yangfy0627 commited on
Commit
cc52a28
·
verified ·
1 Parent(s): b86b04c

Visual BCs: use epoch 1400 for all nine locally trained jobs (user decision 2026-10-27); add ToolHang diverse100; archive prior releases

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 +1 -0
  2. README.md +1 -1
  3. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/README.md +4 -14
  4. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/SHA256SUMS.json +41 -45
  5. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/best.json +3 -153
  6. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1400.json +188 -0
  7. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1400.pkl +3 -0
  8. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/epoch_1400.parquet +3 -0
  9. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/protocol.json +3 -3
  10. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/result.json +0 -0
  11. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/provenance/evaluate_best.py +70 -0
  12. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/release.json +36 -29
  13. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/README.md +17 -0
  14. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/SHA256SUMS.json +102 -0
  15. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/best.json +157 -0
  16. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/checkpoints/epoch_1050.json +188 -0
  17. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/checkpoints/epoch_1050.pkl +3 -0
  18. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/data_manifest.json +436 -0
  19. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/evaluation_schedule.json +17 -0
  20. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/evaluations.jsonl +58 -0
  21. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/epoch_1050.parquet +3 -0
  22. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/protocol.json +226 -0
  23. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/result.json +0 -0
  24. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/first_phase_complete.json +10 -0
  25. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/normalizer.json +188 -0
  26. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/prior_index_record.json +59 -0
  27. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/bc_eval_schedule.py +68 -0
  28. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/effective_train_eval50.py +132 -0
  29. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/evaluation_schedule_active.json +9 -0
  30. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/native_training_status.json +12 -0
  31. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/train_bc_noleak.py +54 -0
  32. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/release.json +59 -0
  33. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/resolved_config.json +160 -0
  34. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/selection_complete.json +186 -0
  35. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/selection_supervision.json +12 -0
  36. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/split.json +61 -0
  37. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/train_metrics.jsonl +0 -0
  38. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/training_status.json +6 -0
  39. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/validation_metrics.jsonl +0 -0
  40. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json +36 -179
  41. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_supervision.json +8 -7
  42. visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/training_status.json +2 -3
  43. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/README.md +4 -16
  44. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/SHA256SUMS.json +40 -96
  45. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/best.json +3 -153
  46. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/checkpoints/epoch_1400.json +188 -0
  47. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/checkpoints/epoch_1400.pkl +3 -0
  48. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/epoch_1400.parquet +3 -0
  49. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/protocol.json +4 -6
  50. visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/result.json +0 -0
.gitattributes CHANGED
@@ -43,3 +43,4 @@ visual_no_collision/reviews/20261026_remaining2/joint_curve_review.pdf filter=lf
43
  visual_no_collision/reviews/20261026_remaining2/joint_curves.png filter=lfs diff=lfs merge=lfs -text
44
  visual_no_collision/Robomimic/study1/ToolHang/focused100/seed_1/selection_validation/candidates.png filter=lfs diff=lfs merge=lfs -text
45
  visual_no_collision/reviews/20261026_toolhang_f100_reselect/candidates.png filter=lfs diff=lfs merge=lfs -text
 
 
43
  visual_no_collision/reviews/20261026_remaining2/joint_curves.png filter=lfs diff=lfs merge=lfs -text
44
  visual_no_collision/Robomimic/study1/ToolHang/focused100/seed_1/selection_validation/candidates.png filter=lfs diff=lfs merge=lfs -text
45
  visual_no_collision/reviews/20261026_toolhang_f100_reselect/candidates.png filter=lfs diff=lfs merge=lfs -text
46
+ visual_no_collision/Robomimic/study1/ToolHang/focused100/seed_1/releases/before_20261027_e1400/selection_validation/candidates.png filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -81,5 +81,5 @@ DexMimicGen/
81
  <!-- visual-no-collision-bc:start -->
82
  ## 纯视觉输入 BC(不渲染碰撞网格)
83
 
84
- 当前已定稿 12 项。状态、权重与独立 200 局评估见 [visual_no_collision](visual_no_collision/README.md)。请按 selected_after_joint_review 标记识别定稿权重;定稿不表示零过拟合或全局最优。旧候选及历史碰撞渲染模型均保留。
85
  <!-- visual-no-collision-bc:end -->
 
81
  <!-- visual-no-collision-bc:start -->
82
  ## 纯视觉输入 BC(不渲染碰撞网格)
83
 
84
+ 当前已定稿 9 项;本机训练的 9 项按用户决定(2026-10-27)统一使用 epoch 1400。状态、权重与 200 局评估见 [visual_no_collision](visual_no_collision/README.md)。请按 selected_after_joint_review 标记识别定稿权重;定稿不表示零过拟合或全局最优。旧版本均在各自 releases/ 下保留。
85
  <!-- visual-no-collision-bc:end -->
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/README.md CHANGED
@@ -1,17 +1,7 @@
1
- # DrawerCleanup diverse50: reviewed visual-only BC
2
 
3
- Selected **epoch1050**, full EMA encoder + flow policy, collision_geoms=false. The2000-epoch training phase is complete.
4
 
5
- Train mean e951-1000 0.0677247 -> e1001-1050 0.0664753 -> e1051-1100 0.0655113; improvement has slowed to about1.5-1.8% per50 epochs.
6
 
7
- Validation mean e951-1000 0.143344 -> e1001-1050 0.145048 -> e1051-1100 0.146439; e1801-1950 0.162890. Prefer e1050 to later more overfit weights.
8
-
9
- Periodic success e1020/1050/1080=62/64/56%, window60.67%. Later150-epoch blocks55.6/54/58/55.33/56.67/56.67%, endpoint56%. No sustained later gain.
10
-
11
- This is a practical joint curve selection. Validation loss already exceeds its historical minimum; no claim of zero overfitting or globally optimal success is made.
12
-
13
- Independent final200: **122/200 = 61.0%**, Wilson95% CI [54.1%, 67.5%]. The checkpoint was fixed before evaluating this panel; the result never drives reselection. The unchanged prior report is reused.
14
-
15
- SHA256: `bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5`.
16
-
17
- `selection_complete.json` records the decision; `evaluation_schedule.json` and `provenance/` describe the explicit evaluation-cadence change while retaining the original sealed training contract. `SHA256SUMS.json` covers this current release. The exact prior release is preserved under [releases/before_20261026_e1050](releases/before_20261026_e1050/README.md).
 
1
+ # DrawerCleanup diverse50: visual-only BC, epoch 1400
2
 
3
+ Selected **epoch 1400** by user decision (2026-10-27): every locally trained visual BC uses e1400. Complete EMA encoder and flow policy; collision_geoms=false. `release.json` names the canonical checkpoint; any other checkpoint left in `checkpoints/` belongs to an archived release.
4
 
5
+ Final200: **114/200 = 57.0%**, Wilson 95% CI [50.1%, 63.7%]. Same panel, prior selection e1050: 122/200 vs e1400 114/200 (only prior 38, only e1400 30, exact McNemar p=0.40). This panel was seen before the decision, so it is not an independent test. Trained to the e2000 first-phase boundary; e1400 is an intermediate saved checkpoint. Single training seed; zero overfitting is not established.
6
 
7
+ SHA256: `06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383`. The prior release is preserved under [releases/before_20261027_e1400](releases/before_20261027_e1400/README.md).
 
 
 
 
 
 
 
 
 
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/SHA256SUMS.json CHANGED
@@ -1,20 +1,12 @@
1
  {
2
- "checkpoints/epoch_1050.pkl": {
3
- "sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
4
- "bytes": 138495554
5
  },
6
- "checkpoints/epoch_1050.json": {
7
- "sha256": "12f4753d1b6471d5bf3d78d68017a325ded95e4c45ea1c4d966a0ff148d48041",
8
  "bytes": 7966
9
  },
10
- "best.json": {
11
- "sha256": "36a7c38e2d329aceaea261784d39b7f2c3dbe7258127733e020a16b4370d1911",
12
- "bytes": 3996
13
- },
14
- "selection_complete.json": {
15
- "sha256": "dd7b9eb566691013bf815fe3eb1bd13814e17821a069eeaa377dd31de9a25e13",
16
- "bytes": 6372
17
- },
18
  "resolved_config.json": {
19
  "sha256": "3e7b7c521a37e9040090b90718d27ec9ea0a66259ce222ed2f6a97c5013880cc",
20
  "bytes": 6617
@@ -43,14 +35,30 @@
43
  "sha256": "fa560dba6e2c8a2d46a1618f41423b97fac6b8dcadcfd22dc38ca205cd08b524",
44
  "bytes": 82794
45
  },
46
- "first_phase_complete.json": {
47
- "sha256": "139f202e3772aa538ab4aa0a3ed0f5202e881b9254b3a8d6e171c5b79618a7de",
48
- "bytes": 523
49
- },
50
  "provenance/native_training_status.json": {
51
  "sha256": "69c9110b3e73ecbe48da0b507acab1ed89b5dbd0ecd58106642ffc403ede3719",
52
  "bytes": 334
53
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  "evaluation_schedule.json": {
55
  "sha256": "30754536ae01c1ba0c0696c815623c035640798e025ee6ea068e46c53629b0a6",
56
  "bytes": 1009
@@ -63,40 +71,28 @@
63
  "sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
64
  "bytes": 9987
65
  },
66
- "provenance/bc_eval_schedule.py": {
67
- "sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
68
- "bytes": 3539
69
- },
70
- "provenance/train_bc_noleak.py": {
71
- "sha256": "587b66eca4c82c8ee519ca7df23874ca9a9acdbb8cdabd8a36afb68ebcf891d5",
72
- "bytes": 2186
73
- },
74
- "final200/result.json": {
75
- "sha256": "61b57c921c11de832c70836f4c1987dbe7dfb4f614b314ca2cc8e2cec02a401c",
76
- "bytes": 119075
77
- },
78
- "final200/protocol.json": {
79
- "sha256": "b30eeb8b6b735db58d581ea2ac22934431c3f4dd24c50178a370f2f5f5626f07",
80
- "bytes": 3993
81
- },
82
- "final200/epoch_1050.parquet": {
83
- "sha256": "36e242073f9d8296209dacd2221969a4de26710cb14492ad5a2707a0d66b7b85",
84
- "bytes": 6554
85
- },
86
  "release.json": {
87
- "sha256": "dbdf52f39a7cc345057c3ae088bfddc006226f91e20b42818b8a50a7dca6a6b4",
88
- "bytes": 2920
89
  },
90
  "selection_supervision.json": {
91
- "sha256": "b4badf04f9652bf6f6d1fb16bd96e7b82cb416e50a47b407bd680137ea8f41c9",
92
- "bytes": 732
 
 
 
 
93
  },
94
  "training_status.json": {
95
- "sha256": "fe5363e125bcc6d4521ffa79ded02b5b4534e869e165f9964e1bc1109414b4d1",
96
- "bytes": 123
 
 
 
 
97
  },
98
  "README.md": {
99
- "sha256": "76f47ae63cea1833c0617db22f84b62e185956257ded463e4de5b856d10aeb63",
100
- "bytes": 1414
101
  }
102
  }
 
1
  {
2
+ "checkpoints/epoch_1400.pkl": {
3
+ "sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
4
+ "bytes": 138482745
5
  },
6
+ "checkpoints/epoch_1400.json": {
7
+ "sha256": "3a7a936ab2749d89eea8c986d33063361bac7a9aee7ebe6cbe3058da841b51bf",
8
  "bytes": 7966
9
  },
 
 
 
 
 
 
 
 
10
  "resolved_config.json": {
11
  "sha256": "3e7b7c521a37e9040090b90718d27ec9ea0a66259ce222ed2f6a97c5013880cc",
12
  "bytes": 6617
 
35
  "sha256": "fa560dba6e2c8a2d46a1618f41423b97fac6b8dcadcfd22dc38ca205cd08b524",
36
  "bytes": 82794
37
  },
 
 
 
 
38
  "provenance/native_training_status.json": {
39
  "sha256": "69c9110b3e73ecbe48da0b507acab1ed89b5dbd0ecd58106642ffc403ede3719",
40
  "bytes": 334
41
  },
42
+ "provenance/train_bc_noleak.py": {
43
+ "sha256": "587b66eca4c82c8ee519ca7df23874ca9a9acdbb8cdabd8a36afb68ebcf891d5",
44
+ "bytes": 2186
45
+ },
46
+ "provenance/evaluate_best.py": {
47
+ "sha256": "7be8b5440b354a862228f5a3d3855ec0ba7267c78c4e04158fa028e9633e8d32",
48
+ "bytes": 3370
49
+ },
50
+ "final200/result.json": {
51
+ "sha256": "b05bfd80caebc7eb7d5b80697157ebad136806e47ea213b28c1f025455ef3097",
52
+ "bytes": 119096
53
+ },
54
+ "final200/protocol.json": {
55
+ "sha256": "f7b23c7fd4098b47c65bb77a10a849ec92bb91fa821cfdcf2e4b8b3ad1f78ff0",
56
+ "bytes": 3993
57
+ },
58
+ "final200/epoch_1400.parquet": {
59
+ "sha256": "d6275f867f9734a9fdb74d35d5ee6f16ee9cbc8d75f207f6d2b483f394daa281",
60
+ "bytes": 6510
61
+ },
62
  "evaluation_schedule.json": {
63
  "sha256": "30754536ae01c1ba0c0696c815623c035640798e025ee6ea068e46c53629b0a6",
64
  "bytes": 1009
 
71
  "sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
72
  "bytes": 9987
73
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
74
  "release.json": {
75
+ "sha256": "2b1b7b75eb4b7708d3297a3867597b2dac964ade6498450cf5aadc26d1bb8478",
76
+ "bytes": 3067
77
  },
78
  "selection_supervision.json": {
79
+ "sha256": "d45733d43e9ad628a2b9f9c3983c8a74a5a0e927a266f65ed5182ffc97c04d90",
80
+ "bytes": 861
81
+ },
82
+ "selection_complete.json": {
83
+ "sha256": "47d5a9cdd5794e452aea786b4801c6a6f87ab0e89576e1cf4281327b6d1a3ccb",
84
+ "bytes": 1405
85
  },
86
  "training_status.json": {
87
+ "sha256": "e9f7bc09f87602522f422177c5cffa16e9dc06c0b5a0a544af71a2c6c60c1f77",
88
+ "bytes": 97
89
+ },
90
+ "best.json": {
91
+ "sha256": "9be955f3cc16595e22b5325359f0eb97e94d7dfac0d3402694edc1e99408d430",
92
+ "bytes": 305
93
  },
94
  "README.md": {
95
+ "sha256": "717ae342462a8083ba9c0ee00ea3bf8c8520e8d03fdcceb4b575a8426eb9b577",
96
+ "bytes": 924
97
  }
98
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/best.json CHANGED
@@ -1,157 +1,7 @@
1
  {
2
  "job": "dmg_drawer_cleanup/diverse50",
3
- "stopped": "2026-10-26T20:18:15.212073+00:00",
4
  "reason": "selected_after_joint_review",
5
- "review_note": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
6
- "rule": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
7
- "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
8
- "window": 50,
9
- "evaluation_schedule": {
10
- "schema_version": 1,
11
- "job": "dmg_drawer_cleanup/diverse50",
12
- "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
13
- "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
14
- "recorded_at": "2026-10-26T04:57:55.268092+00:00",
15
- "original_interval_epochs": 30,
16
- "interval_epochs": 50,
17
- "effective_after_epoch": 1350,
18
- "resume_sha256_at_change": "e64fef31c951965087a4ea6118ff08623fce179d1b4aa5ea8a362c0fbf7391fa",
19
- "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
20
- "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
21
- "validation_loss_interval_epochs": 1,
22
- "periodic_episodes": 50,
23
- "original_job_preserved": true,
24
- "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
25
- },
26
- "evaluated_epoch": 2000,
27
- "saved": 40,
28
- "best": {
29
- "epoch": 1050,
30
- "window_success": 0.6066666666666667,
31
- "episodes": 150,
32
- "window_epochs": [
33
- 1020,
34
- 1050,
35
- 1080
36
- ],
37
- "interval_epochs": 30,
38
- "window_radius": 45,
39
- "success_at_epoch": 0.64
40
- },
41
- "blocks": [
42
- {
43
- "epochs": [
44
- 1,
45
- 150
46
- ],
47
- "success": 0.228
48
- },
49
- {
50
- "epochs": [
51
- 151,
52
- 300
53
- ],
54
- "success": 0.404
55
- },
56
- {
57
- "epochs": [
58
- 301,
59
- 450
60
- ],
61
- "success": 0.496
62
- },
63
- {
64
- "epochs": [
65
- 451,
66
- 600
67
- ],
68
- "success": 0.484
69
- },
70
- {
71
- "epochs": [
72
- 601,
73
- 750
74
- ],
75
- "success": 0.564
76
- },
77
- {
78
- "epochs": [
79
- 751,
80
- 900
81
- ],
82
- "success": 0.536
83
- },
84
- {
85
- "epochs": [
86
- 901,
87
- 1050
88
- ],
89
- "success": 0.588
90
- },
91
- {
92
- "epochs": [
93
- 1051,
94
- 1200
95
- ],
96
- "success": 0.556
97
- },
98
- {
99
- "epochs": [
100
- 1201,
101
- 1350
102
- ],
103
- "success": 0.54
104
- },
105
- {
106
- "epochs": [
107
- 1351,
108
- 1500
109
- ],
110
- "success": 0.58
111
- },
112
- {
113
- "epochs": [
114
- 1501,
115
- 1650
116
- ],
117
- "success": 0.5533333333333333
118
- },
119
- {
120
- "epochs": [
121
- 1651,
122
- 1800
123
- ],
124
- "success": 0.5666666666666667
125
- },
126
- {
127
- "epochs": [
128
- 1801,
129
- 1950
130
- ],
131
- "success": 0.5666666666666667
132
- }
133
- ],
134
- "plateau": true,
135
- "loss_trends": {
136
- "train": {
137
- "last_epoch": 2001,
138
- "preceding_50_mean": 0.0558663970049623,
139
- "recent_50_mean": 0.05525940129480686,
140
- "relative_change": -0.010865130788755284,
141
- "complete_100_epoch_window": true
142
- },
143
- "validation": {
144
- "last_epoch": 2001,
145
- "preceding_50_mean": 0.1629074254631996,
146
- "recent_50_mean": 0.16240808933973314,
147
- "relative_change": -0.0030651526291493036,
148
- "complete_100_epoch_window": true
149
- }
150
- },
151
- "validation_min": {
152
- "epoch": 284,
153
- "loss": 0.12474054843187332
154
- },
155
- "validation_at_best": 0.14491789042949677,
156
- "validation_last": 0.1636236011981964
157
  }
 
