816
202
+ ],
203
+ "std": [
204
+ 0.44472837448120117
205
+ ],
206
+ "q01": [
207
+ -0.9375
208
+ ],
209
+ "q99": [
210
+ 0.9375
211
+ ]
212
+ },
213
+ "roll": {
214
+ "max": [
215
+ 0.3557142913341522
216
+ ],
217
+ "min": [
218
+ -0.2582142949104309
219
+ ],
220
+ "mean": [
221
+ 0.0005407406715676188
222
+ ],
223
+ "std": [
224
+ 0.03924351558089256
225
+ ],
226
+ "q01": [
227
+ -0.11464285850524902
228
+ ],
229
+ "q99": [
230
+ 0.13178572058677673
231
+ ]
232
+ },
233
+ "pitch": {
234
+ "max": [
235
+ 0.375
236
+ ],
237
+ "min": [
238
+ -0.375
239
+ ],
240
+ "mean": [
241
+ 0.005643361248075962
242
+ ],
243
+ "std": [
244
+ 0.06339313089847565
245
+ ],
246
+ "q01": [
247
+ -0.1639285683631897
248
+ ],
249
+ "q99": [
250
+ 0.19285714626312256
251
+ ]
252
+ },
253
+ "yaw": {
254
+ "max": [
255
+ 0.375
256
+ ],
257
+ "min": [
258
+ -0.3675000071525574
259
+ ],
260
+ "mean": [
261
+ -0.005229088477790356
262
+ ],
263
+ "std": [
264
+ 0.07797032594680786
265
+ ],
266
+ "q01": [
267
+ -0.2239285707473755
268
+ ],
269
+ "q99": [
270
+ 0.335357129573822
271
+ ]
272
+ },
273
+ "gripper": {
274
+ "max": [
275
+ 1.0
276
+ ],
277
+ "min": [
278
+ -1.0
279
+ ],
280
+ "mean": [
281
+ -0.04964079707860947
282
+ ],
283
+ "std": [
284
+ 0.9987710118293762
285
+ ],
286
+ "q01": [
287
+ -1.0
288
+ ],
289
+ "q99": [
290
+ 1.0
291
+ ]
292
+ }
293
+ }
294
+ },
295
+ "modalities": {
296
+ "video": {
297
+ "image": {
298
+ "resolution": [
299
+ 256,
300
+ 256
301
+ ],
302
+ "channels": 3,
303
+ "fps": 10.0
304
+ },
305
+ "wrist_image": {
306
+ "resolution": [
307
+ 256,
308
+ 256
309
+ ],
310
+ "channels": 3,
311
+ "fps": 10.0
312
+ }
313
+ },
314
+ "state": {
315
+ "x": {
316
+ "absolute": true,
317
+ "rotation_type": null,
318
+ "shape": [
319
+ 1
320
+ ],
321
+ "continuous": true
322
+ },
323
+ "y": {
324
+ "absolute": true,
325
+ "rotation_type": null,
326
+ "shape": [
327
+ 1
328
+ ],
329
+ "continuous": true
330
+ },
331
+ "z": {
332
+ "absolute": true,
333
+ "rotation_type": null,
334
+ "shape": [
335
+ 1
336
+ ],
337
+ "continuous": true
338
+ },
339
+ "roll": {
340
+ "absolute": true,
341
+ "rotation_type": null,
342
+ "shape": [
343
+ 1
344
+ ],
345
+ "continuous": true
346
+ },
347
+ "pitch": {
348
+ "absolute": true,
349
+ "rotation_type": null,
350
+ "shape": [
351
+ 1
352
+ ],
353
+ "continuous": true
354
+ },
355
+ "yaw": {
356
+ "absolute": true,
357
+ "rotation_type": null,
358
+ "shape": [
359
+ 1
360
+ ],
361
+ "continuous": true
362
+ },
363
+ "gripper": {
364
+ "absolute": true,
365
+ "rotation_type": null,
366
+ "shape": [
367
+ 2
368
+ ],
369
+ "continuous": true
370
+ }
371
+ },
372
+ "action": {
373
+ "x": {
374
+ "absolute": true,
375
+ "rotation_type": null,
376
+ "shape": [
377
+ 1
378
+ ],
379
+ "continuous": true
380
+ },
381
+ "y": {
382
+ "absolute": true,
383
+ "rotation_type": null,
384
+ "shape": [
385
+ 1
386
+ ],
387
+ "continuous": true
388
+ },
389
+ "z": {
390
+ "absolute": true,
391