1
  {
2
  "job": "dmg_drawer_cleanup/diverse50",
 
3
  "reason": "selected_after_joint_review",
4
+ "selected_epoch": 1400,
5
+ "checkpoint_sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
6
+ "rule": "User decision 2026-10-27: a uniform epoch 1400 for all nine locally trained visual BCs."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1400.json ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bc_identity": {
3
+ "architecture": "dice_vision_resnet_flow_unet_v1",
4
+ "family": "StandardBaseBC",
5
+ "framework": "jax"
6
+ },
7
+ "data_manifest_hash": "57937c542aa6790d0dacfb9e3fbae03147553d3ecea84c4b61692944aa942199",
8
+ "encoding": "gzip_pickle_numpy_float32_lossless",
9
+ "kind": "pixel_state_bc",
10
+ "metadata": {
11
+ "epoch": 1400,
12
+ "epochs": 3000,
13
+ "experiment": "dmg_drawer_cleanup",
14
+ "regime": "diverse50",
15
+ "seed": 1,
16
+ "smoke": false,
17
+ "stage": "fixed_epochs",
18
+ "step": 74200,
19
+ "task": "dmg_drawer_cleanup"
20
+ },
21
+ "normalizer_hash": "e224f78a15bbee112e82d8544f735d9253c42dff6c5f5f1f5730100fe3a0b2d2",
22
+ "observation_mode": "pixel_state",
23
+ "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
24
+ "resolved_config": {
25
+ "bc_identity": {
26
+ "family": "StandardBaseBC",
27
+ "framework": "jax"
28
+ },
29
+ "checkpoint_encoding": "lossless_gzip_ema_only",
30
+ "checkpoint_interval_epochs": 50,
31
+ "ema_buffers": "copy_live_running_statistics",
32
+ "evaluation": {
33
+ "episodes": 50,
34
+ "final": {
35
+ "episodes": 200,
36
+ "seed_panel_seed": 20261021,
37
+ "seed_range": [
38
+ 10000000,
39
+ 90000000
40
+ ],
41
+ "seed_sampling": "uniform_without_replacement"
42
+ },
43
+ "interval_epochs": 30,
44
+ "policy_seed": 700001,
45
+ "reset_seed_start": 880000,
46
+ "workers": 8
47
+ },
48
+ "experiment": "dmg_drawer_cleanup",
49
+ "family": "StandardBaseBC",
50
+ "framework": "jax",
51
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
52
+ "index": 0,
53
+ "manifest": "/home/nhy/RL-postTrain/repo_visual/data/bc/versions/dmg_distribution_v1_local_20261025/manifests/dmg_drawer_cleanup.json",
54
+ "manifest_sha256": "110e1f8373c49813ff466b7f24ac25347ae7564ff970efb01556e4fa05e57bba",
55
+ "model_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
56
+ "normalizer_hash": "e224f78a15bbee112e82d8544f735d9253c42dff6c5f5f1f5730100fe3a0b2d2",
57
+ "observation_mode": "pixel_state",
58
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
59
+ "pixel_cache": "/home/nhy/RL-postTrain/repo_visual/data/pixels/dmg_drawer_cleanup/visual_20261025",
60
+ "pixel_spec": {
61
+ "action_dim": 24,
62
+ "architecture": "dice_vision_resnet_flow_unet_v1",
63
+ "backbone_initialization": "random",
64
+ "backbone_norm": "batch_norm",
65
+ "cameras": [
66
+ "agentview",
67
+ "robot0_eye_in_hand",
68
+ "robot1_eye_in_hand"
69
+ ],
70
+ "collision_geoms": false,
71
+ "crop_size": 84,
72
+ "family": "StandardBaseBC",
73
+ "feature_dim": 166,
74
+ "features_per_camera": 64,
75
+ "flow_steps": 10,
76
+ "horizon": 8,
77
+ "image_size": 96,
78
+ "proprio_dim": 38,
79
+ "proprio_keys": [
80
+ "robot0_eef_pos",
81
+ "robot0_eef_quat",
82
+ "robot0_gripper_qpos",
83
+ "robot1_eef_pos",
84
+ "robot1_eef_quat",
85
+ "robot1_gripper_qpos"
86
+ ],
87
+ "task": "dmg_drawer_cleanup",
88
+ "time_embedding_dim": 64
89
+ },
90
+ "regime": "diverse50",
91
+ "runtime": {
92
+ "framework_version": "0.10.1",
93
+ "python": "/home/nhy/RL-postTrain/repo/.venvs/ogpo/bin/python",
94
+ "requested_device": "cuda:0"
95
+ },
96
+ "seed": 1,
97
+ "sources_sha256": {
98
+ "configs/bc_models.yaml": "ce58a903c4270328985ab8677babdaa12748ad18efc3749faed72425024e61c1",
99
+ "configs/bc_pretraining.yaml": "7d1b9be73f25aeb40cd342a3e3ca2543ea865b38b2abe8fb28a7da6588d419a6",
100
+ "configs/bc_pretraining_ph_full.yaml": "a93c7d4a50e9afa836d98a6193ba1bc2fb3827a259cf5de39f5dc7a2504d6728",
101
+ "configs/bc_standard_init_schema.json": "e2c70eee2a1ce8d1a812ec427d24ba3cc45519434331354318d97415fd559bc8",
102
+ "configs/study1.yaml": "35d065e242b205a4c99f8add2eab35bb85f80401f4e07d49f063a0eddf4de608",
103
+ "configs/study1_ph_full.yaml": "f9cebc4b66279d4ea7492f7a80a09f1adca7c1cae7b242374305be4f701e76ef",
104
+ "scripts/pretrain_bc.py": "9972b77f7b8ed964e7e04fa3ba3821b767f0b7a5d697625ed1c6f8fcdf990890",
105
+ "src/rl_pt_analyze/bc_evaluation.py": "7319c346f453263a99cdc5d6fd39e6a0a88783d2c42f68a312ee36ab6d656c6f",
106
+ "src/rl_pt_analyze/bc_fleet.py": "a49c24c66ec0bb1a5548d384deefda429ff1ca693cb19f1c99dbd661148d7ffc",
107
+ "src/rl_pt_analyze/bc_full_cut.py": "a8170de58bc0a16dd5c6358a6429d96afb5227e5701602806d68e42727f671bc",
108
+ "src/rl_pt_analyze/bc_identity.py": "f94c284c71df7ed06eaf4553297de53f2b801a5fac743beb3b98d1bc41314fe0",
109
+ "src/rl_pt_analyze/bc_policy.py": "aff3c4613a0dfc46d6491bdba6478261049c2c9631eca446fa8939a1e498fc30",
110
+ "src/rl_pt_analyze/bc_pretrain.py": "703b2e50fe53d62b23910b907d75c09cec49463e5e9e2c08d8f77d2b6b77a5bd",
111
+ "src/rl_pt_analyze/bc_runtime.py": "29d85733684176bd7328c01921ed1eb45857a914d8ee998e6df4fb0a5be5315d",
112
+ "src/rl_pt_analyze/bc_standard_jax.py": "9063d0d4160e972bb9f7b1acd9e508a2ea1118ceeed7647614c787abcb7b32ce",
113
+ "src/rl_pt_analyze/bc_standard_policy.py": "54d733549ec571f709d49698520f2461def5a20fc4146d155b3d56fbbcf5deb1",
114
+ "src/rl_pt_analyze/bc_storage.py": "83b79d5beaaaa46310d756476f5da90c9227a656533dc9414334d395a6749e54",
115
+ "src/rl_pt_analyze/bc_training.py": "bf20475ceeef114e3444c2610ea6d490389b02d147f22c962e2f63fd94b9dafa",
116
+ "src/rl_pt_analyze/pixel.py": "ec46a0b0e3c40bca7ea2d0c76b086b482fedaf98293b06f7ab719b86b81141e6",
117
+ "src/rl_pt_analyze/pixel_data.py": "74ed81ff034d0782671ff9e48157a3be2d274e7db3735a955705b8e8fcf42b34",
118
+ "src/rl_pt_analyze/pixel_vec.py": "f247a6d9ed1791b82880ecef7505275959835fc3b69bfab2e85dedd096416bc8",
119
+ "src/rl_pt_analyze/r0.py": "e0a13e9c02d935dbab6a0e091fd925f08ce2d00e661b6cf09aff0ee95925249d",
120
+ "third_party/dice/model/diffusion/modules.py": "3b6b2efae0bb67cc30a061a8dd7c6b2d8d6905dac7cecac92b3dbfa5da0067cd",
121
+ "third_party/dice/model/flow_matching/vision_resnet_mlp_flow_matching.py": "2f0d3aa0b6e1624e32a540e66a5803e613481f35bb190926e51560856249981d"
122
+ },
123
+ "spec": {
124
+ "action_dim": 24,
125
+ "architecture": "dice_vision_resnet_flow_unet_v1",
126
+ "backbone_initialization": "random",
127
+ "backbone_norm": "batch_norm",
128
+ "cameras": [
129
+ "agentview",
130
+ "robot0_eye_in_hand",
131
+ "robot1_eye_in_hand"
132
+ ],
133
+ "collision_geoms": false,
134
+ "crop_size": 84,
135
+ "family": "StandardBaseBC",
136
+ "feature_dim": 166,
137
+ "features_per_camera": 64,
138
+ "flow_steps": 10,
139
+ "horizon": 8,
140
+ "image_size": 96,
141
+ "proprio_dim": 38,
142
+ "proprio_keys": [
143
+ "robot0_eef_pos",
144
+ "robot0_eef_quat",
145
+ "robot0_gripper_qpos",
146
+ "robot1_eef_pos",
147
+ "robot1_eef_quat",
148
+ "robot1_gripper_qpos"
149
+ ],
150
+ "task": "dmg_drawer_cleanup",
151
+ "time_embedding_dim": 64
152
+ },
153
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b",
154
+ "task": "dmg_drawer_cleanup",
155
+ "training": {
156
+ "batch_size": 256,
157
+ "bc_only": true,
158
+ "drop_last": false,
159
+ "early_stopping": false,
160
+ "ema_decay": 0.999,
161
+ "epoch_definition": "shuffled_full_training_split_without_replacement",
162
+ "epochs": 3000,
163
+ "learning_rate": 0.0001,
164
+ "online_steps": 0,
165
+ "optimizer": "adamw",
166
+ "padding_loss_mask": true,
167
+ "stopping": "none_fixed_epochs",
168
+ "weight_decay": 1e-06
169
+ },
170
+ "training_samples": 13354,
171
+ "validation": {
172
+ "batch_seed": 771,
173
+ "batch_size": 256,
174
+ "interval_epochs": 1,
175
+ "loss_seed": 882
176
+ },
177
+ "validation_only": false,
178
+ "validation_samples": 1501,
179
+ "wandb": {
180
+ "mode": "offline",
181
+ "project": "standard_bc_pretrain"
182
+ }
183
+ },
184
+ "schema_version": 2,
185
+ "sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
186
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b",
187
+ "weights_used_for_deployment": "complete_model_EMA"
188
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1400.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383
3
+ size 138482745
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/epoch_1400.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6275f867f9734a9fdb74d35d5ee6f16ee9cbc8d75f207f6d2b483f394daa281
3
+ size 6510
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/protocol.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl",
3
- "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
4
  "collision_geoms": false,
5
- "epoch": 1050,
6
  "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
7
  "job": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/jobs/00.json",
8
  "panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
 
1
  {
2
+ "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1400.pkl",
3
+ "checkpoint_sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
4
  "collision_geoms": false,
5
+ "epoch": 1400,
6
  "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
7
  "job": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/jobs/00.json",
8
  "panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/final200/result.json CHANGED
The diff for this file is too large to render. See raw diff
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/provenance/evaluate_best.py ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Evaluate a curve-selected checkpoint on its job's untouched final200 panel.
2
+
3
+ Run from the corresponding frozen checkout. Import JAX only inside main, since
4
+ PixelVecEnv's spawned workers import this entry point before configuring themselves.
5
+ The holdout panel is for reporting, never for reselecting the checkpoint.
6
+ """
7
+ import argparse
8
+ import copy
9
+ import json
10
+ import os
11
+ import sys
12
+ import time
13
+ from pathlib import Path
14
+
15
+
16
+ def main():
17
+ p = argparse.ArgumentParser()
18
+ p.add_argument("--job", required=True, type=Path)
19
+ p.add_argument("--checkpoint", required=True, type=Path)
20
+ p.add_argument("--output", required=True, type=Path)
21
+ args = p.parse_args()
22
+ sys.path.insert(0, str(Path.cwd() / "src"))
23
+ job = json.loads(args.job.read_text())
24
+ os.environ["RL_PT_EXPERIMENT"] = job["experiment"]
25
+ from rl_pt_analyze.bc_pretrain import validate_job, _evaluate
26
+ from rl_pt_analyze.bc_evaluation import evaluation_settings, evaluation_seeds
27
+ from rl_pt_analyze.io import canonical_hash, sha256_file, write_json
28
+ spec = validate_job(job)
29
+ assert spec.collision_geoms is False
30
+ meta = json.loads(args.checkpoint.with_suffix(".json").read_text())
31
+ digest = sha256_file(args.checkpoint)
32
+ assert digest == meta["sha256"]
33
+ assert meta["resolved_config"]["identity_hash"] == job["identity_hash"]
34
+ epoch = meta["metadata"]["epoch"]
35
+ settings = evaluation_settings(job, job["training"]["epochs"])
36
+ seeds = evaluation_seeds(settings)
37
+ periodic = evaluation_seeds(evaluation_settings(job, 0))
38
+ assert len(seeds) == 200 and not set(seeds).intersection(periodic)
39
+ assert not args.output.exists(), f"Refusing to overwrite {args.output}"
40
+ args.output.mkdir(parents=True)
41
+ protocol = dict(job=str(args.job), identity_hash=job["identity_hash"],
42
+ checkpoint=str(args.checkpoint), checkpoint_sha256=digest,
43
+ epoch=epoch, settings=settings, reset_seeds=seeds,
44
+ panel_hash=canonical_hash(seeds), collision_geoms=False,
45
+ selection="frozen before this independent panel was evaluated",
46
+ script_sha256=sha256_file(__file__))
47
+ write_json(args.output / "protocol.json", protocol)
48
+ write_json(args.output / "status.json", dict(status="running", pid=os.getpid()))
49
+ evaluation_job = copy.deepcopy(job)
50
+ evaluation_job["evaluation"] = settings
51
+ evaluation_job["output"] = str(args.output)
52
+ started = time.monotonic()
53
+ try:
54
+ result = _evaluate(args.checkpoint, evaluation_job, epoch)
55
+ assert result["total"] == 200
56
+ assert len(result["successes_per_seed"]) == 200
57
+ assert sum(result["successes_per_seed"]) == result["successes"]
58
+ assert sha256_file(args.checkpoint) == digest
59
+ write_json(args.output / "result.json", dict(protocol=protocol,
60
+ seconds=time.monotonic() - started, **result))
61
+ write_json(args.output / "status.json", dict(status="completed", epoch=epoch,
62
+ success=result["success"], successes=result["successes"], total=200))
63
+ print(json.dumps({k: result[k] for k in ("success", "successes", "total", "ci_low", "ci_high")}), flush=True)
64
+ except BaseException as error:
65
+ write_json(args.output / "status.json", dict(status="failed", error=repr(error)))
66
+ raise
67
+
68
+
69
+ if __name__ == "__main__":
70
+ main()
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/release.json CHANGED
@@ -4,56 +4,63 @@
4
  "experiment": "dmg_drawer_cleanup",
5
  "regime": "diverse50",
6
  "seed": 1,
7
- "epoch": 1050,
8
  "trained_epochs_at_selection": 2001,
9
  "planned_epochs": 3000,
10
- "first_phase_epochs": 2000,
11
  "training_status_at_selection": "selected_after_joint_review",
12
  "family": "StandardBaseBC",
13
  "framework": "jax",
14
  "weights": "complete_model_EMA",
15
  "horizon": 8,
16
  "collision_geoms": false,
17
- "selection_rule": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
18
- "window_success_rate": 0.6066666666666667,
19
- "window_epochs": [
20
- 1020,
21
- 1050,
22
- 1080
23
- ],
24
- "window_episodes": 150,
25
- "selection_note": "Neighborhood score combines different checkpoints on reused periodic seeds; use final200 for this frozen checkpoint performance.",
26
  "final200": {
27
- "success": 0.61,
28
- "successes": 122,
29
  "total": 200,
30
- "ci_low": 0.5409374595716976,
31
- "ci_high": 0.6749165686253038
32
  },
33
  "final200_panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
34
- "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
35
- "checkpoint_bytes": 138495554,
 
 
 
 
 
 
 
 
36
  "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
37
  "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
38
- "path": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl",
39
  "directory": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1",
40
  "supervision": {
41
  "status": "selected_after_joint_review",
42
- "training": "stopped",
43
- "joint_curve_review_status": "finalized_after_phase2000_review",
44
- "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
45
- "review_evidence": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json",
46
- "joint_curve_review_reason": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
47
- "final200_role": "report_only_not_used_for_selection",
48
  "zero_overfitting_claim": false,
49
  "completed_train_epoch": 2001,
50
- "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
 
 
51
  },
52
- "evaluation_interval_at_selected_epoch": 30,
53
- "evaluation_schedule": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/evaluation_schedule.json",
54
  "supersedes_or_reconfirms": {
55
  "epoch": 1050,
56
  "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
57
- "archive": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
58
- }
 
 
 
 
 
 
 
 
59
  }
 
4
  "experiment": "dmg_drawer_cleanup",
5
  "regime": "diverse50",
6
  "seed": 1,
7
+ "epoch": 1400,
8
  "trained_epochs_at_selection": 2001,
9
  "planned_epochs": 3000,
 
10
  "training_status_at_selection": "selected_after_joint_review",
11
  "family": "StandardBaseBC",
12
  "framework": "jax",
13
  "weights": "complete_model_EMA",
14
  "horizon": 8,
15
  "collision_geoms": false,
16
+ "selection_rule": "User decision 2026-10-27: a uniform epoch 1400 for all nine locally trained visual BCs.",
17
+ "selection_note": "Final200 used the same panel as the prior selection and informed the decision, so it is not an independent estimate.",
 
 
 
 
 
 
 
18
  "final200": {
19
+ "success": 0.57,
20
+ "successes": 114,
21
  "total": 200,
22
+ "ci_low": 0.5007048776415268,
23
+ "ci_high": 0.6366567766656557
24
  },
25
  "final200_panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
26
+ "paired_vs_prior_on_final200": {
27
+ "prior_epoch": 1050,
28
+ "prior_successes": 122,
29
+ "e1400_successes": 114,
30
+ "prior_only": 38,
31
+ "e1400_only": 30,
32
+ "exact_mcnemar_p": 0.39612484029702977
33
+ },
34
+ "checkpoint_sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
35
+ "checkpoint_bytes": 138482745,
36
  "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
37
  "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
38
+ "path": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1400.pkl",
39
  "directory": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1",
40
  "supervision": {
41
  "status": "selected_after_joint_review",
42
+ "decision": "user_uniform_e1400_20261027",
43
+ "reviewed_at": "2026-10-27T08:29:29.361939+00:00",
44
+ "joint_curve_review_status": "replaced_by_user_uniform_epoch",
45
+ "joint_curve_review_reason": "User 2026-10-27: use epoch 1400 for every locally owned visual BC, after seeing e1400 vs prior-selection final200 on the same 200 seeds.",
46
+ "final200_role": "decision_evidence_same_panel_as_prior_selection_not_independent",
 
47
  "zero_overfitting_claim": false,
48
  "completed_train_epoch": 2001,
49
+ "training_note": "Trained to the e2000 first-phase boundary; e1400 is an intermediate saved checkpoint.",
50
+ "review_evidence": "visual_no_collision/reviews/20261027_uniform_e1400/comparison.json",
51
+ "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400"
52
  },
 