+ "rotation_type": null,
392
+ "shape": [
393
+ 1
394
+ ],
395
+ "continuous": true
396
+ },
397
+ "roll": {
398
+ "absolute": true,
399
+ "rotation_type": null,
400
+ "shape": [
401
+ 1
402
+ ],
403
+ "continuous": true
404
+ },
405
+ "pitch": {
406
+ "absolute": true,
407
+ "rotation_type": null,
408
+ "shape": [
409
+ 1
410
+ ],
411
+ "continuous": true
412
+ },
413
+ "yaw": {
414
+ "absolute": true,
415
+ "rotation_type": null,
416
+ "shape": [
417
+ 1
418
+ ],
419
+ "continuous": true
420
+ },
421
+ "gripper": {
422
+ "absolute": true,
423
+ "rotation_type": null,
424
+ "shape": [
425
+ 1
426
+ ],
427
+ "continuous": true
428
+ }
429
+ }
430
+ },
431
+ "embodiment_tag": "new_embodiment"
432
+ }
433
+ }
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45bcd4d4bcef809d7e7ba816a13f18627fd60a91306beff85f044981e85f49fe
3
+ size 4999367032
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d2ddc8f674e3349e6cae9e9ddebca1925288207d9bf53767a8cfb4d8e689502
3
+ size 2586705312
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:488db06e67684d5152a6e2d8fd7e29f20a759a410c1c5e86b7bce728429b275c
3
+ size 8550720062
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb56d629866c90996018f1f830acd404ef91d919d6f113d46f65a9fdb983c449
3
+ size 14244
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe8c620e9a7808e10f2bc64fa81fe5b1efb6dcc2acb3596ed32f58119a5368c1
3
+ size 1064
GuideVLA/headgate_sparse_bs128_30k_20261023/checkpoint-30000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
baseline_bs128_30k/checkpoint-30000/config.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "compute_dtype": "bfloat16",
59
+ "hidden_size": 2048,
60
+ "model_dtype": "float32",
61
+ "model_type": "gr00t_n1_5",
62
+ "torch_dtype": "bfloat16",
63
+ "transformers_version": "4.51.3"
64
+ }
baseline_bs128_30k/checkpoint-30000/experiment_cfg/clvla_contract.yaml ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: clvla_n15_v1
2
+ variant: baseline
3
+ relation_mode: null
4
+ future_token_mode: checkpoint_weight
5
+ vision_tuning: {}
6
+ auxiliary: {}
7
+ task_loss: {}
8
+ state_rkd: {}
9
+ state_subspace: {}
10
+ action_rkd: {}
11
+ relation_mismatch: {}
12
+ objective: {}
13
+ input_cache: {}
14
+ episode_shard: {}
15
+ n17_official_shard: {}
16
+ training:
17
+ dataset_path:
18
+ - /home/work/workspace/test/seonho/libero
19
+ output_dir: /home/work/workspace/test/seonho/groot_libero/train_ckpt/baseline_bs128_30k
20
+ experiment_name: Baseline_BS128_30K
21
+ wandb_project: finetune-gr00t-n1d5-libero
22
+ data_config: examples.Libero.custom_data_config:LiberoDataConfig
23
+ base_model_path: nvidia/GR00T-N1.5-3B
24
+ embodiment_tag: new_embodiment
25
+ video_backend: torchcodec
26
+ batch_size: 128
27
+ max_steps: 30000
28
+ num_gpus: 1
29
+ save_steps: 10000
30
+ save_total_limit: 1
31
+ tune_llm: false
32
+ tune_visual: false
33
+ tune_projector: true
34
+ tune_diffusion_model: true
35
+ learning_rate: 0.0001
36
+ gradient_accumulation_steps: 1