 
53
  "supersedes_or_reconfirms": {
54
  "epoch": 1050,
55
  "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
56
+ "final200": {
57
+ "success": 0.61,
58
+ "successes": 122,
59
+ "total": 200,
60
+ "ci_low": 0.5409374595716976,
61
+ "ci_high": 0.6749165686253038
62
+ },
63
+ "archive": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400"
64
+ },
65
+ "evaluation_schedule": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/evaluation_schedule.json"
66
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/README.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # DrawerCleanup diverse50: reviewed visual-only BC
2
+
3
+ Selected **epoch1050**, full EMA encoder + flow policy, collision_geoms=false. The2000-epoch training phase is complete.
4
+
5
+ Train mean e951-1000 0.0677247 -> e1001-1050 0.0664753 -> e1051-1100 0.0655113; improvement has slowed to about1.5-1.8% per50 epochs.
6
+
7
+ Validation mean e951-1000 0.143344 -> e1001-1050 0.145048 -> e1051-1100 0.146439; e1801-1950 0.162890. Prefer e1050 to later more overfit weights.
8
+
9
+ Periodic success e1020/1050/1080=62/64/56%, window60.67%. Later150-epoch blocks55.6/54/58/55.33/56.67/56.67%, endpoint56%. No sustained later gain.
10
+
11
+ This is a practical joint curve selection. Validation loss already exceeds its historical minimum; no claim of zero overfitting or globally optimal success is made.
12
+
13
+ Independent final200: **122/200 = 61.0%**, Wilson95% CI [54.1%, 67.5%]. The checkpoint was fixed before evaluating this panel; the result never drives reselection. The unchanged prior report is reused.
14
+
15
+ SHA256: `bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5`.
16
+
17
+ `selection_complete.json` records the decision; `evaluation_schedule.json` and `provenance/` describe the explicit evaluation-cadence change while retaining the original sealed training contract. `SHA256SUMS.json` covers this current release. The exact prior release is preserved under [releases/before_20261026_e1050](releases/before_20261026_e1050/README.md).
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/SHA256SUMS.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checkpoints/epoch_1050.pkl": {
3
+ "sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
4
+ "bytes": 138495554
5
+ },
6
+ "checkpoints/epoch_1050.json": {
7
+ "sha256": "12f4753d1b6471d5bf3d78d68017a325ded95e4c45ea1c4d966a0ff148d48041",
8
+ "bytes": 7966
9
+ },
10
+ "best.json": {
11
+ "sha256": "36a7c38e2d329aceaea261784d39b7f2c3dbe7258127733e020a16b4370d1911",
12
+ "bytes": 3996
13
+ },
14
+ "selection_complete.json": {
15
+ "sha256": "dd7b9eb566691013bf815fe3eb1bd13814e17821a069eeaa377dd31de9a25e13",
16
+ "bytes": 6372
17
+ },
18
+ "resolved_config.json": {
19
+ "sha256": "3e7b7c521a37e9040090b90718d27ec9ea0a66259ce222ed2f6a97c5013880cc",
20
+ "bytes": 6617
21
+ },
22
+ "split.json": {
23
+ "sha256": "726f23d9a89f85d21c63befc3e8b65a6278bb2464c30334117b28f26d3a82b88",
24
+ "bytes": 1178
25
+ },
26
+ "normalizer.json": {
27
+ "sha256": "ba4651b4ed8ba35431fe7c04bcea11395773284bf9f42d55d79054a14abb460d",
28
+ "bytes": 3605
29
+ },
30
+ "data_manifest.json": {
31
+ "sha256": "110e1f8373c49813ff466b7f24ac25347ae7564ff970efb01556e4fa05e57bba",
32
+ "bytes": 9620
33
+ },
34
+ "train_metrics.jsonl": {
35
+ "sha256": "0f158a88fea8855305cafae564d2e3adfa33877b873108ca394fb9dbea9b90b9",
36
+ "bytes": 365163
37
+ },
38
+ "validation_metrics.jsonl": {
39
+ "sha256": "3d264a1ee42c803bcb107f677abd16560827f39765953d1888940c185fbcc57e",
40
+ "bytes": 140235
41
+ },
42
+ "evaluations.jsonl": {
43
+ "sha256": "fa560dba6e2c8a2d46a1618f41423b97fac6b8dcadcfd22dc38ca205cd08b524",
44
+ "bytes": 82794
45
+ },
46
+ "first_phase_complete.json": {
47
+ "sha256": "139f202e3772aa538ab4aa0a3ed0f5202e881b9254b3a8d6e171c5b79618a7de",
48
+ "bytes": 523
49
+ },
50
+ "provenance/native_training_status.json": {
51
+ "sha256": "69c9110b3e73ecbe48da0b507acab1ed89b5dbd0ecd58106642ffc403ede3719",
52
+ "bytes": 334
53
+ },
54
+ "evaluation_schedule.json": {
55
+ "sha256": "30754536ae01c1ba0c0696c815623c035640798e025ee6ea068e46c53629b0a6",
56
+ "bytes": 1009
57
+ },
58
+ "provenance/evaluation_schedule_active.json": {
59
+ "sha256": "3e2dc97b350a15877b0b36837ea81fd26df81cdb0f8c40ab689c6a8bf19b01fe",
60
+ "bytes": 455
61
+ },
62
+ "provenance/effective_train_eval50.py": {
63
+ "sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
64
+ "bytes": 9987
65
+ },
66
+ "provenance/bc_eval_schedule.py": {
67
+ "sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
68
+ "bytes": 3539
69
+ },
70
+ "provenance/train_bc_noleak.py": {
71
+ "sha256": "587b66eca4c82c8ee519ca7df23874ca9a9acdbb8cdabd8a36afb68ebcf891d5",
72
+ "bytes": 2186
73
+ },
74
+ "final200/result.json": {
75
+ "sha256": "61b57c921c11de832c70836f4c1987dbe7dfb4f614b314ca2cc8e2cec02a401c",
76
+ "bytes": 119075
77
+ },
78
+ "final200/protocol.json": {
79
+ "sha256": "b30eeb8b6b735db58d581ea2ac22934431c3f4dd24c50178a370f2f5f5626f07",
80
+ "bytes": 3993
81
+ },
82
+ "final200/epoch_1050.parquet": {
83
+ "sha256": "36e242073f9d8296209dacd2221969a4de26710cb14492ad5a2707a0d66b7b85",
84
+ "bytes": 6554
85
+ },
86
+ "release.json": {
87
+ "sha256": "dbdf52f39a7cc345057c3ae088bfddc006226f91e20b42818b8a50a7dca6a6b4",
88
+ "bytes": 2920
89
+ },
90
+ "selection_supervision.json": {
91
+ "sha256": "b4badf04f9652bf6f6d1fb16bd96e7b82cb416e50a47b407bd680137ea8f41c9",
92
+ "bytes": 732
93
+ },
94
+ "training_status.json": {
95
+ "sha256": "fe5363e125bcc6d4521ffa79ded02b5b4534e869e165f9964e1bc1109414b4d1",
96
+ "bytes": 123
97
+ },
98
+ "README.md": {
99
+ "sha256": "76f47ae63cea1833c0617db22f84b62e185956257ded463e4de5b856d10aeb63",
100
+ "bytes": 1414
101
+ }
102
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/best.json ADDED
@@ -0,0 +1,157 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "job": "dmg_drawer_cleanup/diverse50",
3
+ "stopped": "2026-10-26T20:18:15.212073+00:00",
4
+ "reason": "selected_after_joint_review",
5
+ "review_note": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
6
+ "rule": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
7
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
8
+ "window": 50,
9
+ "evaluation_schedule": {
10
+ "schema_version": 1,
11
+ "job": "dmg_drawer_cleanup/diverse50",
12
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
13
+ "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
14
+ "recorded_at": "2026-10-26T04:57:55.268092+00:00",
15
+ "original_interval_epochs": 30,
16
+ "interval_epochs": 50,
17
+ "effective_after_epoch": 1350,
18
+ "resume_sha256_at_change": "e64fef31c951965087a4ea6118ff08623fce179d1b4aa5ea8a362c0fbf7391fa",
19
+ "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
20
+ "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
21
+ "validation_loss_interval_epochs": 1,
22
+ "periodic_episodes": 50,
23
+ "original_job_preserved": true,
24
+ "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
25
+ },
26
+ "evaluated_epoch": 2000,
27
+ "saved": 40,
28
+ "best": {
29
+ "epoch": 1050,
30
+ "window_success": 0.6066666666666667,
31
+ "episodes": 150,
32
+ "window_epochs": [
33
+ 1020,
34
+ 1050,
35
+ 1080
36
+ ],
37
+ "interval_epochs": 30,
38
+ "window_radius": 45,
39
+ "success_at_epoch": 0.64
40
+ },
41
+ "blocks": [
42
+ {
43
+ "epochs": [
44
+ 1,
45
+ 150
46
+ ],
47
+ "success": 0.228
48
+ },
49
+ {
50
+ "epochs": [
51
+ 151,
52
+ 300
53
+ ],
54
+ "success": 0.404
55
+ },
56
+ {
57
+ "epochs": [
58
+ 301,
59
+ 450
60
+ ],
61
+ "success": 0.496
62
+ },
63
+ {
64
+ "epochs": [
65
+ 451,
66
+ 600
67
+ ],
68
+ "success": 0.484
69
+ },
70
+ {
71
+ "epochs": [
72
+ 601,
73
+ 750
74
+ ],
75
+ "success": 0.564
76
+ },
77
+ {
78
+ "epochs": [
79
+ 751,
80
+ 900
81
+ ],
82
+ "success": 0.536
83
+ },
84
+ {
85
+ "epochs": [
86
+ 901,
87
+ 1050
88
+ ],
89
+ "success": 0.588
90
+ },
91
+ {
92
+ "epochs": [
93
+ 1051,
94
+ 1200
95
+ ],
96
+ "success": 0.556
97
+ },
98
+ {
99
+ "epochs": [
100
+ 1201,
101
+ 1350
102
+ ],
103
+ "success": 0.54
104
+ },
105
+ {
106
+ "epochs": [
107
+ 1351,
108
+ 1500
109
+ ],
110
+ "success": 0.58
111
+ },
112
+ {
113
+ "epochs": [
114
+ 1501,
115
+ 1650
116
+ ],
117
+ "success": 0.5533333333333333
118
+ },
119
+ {
120
+ "epochs": [
121
+ 1651,
122
+ 1800
123
+ ],
124
+ "success": 0.5666666666666667
125
+ },
126
+ {
127
+ "epochs": [
128
+ 1801,
129
+ 1950
130
+ ],
131
+ "success": 0.5666666666666667
132
+ }
133
+ ],
134
+ "plateau": true,
135
+ "loss_trends": {
136
+ "train": {
137
+ "last_epoch": 2001,
138
+ "preceding_50_mean": 0.0558663970049623,
139
+ "recent_50_mean": 0.05525940129480686,
140
+ "relative_change": -0.010865130788755284,
141
+ "complete_100_epoch_window": true
142
+ },
143
+ "validation": {
144
+ "last_epoch": 2001,
145
+ "preceding_50_mean": 0.1629074254631996,
146
+ "recent_50_mean": 0.16240808933973314,
147
+ "relative_change": -0.0030651526291493036,
148
+ "complete_100_epoch_window": true
149
+ }
150
+ },
151
+ "validation_min": {
152
+ "epoch": 284,
153
+ "loss": 0.12474054843187332
154
+ },
155
+ "validation_at_best": 0.14491789042949677,
156
+ "validation_last": 0.1636236011981964
157
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/checkpoints/epoch_1050.json ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bc_identity": {
3
+ "architecture": "dice_vision_resnet_flow_unet_v1",
4
+ "family": "StandardBaseBC",
5
+ "framework": "jax"
6
+ },
7
+ "data_manifest_hash": "57937c542aa6790d0dacfb9e3fbae03147553d3ecea84c4b61692944aa942199",
8
+ "encoding": "gzip_pickle_numpy_float32_lossless",
9
+ "kind": "pixel_state_bc",
10
+ "metadata": {
11
+ "epoch": 1050,
12
+ "epochs": 3000,
13
+ "experiment": "dmg_drawer_cleanup",
14
+ "regime": "diverse50",
15
+ "seed": 1,
16
+ "smoke": false,
17
+ "stage": "fixed_epochs",
18
+ "step": 55650,
19
+ "task": "dmg_drawer_cleanup"
20
+ },
21
+ "normalizer_hash": "e224f78a15bbee112e82d8544f735d9253c42dff6c5f5f1f5730100fe3a0b2d2",
22
+ "observation_mode": "pixel_state",
23
+ "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
24
+ "resolved_config": {
25
+ "bc_identity": {
26
+ "family": "StandardBaseBC",
27
+ "framework": "jax"
28
+ },
29
+ "checkpoint_encoding": "lossless_gzip_ema_only",
30
+ "checkpoint_interval_epochs": 50,
31
+ "ema_buffers": "copy_live_running_statistics",
32
+ "evaluation": {
33
+ "episodes": 50,
34
+ "final": {
35
+ "episodes": 200,
36
+ "seed_panel_seed": 20261021,
37
+ "seed_range": [
38
+ 10000000,
39
+ 90000000
40
+ ],
41
+ "seed_sampling": "uniform_without_replacement"
42
+ },
43
+ "interval_epochs": 30,
44
+ "policy_seed": 700001,
45
+ "reset_seed_start": 880000,
46
+ "workers": 8
47
+ },
48
+ "experiment": "dmg_drawer_cleanup",
49
+ "family": "StandardBaseBC",
50
+ "framework": "jax",
51
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
52
+ "index": 0,
53
+ "manifest": "/home/nhy/RL-postTrain/repo_visual/data/bc/versions/dmg_distribution_v1_local_20261025/manifests/dmg_drawer_cleanup.json",
54
+ "manifest_sha256": "110e1f8373c49813ff466b7f24ac25347ae7564ff970efb01556e4fa05e57bba",
55
+ "model_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
56
+ "normalizer_hash": "e224f78a15bbee112e82d8544f735d9253c42dff6c5f5f1f5730100fe3a0b2d2",
57
+ "observation_mode": "pixel_state",
58
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
59
+ "pixel_cache": "/home/nhy/RL-postTrain/repo_visual/data/pixels/dmg_drawer_cleanup/visual_20261025",
60
+ "pixel_spec": {
61
+ "action_dim": 24,
62
+ "architecture": "dice_vision_resnet_flow_unet_v1",
63
+ "backbone_initialization": "random",
64
+ "backbone_norm": "batch_norm",
65
+ "cameras": [
66
+ "agentview",
67
+ "robot0_eye_in_hand",
68
+ "robot1_eye_in_hand"
69
+ ],
70
+ "collision_geoms": false,
71
+ "crop_size": 84,
72
+ "family": "StandardBaseBC",
73
+ "feature_dim": 166,
74
+ "features_per_camera": 64,
75
+ "flow_steps": 10,
76
+ "horizon": 8,
77
+ "image_size": 96,
78
+ "proprio_dim": 38,
79
+ "proprio_keys": [
80
+ "robot0_eef_pos",
81
+ "robot0_eef_quat",
82
+ "robot0_gripper_qpos",
83
+ "robot1_eef_pos",
84
+ "robot1_eef_quat",
85
+ "robot1_gripper_qpos"
86
+ ],
87
+ "task": "dmg_drawer_cleanup",
88
+ "time_embedding_dim": 64
89
+ },
90
+ "regime": "diverse50",
91
+ "runtime": {
92
+ "framework_version": "0.10.1",
93
+ "python": "/home/nhy/RL-postTrain/repo/.venvs/ogpo/bin/python",
94
+ "requested_device": "cuda:0"
95
+ },
96
+ "seed": 1,
97
+ "sources_sha256": {
98
+ "configs/bc_models.yaml": "ce58a903c4270328985ab8677babdaa12748ad18efc3749faed72425024e61c1",
99
+ "configs/bc_pretraining.yaml": "7d1b9be73f25aeb40cd342a3e3ca2543ea865b38b2abe8fb28a7da6588d419a6",
100
+ "configs/bc_pretraining_ph_full.yaml": "a93c7d4a50e9afa836d98a6193ba1bc2fb3827a259cf5de39f5dc7a2504d6728",
101
+ "configs/bc_standard_init_schema.json": "e2c70eee2a1ce8d1a812ec427d24ba3cc45519434331354318d97415fd559bc8",
102
+ "configs/study1.yaml": "35d065e242b205a4c99f8add2eab35bb85f80401f4e07d49f063a0eddf4de608",
103
+ "configs/study1_ph_full.yaml": "f9cebc4b66279d4ea7492f7a80a09f1adca7c1cae7b242374305be4f701e76ef",
104
+ "scripts/pretrain_bc.py": "9972b77f7b8ed964e7e04fa3ba3821b767f0b7a5d697625ed1c6f8fcdf990890",
105
+ "src/rl_pt_analyze/bc_evaluation.py": "7319c346f453263a99cdc5d6fd39e6a0a88783d2c42f68a312ee36ab6d656c6f",
106
+ "src/rl_pt_analyze/bc_fleet.py": "a49c24c66ec0bb1a5548d384deefda429ff1ca693cb19f1c99dbd661148d7ffc",
107
+ "src/rl_pt_analyze/bc_full_cut.py": "a8170de58bc0a16dd5c6358a6429d96afb5227e5701602806d68e42727f671bc",
108
+ "src/rl_pt_analyze/bc_identity.py": "f94c284c71df7ed06eaf4553297de53f2b801a5fac743beb3b98d1bc41314fe0",
109
+ "src/rl_pt_analyze/bc_policy.py": "aff3c4613a0dfc46d6491bdba6478261049c2c9631eca446fa8939a1e498fc30",
110
+ "src/rl_pt_analyze/bc_pretrain.py": "703b2e50fe53d62b23910b907d75c09cec49463e5e9e2c08d8f77d2b6b77a5bd",
111
+ "src/rl_pt_analyze/bc_runtime.py": "29d85733684176bd7328c01921ed1eb45857a914d8ee998e6df4fb0a5be5315d",
112
+ "src/rl_pt_analyze/bc_standard_jax.py": "9063d0d4160e972bb9f7b1acd9e508a2ea1118ceeed7647614c787abcb7b32ce",
113
+ "src/rl_pt_analyze/bc_standard_policy.py": "54d733549ec571f709d49698520f2461def5a20fc4146d155b3d56fbbcf5deb1",
114
+ "src/rl_pt_analyze/bc_storage.py": "83b79d5beaaaa46310d756476f5da90c9227a656533dc9414334d395a6749e54",
115
+ "src/rl_pt_analyze/bc_training.py": "bf20475ceeef114e3444c2610ea6d490389b02d147f22c962e2f63fd94b9dafa",
116
+ "src/rl_pt_analyze/pixel.py": "ec46a0b0e3c40bca7ea2d0c76b086b482fedaf98293b06f7ab719b86b81141e6",
117
+ "src/rl_pt_analyze/pixel_data.py": "74ed81ff034d0782671ff9e48157a3be2d274e7db3735a955705b8e8fcf42b34",
118
+ "src/rl_pt_analyze/pixel_vec.py": "f247a6d9ed1791b82880ecef7505275959835fc3b69bfab2e85dedd096416bc8",
119
+ "src/rl_pt_analyze/r0.py": "e0a13e9c02d935dbab6a0e091fd925f08ce2d00e661b6cf09aff0ee95925249d",
120
+ "third_party/dice/model/diffusion/modules.py": "3b6b2efae0bb67cc30a061a8dd7c6b2d8d6905dac7cecac92b3dbfa5da0067cd",
121
+ "third_party/dice/model/flow_matching/vision_resnet_mlp_flow_matching.py": "2f0d3aa0b6e1624e32a540e66a5803e613481f35bb190926e51560856249981d"
122
+ },
123
+ "spec": {
124
+ "action_dim": 24,
125
+ "architecture": "dice_vision_resnet_flow_unet_v1",
126
+ "backbone_initialization": "random",
127
+ "backbone_norm": "batch_norm",
128
+ "cameras": [
129
+ "agentview",
130
+ "robot0_eye_in_hand",
131
+ "robot1_eye_in_hand"
132
+ ],
133
+ "collision_geoms": false,
134
+ "crop_size": 84,
135
+ "family": "StandardBaseBC",
136
+ "feature_dim": 166,
137
+ "features_per_camera": 64,
138
+ "flow_steps": 10,
139
+ "horizon": 8,
140
+ "image_size": 96,
141
+ "proprio_dim": 38,
142
+ "proprio_keys": [
143
+ "robot0_eef_pos",
144
+ "robot0_eef_quat",
145
+ "robot0_gripper_qpos",
146
+ "robot1_eef_pos",
147
+ "robot1_eef_quat",
148
+ "robot1_gripper_qpos"
149
+ ],
150
+ "task": "dmg_drawer_cleanup",
151
+ "time_embedding_dim": 64
152
+ },
153
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b",
154
+ "task": "dmg_drawer_cleanup",
155
+ "training": {
156
+ "batch_size": 256,
157
+ "bc_only": true,
158
+ "drop_last": false,
159
+ "early_stopping": false,
160
+ "ema_decay": 0.999,
161
+ "epoch_definition": "shuffled_full_training_split_without_replacement",
162
+ "epochs": 3000,
163
+ "learning_rate": 0.0001,
164
+ "online_steps": 0,
165
+ "optimizer": "adamw",
166
+ "padding_loss_mask": true,
167
+ "stopping": "none_fixed_epochs",
168
+ "weight_decay": 1e-06
169
+ },
170
+ "training_samples": 13354,
171
+ "validation": {
172
+ "batch_seed": 771,
173
+ "batch_size": 256,
174
+ "interval_epochs": 1,
175
+ "loss_seed": 882
176
+ },
177
+ "validation_only": false,
178
+ "validation_samples": 1501,
179
+ "wandb": {
180
+ "mode": "offline",
181
+ "project": "standard_bc_pretrain"
182
+ }
183
+ },
184
+ "schema_version": 2,
185
+ "sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
186
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b",
187
+ "weights_used_for_deployment": "complete_model_EMA"
188
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/checkpoints/epoch_1050.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5
3
+ size 138495554
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/data_manifest.json ADDED
@@ -0,0 +1,436 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "datasets": {
3
+ "diverse50": {
4
+ "path": "data/bc/versions/dmg_distribution_v1_local_20261025/dmg_drawer_cleanup/diverse50.hdf5",
5
+ "sha256": "e8eb5f8c69ae6506c2318b387ead72aee1dc8ecadf0b781fda2c8e00cd342bbc",
6
+ "train_transitions": 13354,
7
+ "validation_transitions": 1501
8
+ },
9
+ "focused50": {
10
+ "path": "data/bc/versions/dmg_distribution_v1_local_20261025/dmg_drawer_cleanup/focused50.hdf5",
11
+ "sha256": "0477bf9f3cc845178dcb4a0fb7c042a9120578eb82e65fd020e630d7d51c1878",
12
+ "train_transitions": 13725,
13
+ "validation_transitions": 1525
14
+ },
15
+ "mixed50": {
16
+ "path": "data/bc/versions/dmg_distribution_v1_local_20261025/dmg_drawer_cleanup/mixed50.hdf5",
17
+ "sha256": "adcd9316c1bd3f3bb53dcd464071b048455af3598c01008fc2e965a91390fc93",
18
+ "train_transitions": 13533,
19
+ "validation_transitions": 1558
20
+ }
21
+ },
22
+ "groups": {},
23
+ "hash": "57937c542aa6790d0dacfb9e3fbae03147553d3ecea84c4b61692944aa942199",
24
+ "mixed_components": {
25
+ "diverse": [
26
+ "demo_102",
27
+ "demo_105",
28
+ "demo_145",
29
+ "demo_241",
30
+ "demo_252",
31
+ "demo_320",
32
+ "demo_356",
33
+ "demo_376",
34
+ "demo_407",
35
+ "demo_461",
36
+ "demo_465",
37
+ "demo_475",
38
+ "demo_515",
39
+ "demo_521",
40
+ "demo_580",
41
+ "demo_640",
42
+ "demo_643",
43
+ "demo_649",
44
+ "demo_664",
45
+ "demo_674",
46
+ "demo_758",
47
+ "demo_847",
48
+ "demo_879",
49
+ "demo_923",
50
+ "demo_992"
51
+ ],
52
+ "focused": [
53
+ "demo_50",
54
+ "demo_66",
55
+ "demo_136",
56
+ "demo_169",
57
+ "demo_280",
58
+ "demo_282",
59
+ "demo_307",
60
+ "demo_318",
61
+ "demo_378",
62
+ "demo_381",
63
+ "demo_436",
64
+ "demo_453",
65
+ "demo_485",
66
+ "demo_599",
67
+ "demo_657",
68
+ "demo_685",
69
+ "demo_712",
70
+ "demo_744",
71
+ "demo_802",
72
+ "demo_845",
73
+ "demo_866",
74
+ "demo_959",
75
+ "demo_970",
76
+ "demo_975",
77
+ "demo_1023"
78
+ ]
79
+ },
80
+ "parent_manifest": "data/manifests/dmg_distribution_v1_local_20261025/dmg_drawer_cleanup.json",
81
+ "parent_manifest_sha256": "d8c3631e15dc38a6106d27342e5768005a6d2494102fd4bde377a53e207c6fe2",
82
+ "protocol": "dmg_distribution_v1_local_20261025",
83
+ "regimes": {
84
+ "diverse50": {
85
+ "mg": [
86
+ "demo_5",
87
+ "demo_49",
88
+ "demo_53",
89
+ "demo_85",
90
+ "demo_102",
91
+ "demo_105",
92
+ "demo_140",
93
+ "demo_145",
94
+ "demo_241",
95
+ "demo_252",
96
+ "demo_255",
97
+ "demo_320",
98
+ "demo_356",
99
+ "demo_364",
100
+ "demo_376",
101
+ "demo_407",
102
+ "demo_461",
103
+ "demo_465",
104
+ "demo_475",
105
+ "demo_481",
106
+ "demo_515",
107
+ "demo_521",
108
+ "demo_551",
109
+ "demo_556",
110
+ "demo_580",
111
+ "demo_586",
112
+ "demo_636",
113
+ "demo_637",
114
+ "demo_640",
115
+ "demo_643",
116
+ "demo_649",
117
+ "demo_664",
118
+ "demo_674",
119
+ "demo_696",
120
+ "demo_742",
121
+ "demo_758",
122
+ "demo_791",
123
+ "demo_797",
124
+ "demo_844",
125
+ "demo_847",
126
+ "demo_850",
127
+ "demo_879",
128
+ "demo_880",
129
+ "demo_896",
130
+ "demo_912",
131
+ "demo_923",
132
+ "demo_942",
133
+ "demo_992",
134
+ "demo_1001",
135
+ "demo_1014"
136
+ ]
137
+ },
138
+ "focused50": {
139
+ "mg": [
140
+ "demo_50",
141
+ "demo_66",
142
+ "demo_120",
143
+ "demo_136",
144
+ "demo_167",
145
+ "demo_169",
146
+ "demo_218",
147
+ "demo_226",
148
+ "demo_234",
149
+ "demo_280",
150
+ "demo_282",
151
+ "demo_307",
152
+ "demo_318",
153
+ "demo_333",
154
+ "demo_370",
155
+ "demo_377",
156
+ "demo_378",
157
+ "demo_381",
158
+ "demo_425",
159
+ "demo_428",
160
+ "demo_436",
161
+ "demo_453",
162
+ "demo_485",
163
+ "demo_532",
164
+ "demo_540",
165
+ "demo_599",
166
+ "demo_621",
167
+ "demo_657",
168
+ "demo_660",
169
+ "demo_667",
170
+ "demo_681",
171
+ "demo_682",
172
+ "demo_685",
173
+ "demo_712",
174
+ "demo_718",
175
+ "demo_727",
176
+ "demo_744",
177
+ "demo_776",
178
+ "demo_782",
179
+ "demo_802",
180
+ "demo_845",
181
+ "demo_866",
182
+ "demo_895",
183
+ "demo_926",
184
+ "demo_959",
185
+ "demo_970",
186
+ "demo_975",
187
+ "demo_1009",
188
+ "demo_1016",
189
+ "demo_1023"
190
+ ]
191
+ },
192
+ "mixed50": {
193
+ "mg": [
194
+ "demo_50",
195
+ "demo_66",
196
+ "demo_102",
197
+ "demo_105",
198
+ "demo_136",
199
+ "demo_145",
200
+ "demo_169",
201
+ "demo_241",
202
+ "demo_252",
203
+ "demo_280",
204
+ "demo_282",
205
+ "demo_307",
206
+ "demo_318",
207
+ "demo_320",
208
+ "demo_356",
209
+ "demo_376",
210
+ "demo_378",
211
+ "demo_381",
212
+ "demo_407",
213
+ "demo_436",
214
+ "demo_453",
215
+ "demo_461",
216
+ "demo_465",
217
+ "demo_475",
218
+ "demo_485",
219
+ "demo_515",
220
+ "demo_521",
221
+ "demo_580",
222
+ "demo_599",
223
+ "demo_640",
224
+ "demo_643",
225
+ "demo_649",
226
+ "demo_657",
227
+ "demo_664",
228
+ "demo_674",
229
+ "demo_685",
230
+ "demo_712",
231
+ "demo_744",
232
+ "demo_758",
233
+ "demo_802",
234
+ "demo_845",
235
+ "demo_847",
236
+ "demo_866",
237
+ "demo_879",
238
+ "demo_923",
239
+ "demo_959",
240
+ "demo_970",
241
+ "demo_975",
242
+ "demo_992",
243
+ "demo_1023"
244
+ ]
245
+ }
246
+ },
247
+ "schema_version": 6,
248
+ "sources": {
249
+ "mg": {
250
+ "path": "data/dexmimicgen/local_20261025/dmg_drawer_cleanup/mg/low_dim.hdf5",
251
+ "sha256": "56e19fbe82043792c298c470e1405b227d6a350bf6794b5f5ce167f10916aae5"
252
+ }
253
+ },
254
+ "split_contract": {
255
+ "cohort": 50,
256
+ "normalizer": "training_45_only",
257
+ "seed": 73119,
258
+ "train": 45,
259
+ "unit": "whole_trajectory",
260
+ "validation": 5
261
+ },
262
+ "splits": {
263
+ "diverse50": {
264
+ "training": [
265
+ "mg/demo_923",
266
+ "mg/demo_53",
267
+ "mg/demo_649",
268
+ "mg/demo_49",
269
+ "mg/demo_850",
270
+ "mg/demo_1014",
271
+ "mg/demo_461",
272
+ "mg/demo_252",
273
+ "mg/demo_465",
274
+ "mg/demo_586",
275
+ "mg/demo_551",
276
+ "mg/demo_844",
277
+ "mg/demo_364",
278
+ "mg/demo_637",
279
+ "mg/demo_376",
280
+ "mg/demo_475",
281
+ "mg/demo_255",
282
+ "mg/demo_5",
283
+ "mg/demo_742",
284
+ "mg/demo_241",
285
+ "mg/demo_847",
286
+ "mg/demo_105",
287
+ "mg/demo_356",
288
+ "mg/demo_674",
289
+ "mg/demo_580",
290
+ "mg/demo_879",
291
+ "mg/demo_515",
292
+ "mg/demo_643",
293
+ "mg/demo_320",
294
+ "mg/demo_145",
295
+ "mg/demo_758",
296
+ "mg/demo_640",
297
+ "mg/demo_481",
298
+ "mg/demo_696",
299
+ "mg/demo_140",
300
+ "mg/demo_880",
301
+ "mg/demo_102",
302
+ "mg/demo_896",
303
+ "mg/demo_636",
304
+ "mg/demo_992",
305
+ "mg/demo_797",
306
+ "mg/demo_942",
307
+ "mg/demo_664",
308
+ "mg/demo_791",
309
+ "mg/demo_1001"
310
+ ],
311
+ "validation": [
312
+ "mg/demo_85",
313
+ "mg/demo_556",
314
+ "mg/demo_407",
315
+ "mg/demo_521",
316
+ "mg/demo_912"
317
+ ]
318
+ },
319
+ "focused50": {
320
+ "training": [
321
+ "mg/demo_959",
322
+ "mg/demo_453",
323
+ "mg/demo_660",
324
+ "mg/demo_381",
325
+ "mg/demo_802",
326
+ "mg/demo_1016",
327
+ "mg/demo_333",
328
+ "mg/demo_218",
329
+ "mg/demo_370",
330
+ "mg/demo_540",
331
+ "mg/demo_485",
332
+ "mg/demo_744",
333
+ "mg/demo_282",
334
+ "mg/demo_621",
335
+ "mg/demo_307",
336
+ "mg/demo_377",
337
+ "mg/demo_226",
338
+ "mg/demo_425",
339
+ "mg/demo_685",
340
+ "mg/demo_169",
341
+ "mg/demo_776",
342
+ "mg/demo_120",
343
+ "mg/demo_280",
344
+ "mg/demo_681",
345
+ "mg/demo_532",
346
+ "mg/demo_845",
347
+ "mg/demo_428",
348
+ "mg/demo_66",
349
+ "mg/demo_234",
350
+ "mg/demo_167",
351
+ "mg/demo_712",
352
+ "mg/demo_657",
353
+ "mg/demo_378",
354
+ "mg/demo_682",
355
+ "mg/demo_136",
356
+ "mg/demo_866",
357
+ "mg/demo_1023",
358
+ "mg/demo_895",
359
+ "mg/demo_599",
360
+ "mg/demo_975",
361
+ "mg/demo_727",
362
+ "mg/demo_970",
363
+ "mg/demo_667",
364
+ "mg/demo_718",
365
+ "mg/demo_1009"
366
+ ],
367
+ "validation": [
368
+ "mg/demo_782",
369
+ "mg/demo_50",
370
+ "mg/demo_318",
371
+ "mg/demo_436",
372
+ "mg/demo_926"
373
+ ]
374
+ },
375
+ "mixed50": {
376
+ "training": [
377
+ "mg/demo_970",
378
+ "mg/demo_475",
379
+ "mg/demo_649",
380
+ "mg/demo_436",
381
+ "mg/demo_847",
382
+ "mg/demo_1023",
383
+ "mg/demo_376",
384
+ "mg/demo_252",
385
+ "mg/demo_378",
386
+ "mg/demo_521",
387
+ "mg/demo_485",
388
+ "mg/demo_758",
389
+ "mg/demo_318",
390
+ "mg/demo_599",
391
+ "mg/demo_320",
392
+ "mg/demo_381",
393
+ "mg/demo_280",
394
+ "mg/demo_453",
395
+ "mg/demo_674",
396
+ "mg/demo_241",
397
+ "mg/demo_802",
398
+ "mg/demo_136",
399
+ "mg/demo_307",
400
+ "mg/demo_66",
401
+ "mg/demo_515",
402
+ "mg/demo_866",
403
+ "mg/demo_461",
404
+ "mg/demo_643",
405
+ "mg/demo_282",
406
+ "mg/demo_169",
407
+ "mg/demo_685",
408
+ "mg/demo_640",
409
+ "mg/demo_407",
410
+ "mg/demo_664",
411
+ "mg/demo_145",
412
+ "mg/demo_879",
413
+ "mg/demo_105",
414
+ "mg/demo_923",
415
+ "mg/demo_580",
416
+ "mg/demo_992",
417
+ "mg/demo_744",
418
+ "mg/demo_975",
419
+ "mg/demo_657",
420
+ "mg/demo_712",
421
+ "mg/demo_102"
422
+ ],
423
+ "validation": [
424
+ "mg/demo_845",
425
+ "mg/demo_50",
426
+ "mg/demo_356",
427
+ "mg/demo_465",
428
+ "mg/demo_959"
429
+ ]
430
+ }
431
+ },
432
+ "status": "accepted_for_bc",
433
+ "task": "dmg_drawer_cleanup",
434
+ "task_spec_hash": "f146205010a77e7a5fc5d0e155a5ccfa2a15cbdb0c87c2925b539a5c4a020b3d",
435
+ "unused_mg": []
436
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/evaluation_schedule.json ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "job": "dmg_drawer_cleanup/diverse50",
4
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
5
+ "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
6
+ "recorded_at": "2026-10-26T04:57:55.268092+00:00",
7
+ "original_interval_epochs": 30,
8
+ "interval_epochs": 50,
9
+ "effective_after_epoch": 1350,
10
+ "resume_sha256_at_change": "e64fef31c951965087a4ea6118ff08623fce179d1b4aa5ea8a362c0fbf7391fa",
11
+ "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
12
+ "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
13
+ "validation_loss_interval_epochs": 1,
14
+ "periodic_episodes": 50,
15
+ "original_job_preserved": true,
16
+ "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
17
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/evaluations.jsonl ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"checkpoint": null, "checkpoint_sha256": "3150b9d0268888b74d70853f54dabb27437934704eae7f8194f0866d29153aea", "ci": "wilson95", "ci_high": 0.07134759913335872, "ci_low": 0.0, "environment_steps": 27500, "epoch": 30, "eval_seconds": 164.5955018139939, "length": 550.0, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_30.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 1590, "success": 0.0, "successes": 0, "successes_per_seed": [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false], "total": 50}
2
+ {"checkpoint": null, "checkpoint_sha256": "6726ba9b05677d24ae0828a42f64542c9180194d898b7b613af192d4f73d8234", "ci": "wilson95", "ci_high": 0.1346009068750702, "ci_low": 0.011038884327619805, "environment_steps": 26958, "epoch": 60, "eval_seconds": 173.11066238899366, "length": 539.16, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_60.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 3180, "success": 0.04, "successes": 2, "successes_per_seed": [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, true, false, false, false, false, false, false, false, false, false, false, false, true, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false], "total": 50}
3
+ {"checkpoint": null, "checkpoint_sha256": "fceb46432a2f33cc4c4d896b58b379003c605a0f7bc7bfc707a1c06bdc02ad74", "ci": "wilson95", "ci_high": 0.478461739497674, "ci_low": 0.2243694922250009, "environment_steps": 23601, "epoch": 90, "eval_seconds": 229.4806621030002, "length": 472.02, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_90.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 4770, "success": 0.34, "successes": 17, "successes_per_seed": [true, false, false, false, true, false, false, false, false, false, false, false, true, true, false, false, false, false, false, false, false, false, true, false, true, true, false, true, false, false, false, false, false, false, true, false, false, true, true, false, true, false, false, true, false, true, true, false, true, true], "total": 50}
4
+ {"checkpoint": null, "checkpoint_sha256": "c9d7de9829c775e7baa541e4067a13e00ce0586042b02e2b228c555bec8eca2f", "ci": "wilson95", "ci_high": 0.437503504644534, "ci_low": 0.19103553500880954, "environment_steps": 23628, "epoch": 120, "eval_seconds": 205.30983765400015, "length": 472.56, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_120.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 6360, "success": 0.3, "successes": 15, "successes_per_seed": [false, false, false, true, false, false, false, false, false, false, false, false, true, false, false, false, false, true, false, false, false, false, true, false, false, false, false, true, false, false, false, true, false, false, false, true, true, true, true, false, false, true, false, true, false, false, true, true, false, true], "total": 50}
5
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_150.pkl", "checkpoint_sha256": "8ddcbfc7ae81bb62157034f6e78a317f6ba112954fa528dd7c0f26d96405943c", "ci": "wilson95", "ci_high": 0.5960112860084253, "ci_low": 0.3296965219222435, "environment_steps": 21385, "epoch": 150, "eval_seconds": 205.5440882249968, "length": 427.7, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_150.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 7950, "success": 0.46, "successes": 23, "successes_per_seed": [false, true, false, true, false, true, false, false, false, true, true, false, true, true, false, false, false, false, true, false, true, true, true, false, false, true, false, true, true, false, true, false, true, false, false, true, false, false, true, false, false, false, false, true, true, true, true, false, true, false], "total": 50}
6
+ {"checkpoint": null, "checkpoint_sha256": "ab8bdba62603d8347a965487ab32c68491f2274f49948ccd361598d1110f2d5b", "ci": "wilson95", "ci_high": 0.5184955306900703, "ci_low": 0.25862789310193596, "environment_steps": 22362, "epoch": 180, "eval_seconds": 210.63276211501216, "length": 447.24, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_180.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 9540, "success": 0.38, "successes": 19, "successes_per_seed": [false, false, false, true, false, true, false, false, true, false, true, true, false, true, false, true, false, false, false, true, false, true, true, false, false, true, false, true, true, false, false, true, false, false, true, false, true, false, true, false, false, false, false, true, false, false, true, false, false, false], "total": 50}
7
+ {"checkpoint": null, "checkpoint_sha256": "e802c885868083874a44c3d6fd5234f6066fc0ede6f34ca7e96be16c9c194fc3", "ci": "wilson95", "ci_high": 0.5769397197834315, "ci_low": 0.3116219921125717, "environment_steps": 21834, "epoch": 210, "eval_seconds": 228.04193008999573, "length": 436.68, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_210.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 11130, "success": 0.44, "successes": 22, "successes_per_seed": [false, true, false, false, false, false, false, true, true, true, false, false, false, true, false, false, true, true, false, true, false, true, false, false, true, false, true, true, false, true, true, false, false, true, true, false, false, false, true, false, false, true, true, true, false, false, true, false, true, false], "total": 50}
8
+ {"checkpoint": null, "checkpoint_sha256": "2e3797a00b5ad9322644b65a499ae60e3fffd7c86cc557042951a9459312b76c", "ci": "wilson95", "ci_high": 0.49858983123887307, "ci_low": 0.24138749651846741, "environment_steps": 22776, "epoch": 240, "eval_seconds": 199.59921601600945, "length": 455.52, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_240.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 12720, "success": 0.36, "successes": 18, "successes_per_seed": [false, false, false, true, false, false, false, false, false, true, false, true, false, false, false, false, true, true, false, true, false, false, false, false, false, true, false, false, false, true, true, true, false, true, true, false, false, false, true, false, true, false, false, false, true, false, true, false, true, true], "total": 50}
9
+ {"checkpoint": null, "checkpoint_sha256": "fb208382b96619b02a5ce48c41a152d1f4188a37877619d7007f9689a05fe39d", "ci": "wilson95", "ci_high": 0.5381856225241064, "ci_low": 0.27608389730256544, "environment_steps": 22210, "epoch": 270, "eval_seconds": 224.39349415700417, "length": 444.2, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_270.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 14310, "success": 0.4, "successes": 20, "successes_per_seed": [false, false, false, false, true, true, false, false, false, true, false, true, false, true, false, false, true, true, false, false, true, true, false, true, false, true, false, false, true, true, true, true, false, false, true, false, false, false, false, false, true, true, true, false, false, false, true, false, false, false], "total": 50}
10
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_300.pkl", "checkpoint_sha256": "4b1deaefc770d863f7b9d03f6cf1bae4ced1ec26c08d2183236f23b7300e080f", "ci": "wilson95", "ci_high": 0.5769397197834315, "ci_low": 0.3116219921125717, "environment_steps": 21533, "epoch": 300, "eval_seconds": 217.7945239249966, "length": 430.66, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_300.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 15900, "success": 0.44, "successes": 22, "successes_per_seed": [false, false, false, false, false, false, false, true, true, true, false, true, false, true, false, false, true, false, true, false, false, false, false, true, true, true, false, true, false, true, true, true, false, true, true, false, false, false, true, false, false, false, true, true, true, false, true, false, false, true], "total": 50}
11
+ {"checkpoint": null, "checkpoint_sha256": "e7d6f85fa0d6e4fd87efd5cf72aa8667d04be1f6b8a16ace4e2125ec7a69af0a", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20362, "epoch": 330, "eval_seconds": 213.58882673000335, "length": 407.24, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_330.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 17490, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, false, true, true, true, true, true, false, true, false, true, false, false, true, true, true, true, false, true, true, false, false, true, false, false, false, false, false, true, true, true, true, false, true, true, true, false, false, false, false, true, true, false, false, true, true, true], "total": 50}