37
+ weight_decay: 1.0e-05
38
+ warmup_ratio: 0.05
39
+ lora_rank: 0
40
+ lora_alpha: 16
41
+ lora_dropout: 0.1
42
+ lora_full_model: false
43
+ dataloader_num_workers: 16
44
+ dataloader_prefetch_factor: 4
45
+ report_to: wandb
46
+ resume: false
47
+ balance_dataset_weights: true
48
+ balance_trajectory_weights: true
49
+ state_dropout_prob: 0.0
50
+ dataset_backend: libero_pi_v2
51
+ n15_commit: f2dcd1c0e7fb16260c851e1d3cc9a8a9860b8f05
52
+ config_path: /home/work/workspace/test/seonho/clvla/Isaac-GR00T-N1.5/clvla/cfg_libero/baseline/baseline_bs128_30k.yaml
53
+ config_sha256: 237d50012a8a0fc52e55422034166cdf24f53164b0337cd7f03e4374c569defd
54
+ data_contract: libero_pi_v2
55
+ source_dataset:
56
+ dataset_backend: libero_pi_v2
57
+ dataset_path: /home/work/workspace/test/seonho/libero
58
+ episodes: 1693
59
+ frames: 273465
60
+ tasks: 40
61
+ fps: 10
62
+ info_sha256: 77c0da93c67960bac803a8ea25dabf380d752c26c1d91c54151121dd10e7d82f
63
+ stats_sha256: 1fe6c2d548a7998394488ae68bc2be46a51c8cd3798af6ab74fff79f8b20e293
64
+ action_horizon: 16
baseline_bs128_30k/checkpoint-30000/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "x": {
6
+ "max": [
7
+ 0.21031762659549713
8
+ ],
9
+ "min": [
10
+ -0.4828203022480011
11
+ ],
12
+ "mean": [
13
+ -0.04651879519224167
14
+ ],
15
+ "std": [
16
+ 0.10494378954172134
17
+ ],
18
+ "q01": [
19
+ -0.39912487387657164
20
+ ],
21
+ "q99": [
22
+ 0.13556506335735302
23
+ ]
24
+ },
25
+ "y": {
26
+ "max": [
27
+ 0.39128610491752625
28
+ ],
29
+ "min": [
30
+ -0.3255046010017395
31
+ ],
32
+ "mean": [
33
+ 0.03440921753644943
34
+ ],
35
+ "std": [
36
+ 0.15176637470722198
37
+ ],
38
+ "q01": [
39
+ -0.26883512616157534
40
+ ],
41
+ "q99": [
42
+ 0.33566484451293943
43
+ ]
44
+ },
45
+ "z": {
46
+ "max": [
47
+ 1.3660105466842651
48
+ ],
49
+ "min": [
50
+ 0.008128180168569088
51
+ ],
52
+ "mean": [
53
+ 0.7645525336265564
54
+ ],
55
+ "std": [
56
+ 0.3785160183906555
57
+ ],
58
+ "q01": [
59
+ 0.03826696485280991
60
+ ],
61
+ "q99": [
62
+ 1.2706659078598022
63
+ ]
64
+ },
65
+ "roll": {
66
+ "max": [
67
+ 3.6714255809783936
68
+ ],
69
+ "min": [
70
+ 0.35277295112609863
71
+ ],
72
+ "mean": [
73
+ 2.972202777862549
74
+ ],
75
+ "std": [
76
+ 0.34427398443222046
77
+ ],
78
+ "q01": [
79
+ 1.5089580535888671
80
+ ],
81
+ "q99": [
82
+ 3.2773460674285886
83
+ ]
84
+ },
85
+ "pitch": {
86
+ "max": [
87
+ 3.560650587081909
88
+ ],
89
+ "min": [
90
+ -3.641430377960205
91
+ ],
92
+ "mean": [
93
+ -0.22047005593776703
94
+ ],
95
+ "std": [
96
+ 0.9069469571113586
97
+ ],
98
+ "q01": [
99
+ -2.719791135787964
100
+ ],
101
+ "q99": [
102
+ 2.406109695434569
103
+ ]
104
+ },
105
+ "yaw": {
106
+ "max": [
107
+ 1.386339545249939
108
+ ],
109
+ "min": [
110
+ -1.842738389968872
111
+ ],
112
+ "mean": [
113
+ -0.1255796253681183
114
+ ],
115
+ "std": [
116
+ 0.3253920078277588
117
+ ],
118
+ "q01": [
119
+ -1.08050856590271
120
+ ],
121
+ "q99": [
122
+ 0.59776972055435
123
+ ]
124
+ },
125