12
+ {"checkpoint": null, "checkpoint_sha256": "21266eb65a69145d3b0920266ddaa08b13e35c0eea7d6016b94ac86dcbfacdf6", "ci": "wilson95", "ci_high": 0.5960112860084253, "ci_low": 0.3296965219222435, "environment_steps": 21444, "epoch": 360, "eval_seconds": 204.30266433698125, "length": 428.88, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_360.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 19080, "success": 0.46, "successes": 23, "successes_per_seed": [false, false, false, false, true, false, true, false, false, true, true, true, false, true, false, false, true, false, false, false, false, false, true, true, false, true, false, false, false, false, true, true, true, true, true, true, true, false, false, false, false, false, false, true, true, false, true, true, true, true], "total": 50}
13
+ {"checkpoint": null, "checkpoint_sha256": "5acacabd5ec5891be03f972af4b93650e15dd9d6559546954c8bd801e28ef358", "ci": "wilson95", "ci_high": 0.5184955306900703, "ci_low": 0.25862789310193596, "environment_steps": 22619, "epoch": 390, "eval_seconds": 201.79164686700096, "length": 452.38, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_390.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 20670, "success": 0.38, "successes": 19, "successes_per_seed": [false, false, false, true, true, true, false, true, false, true, false, true, false, true, false, false, false, true, false, true, false, true, false, false, false, true, false, true, false, false, true, false, false, true, false, false, true, false, false, false, false, true, false, true, true, false, true, false, false, false], "total": 50}
14
+ {"checkpoint": null, "checkpoint_sha256": "7b417bd87dd82a8ee8f247f13436519065d391f82be05efe60ddee4cd3256fba", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19683, "epoch": 420, "eval_seconds": 196.62864280701615, "length": 393.66, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_420.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 22260, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, true, true, true, true, true, false, true, true, false, true, false, false, true, false, true, true, true, true, false, true, false, false, false, true, true, false, true, false, true, true, false, false, true, false, true, false, true, false, true, true, false, true, true, true, true, true, true, false], "total": 50}
15
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_450.pkl", "checkpoint_sha256": "c948fa47576462a1ffa00ef5d11f53bf79f3c34afaebfb784be6188f33eac028", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 20832, "epoch": 450, "eval_seconds": 189.1360748510051, "length": 416.64, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_450.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 23850, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, true, true, true, false, false, true, false, false, true, false, true, true, false, false, false, true, true, false, false, true, false, true, true, false, true, false, false, true, true, true, false, true, false, true, false, true, false, false, false, false, true, true, true, true, false, true, true], "total": 50}
16
+ {"checkpoint": null, "checkpoint_sha256": "be02a9d19b1644dc51539dda520c0d2f4a334a8912387e9567f724e98fdfd434", "ci": "wilson95", "ci_high": 0.5769397197834315, "ci_low": 0.3116219921125717, "environment_steps": 21726, "epoch": 480, "eval_seconds": 204.54176059100428, "length": 434.52, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_480.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 25440, "success": 0.44, "successes": 22, "successes_per_seed": [false, true, false, true, true, true, false, false, true, true, true, false, false, false, true, true, true, true, false, true, false, false, false, true, false, true, false, true, true, false, false, true, false, false, false, false, true, false, false, false, true, true, false, true, true, false, false, false, false, false], "total": 50}
17
+ {"checkpoint": null, "checkpoint_sha256": "1c0b737c5083646ea897f058716904df27638ad591633cf4576b2c865cb94e87", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20140, "epoch": 510, "eval_seconds": 202.16419559900532, "length": 402.8, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_510.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 27030, "success": 0.54, "successes": 27, "successes_per_seed": [false, true, false, true, false, true, false, true, true, true, true, false, false, true, true, false, false, false, true, false, false, true, false, false, true, true, false, true, true, false, false, true, false, true, true, false, true, false, true, false, false, true, false, true, true, true, true, false, true, true], "total": 50}
18
+ {"checkpoint": null, "checkpoint_sha256": "1f9d4475c0d94abdd457786b7688f43ab09bebea048a24bb6c3955d1aefbe455", "ci": "wilson95", "ci_high": 0.6520286471342196, "ci_low": 0.38511744890044614, "environment_steps": 20733, "epoch": 540, "eval_seconds": 211.60593838698696, "length": 414.66, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_540.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 28620, "success": 0.52, "successes": 26, "successes_per_seed": [false, false, false, false, true, false, true, false, true, true, false, false, false, true, true, true, true, true, true, false, false, true, false, false, false, false, true, false, true, false, true, true, false, false, true, true, true, false, true, false, true, true, false, true, true, false, true, false, true, true], "total": 50}
19
+ {"checkpoint": null, "checkpoint_sha256": "f2fcab14a00bf879ebcc2b61aa406da61f43ab86c5b2bbc668f96273bbd7c9ed", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 20857, "epoch": 570, "eval_seconds": 207.37533284601523, "length": 417.14, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_570.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 30210, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, false, true, true, false, true, true, true, false, true, false, true, false, true, true, false, true, true, true, true, true, true, false, false, false, false, false, false, false, true, false, true, true, true, true, true, true, false, false, false, false, true, false, false, true, true, false, false], "total": 50}
20
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_600.pkl", "checkpoint_sha256": "093e9f6ef9affe9673e6d1c5ba35d41b51f0e46b1e831758b35c3096d617321b", "ci": "wilson95", "ci_high": 0.5576655823142176, "ci_low": 0.2937500335471198, "environment_steps": 21939, "epoch": 600, "eval_seconds": 183.72156420801184, "length": 438.78, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_600.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 31800, "success": 0.42, "successes": 21, "successes_per_seed": [false, false, false, false, true, false, false, true, false, true, false, false, false, false, false, false, true, true, true, false, true, true, false, false, false, false, false, false, false, false, true, true, false, true, true, false, true, false, true, true, false, true, false, true, true, true, true, false, true, false], "total": 50}
21
+ {"checkpoint": null, "checkpoint_sha256": "ecbd73c4ac063194c0b3a0296e57fbf53844263bf67b38f33640b1548533d453", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20264, "epoch": 630, "eval_seconds": 214.19332466099877, "length": 405.28, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_630.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 33390, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, true, true, true, true, false, true, true, false, false, false, false, true, true, true, false, true, true, true, false, false, false, true, true, true, false, false, true, true, true, true, true, false, false, false, true, false, true, true, false, true, false, false, true, false, true, false], "total": 50}
22
+ {"checkpoint": null, "checkpoint_sha256": "995435f45b674a3b68d81575d4b7ae615b92bedd7ef91f598917eb29f2ec4b9c", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20069, "epoch": 660, "eval_seconds": 200.36977762900642, "length": 401.38, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_660.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 34980, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, true, true, true, true, false, true, true, true, true, false, true, false, false, true, true, true, true, true, false, false, false, false, true, true, false, true, false, true, false, true, true, false, false, true, false, true, false, false, true, false, true, true, true, true, false, false, true], "total": 50}
23
+ {"checkpoint": null, "checkpoint_sha256": "0f4ad9fd4f911f5dd6d8acb27be1dd01450a8215e876b6722b7be161aa1d41ff", "ci": "wilson95", "ci_high": 0.7413721068980641, "ci_low": 0.4815044693099298, "environment_steps": 19217, "epoch": 690, "eval_seconds": 188.69151423400035, "length": 384.34, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_690.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 36570, "success": 0.62, "successes": 31, "successes_per_seed": [false, true, false, false, true, true, false, true, true, true, true, true, false, true, false, false, true, false, true, true, true, false, true, false, false, true, true, false, false, false, false, true, true, true, true, true, true, false, true, false, true, true, false, true, true, true, true, false, true, true], "total": 50}
24
+ {"checkpoint": null, "checkpoint_sha256": "beb6ae1f02883d608153835f0e4e4560bddd9a205b8ffe7f1a11bb5d5dc3e566", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19244, "epoch": 720, "eval_seconds": 176.27786240901332, "length": 384.88, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_720.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 38160, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, false, true, true, true, false, false, true, true, false, true, false, true, false, false, true, true, false, true, true, true, true, false, false, true, false, true, true, false, true, true, false, true, false, false, true, true, true, true, true, true, false, false, true, true, true, false, true, true], "total": 50}
25
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_750.pkl", "checkpoint_sha256": "bce66bc45847f7f2d47bfae4d7e5cb34174526411177e923ee757b1d4c37be4b", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 20738, "epoch": 750, "eval_seconds": 196.59095785298268, "length": 414.76, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_750.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 39750, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, false, true, false, true, true, true, true, true, true, false, true, false, false, true, true, false, true, false, false, true, false, false, true, true, true, true, false, false, false, false, true, true, false, false, false, true, false, true, false, false, false, true, true, false, true, true, true], "total": 50}
26
+ {"checkpoint": null, "checkpoint_sha256": "267b5cbeceda0e4bb669b19c4f0fa161795b68abda4672c34eea74a8d3b1a93a", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20322, "epoch": 780, "eval_seconds": 179.20251053699758, "length": 406.44, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_780.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 41340, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, false, true, false, false, true, true, false, true, false, true, false, true, true, true, true, true, true, false, true, false, false, true, false, true, false, false, true, true, true, true, true, false, false, false, true, false, true, false, false, false, false, true, true, true, true, true], "total": 50}
27
+ {"checkpoint": null, "checkpoint_sha256": "53b196eed1eca42a911395ff6d9c6951a09eb2aafc1c3e1e0cfc795a1bce3b1c", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20057, "epoch": 810, "eval_seconds": 197.0747535309929, "length": 401.14, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_810.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 42930, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, true, true, false, false, false, true, true, true, false, true, false, false, true, false, false, true, true, true, true, false, true, true, false, true, false, false, true, true, true, false, true, false, true, true, true, false, true, false, false, false, true, true, true, false, true, false], "total": 50}
28
+ {"checkpoint": null, "checkpoint_sha256": "bbf386a5f313ed25ddc6683c51093977f7bc287dba034208b70bc16806906f8a", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 19857, "epoch": 840, "eval_seconds": 162.03924702797667, "length": 397.14, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_840.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 44520, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, false, false, true, false, true, true, true, false, true, false, true, false, false, true, true, true, true, false, true, true, false, false, true, true, true, false, false, true, true, false, true, true, true, true, false, true, false, false, false, false, true, false, true, true, true, true, true], "total": 50}
29
+ {"checkpoint": null, "checkpoint_sha256": "132bbf0c7d31f5d03305e232a9535171da7294a78823b47992acc32772dc8bc4", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 20919, "epoch": 870, "eval_seconds": 190.812047385989, "length": 418.38, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_870.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 46110, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, false, false, true, true, true, true, true, true, false, false, true, false, false, true, false, false, true, false, true, true, false, false, false, true, false, true, false, true, true, false, true, true, true, true, false, true, false, false, false, false, false, true, true, true, false, true, true], "total": 50}
30
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_900.pkl", "checkpoint_sha256": "8688c70ee521c9406cb015a16390b99b2632f8ee353ce4f5b1510632b4d8651f", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20113, "epoch": 900, "eval_seconds": 179.4899732079939, "length": 402.26, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_900.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 47700, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, true, true, true, false, true, true, false, false, false, true, false, true, true, false, true, true, true, true, true, false, true, true, true, false, false, false, true, false, false, true, false, true, true, false, false, false, true, false, false, true, false, true, true, true, false, true], "total": 50}
31
+ {"checkpoint": null, "checkpoint_sha256": "a78b900ee7c014e62f46d5274f281d80e7515ee6dda796d04dc2d8086ee02700", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 21086, "epoch": 930, "eval_seconds": 186.08310608498869, "length": 421.72, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_930.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 49290, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, true, true, false, true, true, true, true, false, true, false, true, false, false, true, false, true, false, false, true, false, false, false, true, false, true, true, false, false, false, false, true, true, false, false, false, true, false, true, false, false, true, true, true, true, true, true, true], "total": 50}
32
+ {"checkpoint": null, "checkpoint_sha256": "80ccc4fa6c50d70ddc8fa664bfea016b37324dfb24a3c1355f3a1267336b5c13", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19555, "epoch": 960, "eval_seconds": 185.3208970319829, "length": 391.1, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_960.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 50880, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, false, true, false, true, true, false, true, false, false, true, false, true, false, false, true, true, false, true, true, true, true, false, true, true, false, false, false, false, true, true, true, true, true, true, true, false, true, false, false, true, false, true, true, true, true, true, true, true], "total": 50}
33
+ {"checkpoint": null, "checkpoint_sha256": "a65a3571eb16349c2c6b7ffb7162fee1bcef9d43492cab499f2d10b43b167e53", "ci": "wilson95", "ci_high": 0.7062499664528802, "ci_low": 0.4423344176857823, "environment_steps": 19402, "epoch": 990, "eval_seconds": 201.12720023200382, "length": 388.04, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_990.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 52470, "success": 0.58, "successes": 29, "successes_per_seed": [false, true, false, true, true, true, true, false, true, true, true, false, false, true, false, true, true, true, false, true, false, true, true, false, false, true, false, true, true, false, true, false, false, false, true, true, true, false, true, false, true, false, false, true, false, false, true, true, true, true], "total": 50}
34
+ {"checkpoint": null, "checkpoint_sha256": "e0f1378a19356298def15775fcfa2b47186ac62bd23f9fb04111f5eb373c7caa", "ci": "wilson95", "ci_high": 0.7413721068980641, "ci_low": 0.4815044693099298, "environment_steps": 19413, "epoch": 1020, "eval_seconds": 195.71366488700733, "length": 388.26, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1020.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 54060, "success": 0.62, "successes": 31, "successes_per_seed": [false, false, false, true, false, true, false, true, true, true, false, true, false, true, false, false, true, true, true, true, true, true, true, false, false, true, true, false, true, false, true, true, false, false, true, true, true, true, false, true, false, true, false, true, true, true, true, false, true, true], "total": 50}
35