+ "gripper": {
126
+ "max": [
127
+ 0.04233968257904053,
128
+ 0.0013633022317662835
129
+ ],
130
+ "min": [
131
+ -0.0013586411951109767,
132
+ -0.042040832340717316
133
+ ],
134
+ "mean": [
135
+ 0.026914266869425774,
136
+ -0.02719070389866829
137
+ ],
138
+ "std": [
139
+ 0.014175914227962494,
140
+ 0.014058894477784634
141
+ ],
142
+ "q01": [
143
+ 0.0017423711018636821,
144
+ -0.04002561450004578
145
+ ],
146
+ "q99": [
147
+ 0.04031316146254539,
148
+ -0.0017781084170565016
149
+ ]
150
+ }
151
+ },
152
+ "action": {
153
+ "x": {
154
+ "max": [
155
+ 0.9375
156
+ ],
157
+ "min": [
158
+ -0.9375
159
+ ],
160
+ "mean": [
161
+ 0.06278137117624283
162
+ ],
163
+ "std": [
164
+ 0.3355240225791931
165
+ ],
166
+ "q01": [
167
+ -0.7044642567634583
168
+ ],
169
+ "q99": [
170
+ 0.9375
171
+ ]
172
+ },
173
+ "y": {
174
+ "max": [
175
+ 0.9375
176
+ ],
177
+ "min": [
178
+ -0.9375
179
+ ],
180
+ "mean": [
181
+ 0.0868409126996994
182
+ ],
183
+ "std": [
184
+ 0.3784470558166504
185
+ ],
186
+ "q01": [
187
+ -0.8008928298950195
188
+ ],
189
+ "q99": [
190
+ 0.8678571581840515
191
+ ]
192
+ },
193
+ "z": {
194
+ "max": [
195
+ 0.9375
196
+ ],
197
+ "min": [
198
+ -0.9375
199
+ ],
200
+ "mean": [
201
+ -0.09037282317876816
202
+ ],
203
+ "std": [
204
+ 0.44472837448120117
205
+ ],
206
+ "q01": [
207
+ -0.9375
208
+ ],
209
+ "q99": [
210
+ 0.9375
211
+ ]
212
+ },
213
+ "roll": {
214
+ "max": [
215
+ 0.3557142913341522
216
+ ],
217
+ "min": [
218
+ -0.2582142949104309
219
+ ],
220
+ "mean": [
221
+ 0.0005407406715676188
222
+ ],
223
+ "std": [
224
+ 0.03924351558089256
225
+ ],
226
+ "q01": [
227
+ -0.11464285850524902
228
+ ],
229
+ "q99": [
230
+ 0.13178572058677673
231
+ ]
232
+ },
233
+ "pitch": {
234
+ "max": [
235
+ 0.375
236
+ ],
237
+ "min": [
238
+ -0.375
239
+ ],
240
+ "mean": [
241
+ 0.005643361248075962
242
+ ],
243
+ "std": [
244
+ 0.06339313089847565
245
+ ],
246
+ "q01": [
247
+ -0.1639285683631897
248
+ ],
249
+ "q99": [
250
+ 0.19285714626312256
251
+ ]
252
+ },
253
+ "yaw": {
254
+ "max": [
255
+ 0.375
256
+ ],
257
+ "min": [
258
+ -0.3675000071525574
259
+ ],
260
+ "mean": [
261
+ -0.005229088477790356
262
+ ],
263
+ "std": [
264
+ 0.07797032594680786
265
+ ],
266
+ "q01": [
267
+ -0.2239285707473755
268
+ ],
269
+ "q99": [
270
+ 0.335357129573822
271
+ ]
272
+ },
273
+ "gripper": {
274
+ "max": [
275
+ 1.0
276
+ ],
277
+ "min": [
278
+ -1.0
279
+ ],
280
+ "mean": [
281
+ -0.04964079707860947
282
+ ],
283
+ "std": [
284
+ 0.9987710118293762
285
+ ],
286
+ "q01": [
287
+ -1.0
288
+ ],
289
+ "q99": [
290
+ 1.0
291
+ ]
292
+ }
293
+ }
294
+ },
295
+ "modalities": {
296
+ "video": {
297
+ "image": {
298
+ "resolution": [
299
+ 256,
300
+ 256
301
+ ],
302
+ "channels": 3,
303
+ "fps": 10.0
304
+ },
305
+ "wrist_image": {
306
+ "resolution": [
307
+ 256,
308
+ 256
309
+ ],
310
+ "channels": 3,
311
+ "fps": 10.0
312
+ }
313
+ },
314
+ "state": {
315
+ "x": {
316
+ "absolute": true,
317
+ "rotation_type": null,