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl", "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5", "ci": "wilson95", "ci_high": 0.7586125034815325, "ci_low": 0.501410168761127, "environment_steps": 19340, "epoch": 1050, "eval_seconds": 196.68048957397696, "length": 386.8, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1050.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 55650, "success": 0.64, "successes": 32, "successes_per_seed": [false, false, false, true, false, true, true, false, true, true, true, false, false, true, false, false, true, true, false, true, true, false, true, false, false, true, true, true, true, false, true, true, false, true, true, true, true, true, false, false, true, true, false, true, true, true, true, true, true, true], "total": 50}
36
+ {"checkpoint": null, "checkpoint_sha256": "4c08a9ae256200b8d788cd9ab13722e316ea1b7c80093246830b609e78a86012", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20440, "epoch": 1080, "eval_seconds": 187.9070016159967, "length": 408.8, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1080.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 57240, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, false, false, true, false, true, true, true, false, true, false, true, false, false, true, true, true, true, false, true, true, false, false, true, true, true, false, false, true, true, false, true, true, false, true, true, true, false, false, true, false, true, false, true, true, true, false, true], "total": 50}
37
+ {"checkpoint": null, "checkpoint_sha256": "97a11023448dfe0a7d21244107c87d30aa265094cef89d270f549b78006b0bfc", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20150, "epoch": 1110, "eval_seconds": 187.7733797649853, "length": 403.0, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1110.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 58830, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, true, true, true, false, false, true, true, false, true, false, true, false, true, true, true, false, false, true, false, true, false, true, true, true, false, true, false, true, true, false, false, false, true, true, true, false, true, false, false, false, true, true, true, true, true, false, true], "total": 50}
38
+ {"checkpoint": null, "checkpoint_sha256": "7bdf5584644919cbd98f01fd2efc2e13e16074ff49a402e3e5eb821be5852bdc", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20036, "epoch": 1140, "eval_seconds": 193.6986963720119, "length": 400.72, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1140.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 60420, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, true, false, true, true, false, true, true, true, true, false, true, false, true, false, true, true, true, true, false, false, false, false, true, true, true, true, false, true, true, false, true, true, false, false, false, true, false, false, false, false, true, true, false, true, true, true, true], "total": 50}
39
+ {"checkpoint": null, "checkpoint_sha256": "8346b9cf30b3eb279f52bda6c49494f462deb6bfc5037abcdd6d4d806c2fcc92", "ci": "wilson95", "ci_high": 0.6520286471342196, "ci_low": 0.38511744890044614, "environment_steps": 20512, "epoch": 1170, "eval_seconds": 205.80048555898247, "length": 410.24, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1170.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 62010, "success": 0.52, "successes": 26, "successes_per_seed": [false, true, false, true, true, false, false, true, true, true, false, true, false, true, false, false, true, true, false, true, false, true, true, false, false, true, true, false, true, false, true, true, false, true, true, true, true, false, false, false, true, false, false, false, false, true, true, true, false, false], "total": 50}
40
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1200.pkl", "checkpoint_sha256": "917bbe6325a0d941c496f0e18683e2be931adfb652637ecc86210760b38330dd", "ci": "wilson95", "ci_high": 0.7062499664528802, "ci_low": 0.4423344176857823, "environment_steps": 19908, "epoch": 1200, "eval_seconds": 176.04667832900304, "length": 398.16, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1200.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 63600, "success": 0.58, "successes": 29, "successes_per_seed": [false, false, false, true, true, true, false, false, true, true, false, false, false, true, false, true, true, true, true, true, false, true, true, false, false, true, true, false, false, false, true, true, false, false, true, true, true, false, true, false, true, false, false, true, true, true, true, true, true, true], "total": 50}
41
+ {"checkpoint": null, "checkpoint_sha256": "5ea009be7f4a808b3945a1f716d573a2ce83bbe759d46ab8a245693883e6b411", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20338, "epoch": 1230, "eval_seconds": 178.46198591098073, "length": 406.76, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1230.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 65190, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, true, true, false, false, true, true, false, true, true, true, false, true, true, true, true, true, false, false, true, false, false, true, false, false, true, false, true, true, false, false, true, true, true, false, true, false, true, false, false, false, true, false, true, false, true, true], "total": 50}
42
+ {"checkpoint": null, "checkpoint_sha256": "bf6a5fec3b373964adaf71a2055e1fa273cb7be4444d3bce4170dcbf987034b6", "ci": "wilson95", "ci_high": 0.5769397197834315, "ci_low": 0.3116219921125717, "environment_steps": 21444, "epoch": 1260, "eval_seconds": 187.16456362302415, "length": 428.88, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1260.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 66780, "success": 0.44, "successes": 22, "successes_per_seed": [false, false, true, true, true, true, false, false, true, true, false, true, false, true, false, false, false, true, false, true, false, false, false, false, false, true, true, false, false, false, true, true, false, true, true, true, false, false, false, false, true, false, false, true, false, true, false, true, false, true], "total": 50}
43
+ {"checkpoint": null, "checkpoint_sha256": "71b2952a9bdebe896859b1512f90a9f5ee85be5edd5cfebee75bda2758217187", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19592, "epoch": 1290, "eval_seconds": 183.00042536499677, "length": 391.84, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1290.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 68370, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, true, true, true, true, false, true, true, false, false, true, false, true, true, false, true, true, true, true, false, true, true, false, false, true, true, false, true, false, true, true, false, true, true, true, true, false, true, false, false, true, false, true, false, true, false, false, true, true], "total": 50}
44
+ {"checkpoint": null, "checkpoint_sha256": "060155afe7a52113d59b93e4995932885222009ee79f7c07aa0dac68ca934559", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20045, "epoch": 1320, "eval_seconds": 186.56828709199908, "length": 400.9, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1320.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 69960, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, true, true, true, true, false, false, false, true, true, false, false, true, false, false, true, true, false, true, true, true, false, false, false, true, false, false, true, false, true, false, false, true, true, true, true, false, true, false, false, true, true, true, true, true, true, true, true, false], "total": 50}
45
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1350.pkl", "checkpoint_sha256": "615b1857dfc6c7dcbccd3c80a558bc1e8652567f9b2a8a68554cb1399641fa4e", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 19899, "epoch": 1350, "eval_seconds": 194.9777886540105, "length": 397.98, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1350.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 71550, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, true, true, false, true, true, true, true, false, false, false, true, false, true, false, true, false, true, true, false, true, false, true, true, true, false, true, false, true, true, false, true, true, true, true, false, true, false, false, false, false, true, false, true, true, false, true, true], "total": 50}
46
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1400.pkl", "checkpoint_sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20093, "epoch": 1400, "eval_seconds": 204.52356311600306, "length": 401.86, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1400.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 74200, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, false, true, true, true, false, true, true, true, false, true, false, true, false, false, true, true, false, true, false, true, true, false, false, true, true, false, true, false, true, true, false, false, true, true, true, false, true, false, true, true, false, true, false, false, true, false, false, true], "total": 50}
47
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1450.pkl", "checkpoint_sha256": "0d7379e054c59cfcd79105334d4c5d2d9bcf22ab091beeac89bc50a72adac524", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19399, "epoch": 1450, "eval_seconds": 176.5020790050039, "length": 387.98, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1450.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 76850, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, false, true, false, true, false, true, true, true, false, true, false, true, false, false, true, true, false, false, true, true, true, false, true, true, false, true, true, false, true, true, false, true, true, true, true, true, true, false, false, true, false, false, true, false, true, true, true, true], "total": 50}
48
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1500.pkl", "checkpoint_sha256": "5e32f94decb6362716efba41a01f8e1e96df75915f387328ddec2d1b157a2493", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19759, "epoch": 1500, "eval_seconds": 184.12183918399387, "length": 395.18, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1500.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 79500, "success": 0.6, "successes": 30, "successes_per_seed": [false, true, false, true, true, false, false, true, true, true, false, false, false, true, false, false, true, false, false, true, true, true, true, false, true, true, false, false, true, true, true, true, false, true, true, true, true, false, true, true, false, true, false, false, true, true, true, true, false, true], "total": 50}
49
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1550.pkl", "checkpoint_sha256": "73534e5d8656f74db3854398429afdc773c03e7066b847385e310a7b8862ae30", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20258, "epoch": 1550, "eval_seconds": 192.20501950199832, "length": 405.16, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1550.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 82150, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, true, false, true, false, false, false, true, true, false, true, false, true, false, true, true, true, true, false, false, false, true, false, true, true, false, false, true, false, false, true, false, true, true, true, true, true, true, true, false, true, false, true, false, false, true, true, true, true], "total": 50}
50
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1600.pkl", "checkpoint_sha256": "121d1c83f7bcd2c9860f1966827b32f7a91e3c97418489d3ea6e07ed4654aa0d", "ci": "wilson95", "ci_high": 0.7413721068980641, "ci_low": 0.4815044693099298, "environment_steps": 19636, "epoch": 1600, "eval_seconds": 197.4091302800225, "length": 392.72, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1600.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 84800, "success": 0.62, "successes": 31, "successes_per_seed": [false, false, false, true, true, true, true, true, true, true, false, false, false, true, false, true, true, true, false, true, false, true, true, false, false, true, true, false, false, false, true, false, false, true, true, true, true, true, false, false, false, true, true, true, true, true, true, true, true, true], "total": 50}
51
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1650.pkl", "checkpoint_sha256": "5f9fb00865251bde43db1d2063c96226346a61408674b1bc62bba85c561221c6", "ci": "wilson95", "ci_high": 0.6148825510995539, "ci_low": 0.34797135286578046, "environment_steps": 21331, "epoch": 1650, "eval_seconds": 201.6851862729818, "length": 426.62, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1650.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 87450, "success": 0.48, "successes": 24, "successes_per_seed": [false, false, false, false, true, true, false, true, true, true, false, true, false, true, false, false, true, true, true, true, false, true, true, false, true, false, false, false, false, false, true, true, false, true, true, true, false, false, true, false, false, true, false, true, true, false, false, false, false, true], "total": 50}
52
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1700.pkl", "checkpoint_sha256": "e6a1d814eabc9a002dff4282032ea4e2f05e955db7a54276ccf424748e54fdf1", "ci": "wilson95", "ci_high": 0.7586125034815325, "ci_low": 0.501410168761127, "environment_steps": 19119, "epoch": 1700, "eval_seconds": 173.31722270400496, "length": 382.38, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1700.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 90100, "success": 0.64, "successes": 32, "successes_per_seed": [false, false, false, true, true, true, false, true, true, true, false, true, false, true, true, false, true, true, true, true, true, true, true, false, true, true, false, false, true, false, false, true, false, true, true, true, true, false, true, false, false, false, true, true, true, true, false, true, true, true], "total": 50}
53
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1750.pkl", "checkpoint_sha256": "6c0bd7d3b726ba7164c946b0546efaa06f258b40493ea25773c88d0b800fe498", "ci": "wilson95", "ci_high": 0.6335548568317142, "ci_low": 0.3664451431682858, "environment_steps": 20780, "epoch": 1750, "eval_seconds": 190.4759664760204, "length": 415.6, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1750.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 92750, "success": 0.5, "successes": 25, "successes_per_seed": [false, false, false, true, false, true, false, true, true, true, false, true, false, true, false, false, true, true, false, true, true, true, true, false, true, false, true, false, false, false, true, true, false, true, false, false, true, false, true, false, false, false, false, false, true, false, true, true, true, true], "total": 50}
54
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1800.pkl", "checkpoint_sha256": "0a21f4ed42690ae54b910ed234eb4e763fa7e648893e3e8aa4e47c118c581db7", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 19901, "epoch": 1800, "eval_seconds": 170.34209979700972, "length": 398.02, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1800.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 95400, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, false, false, false, true, false, true, true, true, true, true, false, true, false, false, true, true, true, true, true, true, true, false, true, true, true, true, false, false, false, true, false, false, true, true, true, true, false, false, true, true, false, false, true, false, false, true, false, true], "total": 50}
55
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1850.pkl", "checkpoint_sha256": "acea5fb19ab37fed19a22d7f024f6b44b79e78d5cae91b2c9d5ec8ed3194145d", "ci": "wilson95", "ci_high": 0.7239161026974346, "ci_low": 0.4618143774758936, "environment_steps": 19353, "epoch": 1850, "eval_seconds": 182.57497011701344, "length": 387.06, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1850.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 98050, "success": 0.6, "successes": 30, "successes_per_seed": [false, false, true, false, true, true, false, false, true, true, true, true, false, false, false, true, true, true, true, true, true, true, false, false, false, true, false, true, true, false, false, true, false, false, true, true, false, true, true, false, true, false, false, true, true, true, true, true, true, true], "total": 50}
56
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1900.pkl", "checkpoint_sha256": "c5939369d6198f7fdaf0fee49f7cabe85f19d2144de0de105da98c5c7eb59cfe", "ci": "wilson95", "ci_high": 0.6703034780777566, "ci_low": 0.40398871399157477, "environment_steps": 20127, "epoch": 1900, "eval_seconds": 189.9425573410117, "length": 402.54, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1900.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 100700, "success": 0.54, "successes": 27, "successes_per_seed": [false, false, true, true, false, true, false, true, true, true, true, true, false, true, false, false, true, true, true, false, true, true, true, false, true, true, false, false, false, false, true, false, false, true, false, true, true, false, true, false, false, true, false, false, false, false, true, true, true, true], "total": 50}
57
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1950.pkl", "checkpoint_sha256": "11560dd38b7df55ad45d0c5bfc500dc7c734fa1c071fb376bdf285941ffe1f84", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 20150, "epoch": 1950, "eval_seconds": 207.72391929998412, "length": 403.0, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_1950.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 103350, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, true, false, true, false, false, false, true, true, false, true, false, true, true, false, true, false, true, true, false, true, true, false, true, true, false, true, true, false, true, true, true, true, false, true, false, false, true, false, false, true, false, false, true, false, true, true, true, true], "total": 50}
58
+ {"checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_2000.pkl", "checkpoint_sha256": "bd73a82ecfb7aa8157292f8f819a95c9921a5d8caebd5dbdc6470f8b0cb116d9", "ci": "wilson95", "ci_high": 0.6883780078874284, "ci_low": 0.4230602802165686, "environment_steps": 19979, "epoch": 2000, "eval_seconds": 207.03382295698975, "length": 399.58, "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/eval/epoch_2000.parquet", "seeds": [880000, 880001, 880002, 880003, 880004, 880005, 880006, 880007, 880008, 880009, 880010, 880011, 880012, 880013, 880014, 880015, 880016, 880017, 880018, 880019, 880020, 880021, 880022, 880023, 880024, 880025, 880026, 880027, 880028, 880029, 880030, 880031, 880032, 880033, 880034, 880035, 880036, 880037, 880038, 880039, 880040, 880041, 880042, 880043, 880044, 880045, 880046, 880047, 880048, 880049], "step": 106000, "success": 0.56, "successes": 28, "successes_per_seed": [false, false, true, true, true, true, false, false, true, true, false, true, false, true, false, false, true, true, true, true, false, true, false, false, false, true, true, true, true, false, true, true, true, false, false, true, true, false, true, false, false, false, false, false, true, true, true, true, false, true], "total": 50}
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/epoch_1050.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36e242073f9d8296209dacd2221969a4de26710cb14492ad5a2707a0d66b7b85
3
+ size 6554
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/protocol.json ADDED
@@ -0,0 +1,226 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl",
3
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
4
+ "collision_geoms": false,
5
+ "epoch": 1050,
6
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
7
+ "job": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/jobs/00.json",
8
+ "panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
9
+ "reset_seeds": [
10
+ 70723122,
11
+ 66851989,
12
+ 60005579,
13
+ 32401708,
14
+ 19153477,
15
+ 42749167,
16
+ 46661357,
17
+ 27402198,
18
+ 47810457,
19
+ 18848891,
20
+ 17918997,
21
+ 80363015,
22
+ 64973617,
23
+ 14420984,
24
+ 34943840,
25
+ 20266056,
26
+ 55423323,
27
+ 77426949,
28
+ 21093578,
29
+ 74139179,
30
+ 65396037,
31
+ 66433136,
32
+ 18174832,
33
+ 44050868,
34
+ 79827091,
35
+ 48142557,
36
+ 81058782,
37
+ 57440191,
38
+ 71718518,
39
+ 62288442,
40
+ 38780438,
41
+ 31543247,
42
+ 45815861,
43
+ 69976624,
44
+ 27741928,
45
+ 47860773,
46
+ 68719953,
47
+ 72413983,
48
+ 43596719,
49
+ 25975279,
50
+ 75651978,
51
+ 62538517,
52
+ 49157692,
53
+ 72093687,
54
+ 35658896,
55
+ 34016592,
56
+ 56600068,
57
+ 85354907,
58
+ 64181994,
59
+ 64561502,
60
+ 76858975,
61
+ 17852290,
62
+ 20068153,
63
+ 39072183,
64
+ 55774539,
65
+ 26085971,
66
+ 37837570,
67
+ 88758485,
68
+ 75982683,
69
+ 81377703,
70
+ 57992264,
71
+ 80112931,
72
+ 88742902,
73
+ 47013480,
74
+ 36523859,
75
+ 19498677,
76
+ 20212586,
77
+ 55485421,
78
+ 36767911,
79
+ 50597940,
80
+ 65503466,
81
+ 87299732,
82
+ 61456577,
83
+ 79295759,
84
+ 10605088,
85
+ 21163776,
86
+ 74346251,
87
+ 59990172,
88
+ 40756359,
89
+ 29123831,
90
+ 22339698,
91
+ 55950171,
92
+ 25459058,
93
+ 89169841,
94
+ 51768760,
95
+ 59320974,
96
+ 43210884,
97
+ 19819408,
98
+ 27287306,
99
+ 77579712,
100
+ 33855715,
101
+ 18917811,
102
+ 50238382,
103
+ 29018479,
104
+ 58133015,
105
+ 23864203,
106
+ 49568781,
107
+ 55772368,
108
+ 69087148,
109
+ 43159811,
110
+ 42777958,
111
+ 79686252,
112
+ 88287947,
113
+ 24484856,
114
+ 40522505,
115
+ 18535306,
116
+ 27261685,
117
+ 55157386,
118
+ 73419901,
119
+ 51177153,
120
+ 66197636,
121
+ 89218160,
122
+ 20312137,
123
+ 32815280,
124
+ 38566965,
125
+ 67538690,
126
+ 51444755,
127
+ 49534469,
128
+ 55520272,
129
+ 78917430,
130
+ 53987225,
131
+ 33135526,
132
+ 10641443,
133
+ 83269321,
134
+ 50371327,
135
+ 60499091,
136
+ 14174315,
137
+ 34430409,
138
+ 78665761,
139
+ 55417809,
140
+ 29697446,
141
+ 67300186,
142
+ 14166946,
143
+ 34309048,
144
+ 38651973,
145
+ 63582284,
146
+ 11252552,
147
+ 13083214,
148
+ 12656665,
149
+ 39216146,
150
+ 53058687,
151
+ 26843165,
152
+ 26213599,
153
+ 75250873,
154
+ 86873350,
155
+ 23044120,
156
+ 74475483,
157
+ 73879766,
158
+ 38462205,
159
+ 36712875,
160
+ 61476060,
161
+ 40070375,
162
+ 11657834,
163
+ 12071629,
164
+ 70951113,
165
+ 86273812,
166
+ 33086249,
167
+ 41678564,
168
+ 34536118,
169
+ 27607156,
170
+ 70130490,
171
+ 70448950,
172
+ 63854211,
173
+ 58640013,
174
+ 71222810,
175
+ 77025643,
176
+ 46535790,
177
+ 13450163,
178
+ 48130374,
179
+ 28109111,
180
+ 27277335,
181
+ 14975106,
182
+ 25482760,
183
+ 51368844,
184
+ 65499499,
185
+ 38883080,
186
+ 56579733,
187
+ 81285249,
188
+ 71293324,
189
+ 79868703,
190
+ 78989122,
191
+ 27209169,
192
+ 43304167,
193
+ 84752852,
194
+ 39685313,
195
+ 10941718,
196
+ 17638132,
197
+ 42683951,
198
+ 29174955,
199
+ 13106538,
200
+ 82750942,
201
+ 27595794,
202
+ 85382063,
203
+ 22544583,
204
+ 82667196,
205
+ 53585719,
206
+ 68118062,
207
+ 84752103,
208
+ 10971071,
209
+ 17007174
210
+ ],
211
+ "script_sha256": "7be8b5440b354a862228f5a3d3855ec0ba7267c78c4e04158fa028e9633e8d32",
212
+ "selection": "frozen before this independent panel was evaluated",
213
+ "settings": {
214
+ "episodes": 200,
215
+ "interval_epochs": 30,
216
+ "policy_seed": 700001,
217
+ "reset_seed_start": 880000,
218
+ "seed_panel_seed": 20261021,
219
+ "seed_range": [
220
+ 10000000,
221
+ 90000000
222
+ ],
223
+ "seed_sampling": "uniform_without_replacement",
224
+ "workers": 8
225
+ }
226
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/final200/result.json ADDED
The diff for this file is too large to render. See raw diff
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/first_phase_complete.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "target_epoch": 2000,
3
+ "stopped": "2026-10-26T08:31:50Z",
4
+ "status": "awaiting_joint_curve_review",
5
+ "native_epoch": 2001,
6
+ "historical_selection_preserved": true,
7
+ "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_2000.pkl",
8
+ "sha256": "bd73a82ecfb7aa8157292f8f819a95c9921a5d8caebd5dbdc6470f8b0cb116d9",
9
+ "resume_epoch": 2000
10
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/normalizer.json ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_max": [
3
+ 1.0,
4
+ 1.0,
5
+ 1.0,
6
+ 0.3311271071434021,
7
+ 1.0,
8
+ 0.4967554807662964,
9
+ 0.6288543343544006,
10
+ 0.7448903918266296,
11
+ 0.6859682202339172,
12
+ 0.6408420205116272,
13
+ 0.3885757029056549,
14
+ 1.0,
15
+ 1.0,
16
+ 1.0,
17
+ 1.0,
18
+ 0.6266307234764099,
19
+ 1.0,
20
+ 1.0,
21
+ 0.6799960732460022,
22
+ 0.7951958179473877,
23
+ 0.7104465961456299,
24
+ 0.6693936586380005,
25
+ 1.0,
26
+ 1.0
27
+ ],
28
+ "action_min": [
29
+ -1.0,
30
+ -1.0,
31
+ -1.0,
32
+ -0.7686043381690979,
33
+ -1.0,
34
+ -1.0,
35
+ -1.0,
36
+ -1.0,
37
+ -1.0,
38
+ -1.0,
39
+ -1.0,
40
+ 1.0,
41
+ -1.0,
42
+ -1.0,
43
+ -1.0,
44
+ -0.8547752499580383,
45
+ -1.0,
46
+ -1.0,
47
+ -1.0,
48
+ -1.0,
49
+ -1.0,
50
+ -1.0,
51
+ -1.0,
52
+ 1.0
53
+ ],
54
+ "action_offset": [
55
+ 0.0,
56
+ 0.0,
57
+ 0.0,
58
+ -0.2187386155128479,
59
+ 0.0,
60
+ -0.2516222596168518,
61
+ -0.18557283282279968,
62
+ -0.12755480408668518,
63
+ -0.15701588988304138,
64
+ -0.1795789897441864,
65
+ -0.30571216344833374,
66
+ 1.0,
67
+ 0.0,
68
+ 0.0,
69
+ 0.0,
70
+ -0.11407226324081421,
71
+ 0.0,
72
+ 0.0,
73
+ -0.1600019633769989,
74
+ -0.10240209102630615,
75
+ -0.14477670192718506,
76
+ -0.16530317068099976,
77
+ 0.0,
78
+ 1.0
79
+ ],
80
+ "action_scale": [
81
+ 1.0,
82
+ 1.0,
83
+ 1.0,
84
+ 0.54986572265625,
85
+ 1.0,
86
+ 0.7483777403831482,
87
+ 0.8144271373748779,
88
+ 0.8724452257156372,
89
+ 0.8429840803146362,
90
+ 0.8204209804534912,
91
+ 0.6942878365516663,
92
+ 1.0,
93
+ 1.0,
94
+ 1.0,
95
+ 1.0,
96
+ 0.7407029867172241,
97
+ 1.0,
98
+ 1.0,
99
+ 0.8399980068206787,
100
+ 0.8975979089736938,
101
+ 0.8552232980728149,
102
+ 0.8346968293190002,
103
+ 1.0,
104
+ 1.0
105
+ ],
106
+ "proprio_offset": [
107
+ -0.35657283663749695,
108
+ -0.18467170000076294,
109
+ 0.9951127171516418,
110
+ 0.7680414915084839,
111
+ 0.2349419891834259,
112
+ 0.3196081221103668,
113
+ 0.245033398270607,
114
+ 1.2771103382110596,
115
+ 0.0536290742456913,
116
+ 0.04810062423348427,
117
+ 0.04661514610052109,
118
+ 0.3628089427947998,
119
+ 0.3544827103614807,
120
+ 0.39943715929985046,
121
+ 0.3932478129863739,
122
+ 0.392443984746933,
123
+ 0.4009668529033661,
124
+ 0.28613996505737305,
125
+ 0.31512507796287537,
126
+ -0.2741226851940155,
127
+ 0.2910534143447876,
128
+ 1.0523529052734375,
129
+ 0.8952255249023438,
130
+ 0.018957961350679398,
131
+ 0.03767657279968262,
132
+ -0.007815996184945107,
133
+ 1.234562635421753,
134
+ 0.16546416282653809,
135
+ 0.2870045006275177,
136
+ 0.32762861251831055,
137
+ 0.4227330982685089,
138
+ 0.5281990170478821,
139
+ 0.646793007850647,
140
+ 0.691037118434906,
141
+ 0.7268394231796265,
142
+ 0.738739013671875,
143
+ 0.6841607093811035,
144
+ 0.7076348662376404
145
+ ],
146
+ "proprio_scale": [
147
+ 0.08131919056177139,
148
+ 0.05108539015054703,
149
+ 0.06973035633563995,
150
+ 0.1889827400445938,
151
+ 0.22627729177474976,
152
+ 0.23408500850200653,
153
+ 0.22585122287273407,
154
+ 0.1290702074766159,
155
+ 0.06828603893518448,
156
+ 0.10518860071897507,
157
+ 0.11534391343593597,
158
+ 0.4317815601825714,
159
+ 0.4290821850299835,
160
+ 0.46645617485046387,
161
+ 0.4576132297515869,
162
+ 0.4597686231136322,
163
+ 0.47026729583740234,
164
+ 0.42680424451828003,
165
+ 0.46843960881233215,
166
+ 0.09688738733530045,
167
+ 0.10850110650062561,
168
+ 0.09816151857376099,
169
+ 0.15734872221946716,
170
+ 0.37181636691093445,
171
+ 0.1457068920135498,
172
+ 0.11178682744503021,
173
+ 0.13545534014701843,
174
+ 0.14100651443004608,
175
+ 0.17894944548606873,
176
+ 0.21429292857646942,
177
+ 0.3902991712093353,
178
+ 0.4854361414909363,
179
+ 0.577340304851532,
180
+ 0.6130419373512268,
181
+ 0.6512577533721924,
182
+ 0.6606327295303345,
183
+ 0.6302294731140137,
184
+ 0.6418868899345398
185
+ ],
186
+ "source": "training_trajectories_only",
187
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b"
188
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/prior_index_record.json ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "task": "DrawerCleanup",
3
+ "suite": "DexMimicGen",
4
+ "experiment": "dmg_drawer_cleanup",
5
+ "regime": "diverse50",
6
+ "seed": 1,
7
+ "epoch": 1050,
8
+ "trained_epochs_at_selection": 2001,
9
+ "planned_epochs": 3000,
10
+ "first_phase_epochs": 2000,
11
+ "training_status_at_selection": "selected_after_joint_review",
12
+ "family": "StandardBaseBC",
13
+ "framework": "jax",
14
+ "weights": "complete_model_EMA",
15
+ "horizon": 8,
16
+ "collision_geoms": false,
17
+ "selection_rule": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
18
+ "window_success_rate": 0.6066666666666667,
19
+ "window_epochs": [
20
+ 1020,
21
+ 1050,
22
+ 1080
23
+ ],
24
+ "window_episodes": 150,
25
+ "selection_note": "Neighborhood score combines different checkpoints on reused periodic seeds; use final200 for this frozen checkpoint performance.",
26
+ "final200": {
27
+ "success": 0.61,
28
+ "successes": 122,
29
+ "total": 200,
30
+ "ci_low": 0.5409374595716976,
31
+ "ci_high": 0.6749165686253038
32
+ },
33
+ "final200_panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
34
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
35
+ "checkpoint_bytes": 138495554,
36
+ "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
37
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
38
+ "path": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl",
39
+ "directory": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1",
40
+ "supervision": {
41
+ "status": "selected_after_joint_review",
42
+ "training": "stopped",
43
+ "joint_curve_review_status": "finalized_after_phase2000_review",
44
+ "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
45
+ "review_evidence": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json",
46
+ "joint_curve_review_reason": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
47
+ "final200_role": "report_only_not_used_for_selection",
48
+ "zero_overfitting_claim": false,
49
+ "completed_train_epoch": 2001,
50
+ "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
51
+ },
52
+ "evaluation_interval_at_selected_epoch": 30,
53
+ "evaluation_schedule": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/evaluation_schedule.json",
54
+ "supersedes_or_reconfirms": {
55
+ "epoch": 1050,
56
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
57
+ "archive": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
58
+ }
59
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/bc_eval_schedule.py ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Auditable evaluation-only schedule override for already sealed BC continuations.
2
+
3
+ The original job, resolved contract, optimizer/RNG state and source files stay intact.
4
+ An explicit evaluation_schedule.json accompanies them and all subsequent releases.
5
+ Only the RHS of the native loop's `evaluating` assignment is replaced in memory;
6
+ the complete effective function and its hash are recorded before entering training.
7
+ """
8
+ import ast
9
+ import hashlib
10
+ import inspect
11
+ import json
12
+ from pathlib import Path
13
+
14
+
15
+ ORIGINAL = 'epoch%job["evaluation"]["interval_epochs"]==0 or epoch==epochs'
16
+
17
+
18
+ def transform(source, interval):
19
+ if interval != 50:
20
+ raise ValueError('This approved continuation protocol requires interval 50')
21
+ tree = ast.parse(source)
22
+ assignments = [n for n in ast.walk(tree) if isinstance(n, ast.Assign)
23
+ and any(isinstance(t, ast.Name) and t.id == 'evaluating' for t in n.targets)]
24
+ expected = ast.parse(ORIGINAL, mode='eval').body
25
+ if len(assignments) != 1 or ast.dump(assignments[0].value) != ast.dump(expected):
26
+ raise ValueError('Native evaluation trigger changed; review the runtime adapter')
27
+ assignments[0].value = ast.parse(f'epoch % {interval} == 0 or epoch == epochs', mode='eval').body
28
+ return ast.unparse(ast.fix_missing_locations(tree)) + '\n'
29
+
30
+
31
+ def install(module, job_path, resume):
32
+ job_path = Path(job_path)
33
+ job = json.loads(job_path.read_text())
34
+ output = Path(job['output'])
35
+ path = output / 'evaluation_schedule.json'
36
+ if not path.exists():
37
+ return False
38
+ schedule = json.loads(path.read_text())
39
+ if not resume:
40
+ raise ValueError('Continuation evaluation schedule requires --resume')
41
+ if schedule['identity_hash'] != job['identity_hash']:
42
+ raise ValueError('Evaluation schedule belongs to another job')
43
+ if schedule['original_interval_epochs'] != job['evaluation']['interval_epochs']:
44
+ raise ValueError('Original evaluation interval mismatch')
45
+ if schedule['helper_sha256'] != hashlib.sha256(Path(__file__).read_bytes()).hexdigest():
46
+ raise ValueError('Unreviewed evaluation schedule adapter')
47
+ start = json.loads((output / 'resume.json').read_text())['epoch']
48
+ if start < schedule['effective_after_epoch']:
49
+ raise ValueError('Cannot apply new cadence before its recorded recovery boundary')
50
+ source = inspect.getsource(module.train)
51
+ digest = hashlib.sha256(source.encode()).hexdigest()
52
+ if digest != schedule['native_train_source_sha256']:
53
+ raise ValueError('Native train function changed')
54
+ effective = transform(source, schedule['interval_epochs'])
55
+ effective_path = output / 'effective_train_eval50.py'
56
+ effective_path.write_text(effective)
57
+ namespace = dict(module.__dict__)
58
+ exec(compile(effective, str(effective_path), 'exec'), namespace)
59
+ module.train = namespace['train']
60
+ # evaluate is installed by the cache-release wrapper before this function is called.
61
+ record = dict(resume_epoch=start, effective_after_epoch=schedule['effective_after_epoch'],
62
+ interval_epochs=schedule['interval_epochs'], identity_hash=job['identity_hash'],
63
+ native_train_source_sha256=digest,
64
+ effective_train_sha256=hashlib.sha256(effective.encode()).hexdigest(),
65
+ helper_sha256=schedule['helper_sha256'])
66
+ (output / 'evaluation_schedule_active.json').write_text(json.dumps(record, indent=2) + '\n')
67
+ print('EVALUATION_SCHEDULE ' + json.dumps(record), flush=True)
68
+ return True
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/effective_train_eval50.py ADDED
@@ -0,0 +1,132 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ def train(job, *, resume=False, device='cuda:0'):
2
+ os.environ.setdefault('ROBOMIMIC_DATASET_ROOT', str(ROOT / 'data/robomimic'))
3
+ os.environ.setdefault('XLA_PYTHON_CLIENT_PREALLOCATE', 'false')
4
+ from .bc_storage import export_checkpoint, read_state, write_state
5
+ from .bc_training import make_trainer
6
+ from .bc_queue import training_lock
7
+ from .pixel_data import PixelDataset
8
+ output = Path(job['output'])
9
+ if job['framework'] != 'jax':
10
+ raise ValueError('Torch BC training is archived (legacy/source/20261022_torch); plan JAX jobs')
11
+ with training_lock(output, blocking=False), selection_env(RL_PT_EXPERIMENT=job['experiment']):
12
+ spec = validate_job(job)
13
+ old = output / 'resolved_config.json'
14
+ if old.exists() and (not resume):
15
+ raise FileExistsError('Training output exists; use --resume')
16
+ dataset = PixelDataset(job['manifest'], job['regime'], job['pixel_cache'], job['seed'] + 20000)
17
+ dataset.spec = spec
18
+ config = job['training']
19
+ from .bc_full_cut import prepare_targets, record_crossings
20
+