318
+ "shape": [
319
+ 1
320
+ ],
321
+ "continuous": true
322
+ },
323
+ "y": {
324
+ "absolute": true,
325
+ "rotation_type": null,
326
+ "shape": [
327
+ 1
328
+ ],
329
+ "continuous": true
330
+ },
331
+ "z": {
332
+ "absolute": true,
333
+ "rotation_type": null,
334
+ "shape": [
335
+ 1
336
+ ],
337
+ "continuous": true
338
+ },
339
+ "roll": {
340
+ "absolute": true,
341
+ "rotation_type": null,
342
+ "shape": [
343
+ 1
344
+ ],
345
+ "continuous": true
346
+ },
347
+ "pitch": {
348
+ "absolute": true,
349
+ "rotation_type": null,
350
+ "shape": [
351
+ 1
352
+ ],
353
+ "continuous": true
354
+ },
355
+ "yaw": {
356
+ "absolute": true,
357
+ "rotation_type": null,
358
+ "shape": [
359
+ 1
360
+ ],
361
+ "continuous": true
362
+ },
363
+ "gripper": {
364
+ "absolute": true,
365
+ "rotation_type": null,
366
+ "shape": [
367
+ 2
368
+ ],
369
+ "continuous": true
370
+ }
371
+ },
372
+ "action": {
373
+ "x": {
374
+ "absolute": true,
375
+ "rotation_type": null,
376
+ "shape": [
377
+ 1
378
+ ],
379
+ "continuous": true
380
+ },
381
+ "y": {
382
+ "absolute": true,
383
+ "rotation_type": null,
384
+ "shape": [
385
+ 1
386
+ ],
387
+ "continuous": true
388
+ },
389
+ "z": {
390
+ "absolute": true,
391
+ "rotation_type": null,
392
+ "shape": [
393
+ 1
394
+ ],
395
+ "continuous": true
396
+ },
397
+ "roll": {
398
+ "absolute": true,
399
+ "rotation_type": null,
400
+ "shape": [
401
+ 1
402
+ ],
403
+ "continuous": true
404
+ },
405
+ "pitch": {
406
+ "absolute": true,
407
+ "rotation_type": null,
408
+ "shape": [
409
+ 1
410
+ ],
411
+ "continuous": true
412
+ },
413
+ "yaw": {
414
+ "absolute": true,
415
+ "rotation_type": null,
416
+ "shape": [
417
+ 1
418
+ ],
419
+ "continuous": true
420
+ },
421
+ "gripper": {
422
+ "absolute": true,
423
+ "rotation_type": null,
424
+ "shape": [
425
+ 1
426
+ ],
427
+ "continuous": true
428
+ }
429
+ }
430
+ },
431
+ "embodiment_tag": "new_embodiment"
432
+ }
433
+ }
baseline_bs128_30k/checkpoint-30000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daa1b8992ade69428cbe9be6d9c0319e014a2ee99b547ea992448894b757640a
3
+ size 4999367032
baseline_bs128_30k/checkpoint-30000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ae101b866e3dbef0c3a3e535c28c13f33b2e42a80ff571019a29414d392b10f
3
+ size 2586705312
baseline_bs128_30k/checkpoint-30000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
baseline_bs128_30k/checkpoint-30000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c80b11d5f801e48d01d05f80e4e09efc793f390a7e8bc93a765ba966e38543e
3
+ size 8550720062
baseline_bs128_30k/checkpoint-30000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb56d629866c90996018f1f830acd404ef91d919d6f113d46f65a9fdb983c449
3
+ size 14244
baseline_bs128_30k/checkpoint-30000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe8c620e9a7808e10f2bc64fa81fe5b1efb6dcc2acb3596ed32f58119a5368c1
3
+ size 1064
baseline_bs128_30k/checkpoint-30000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff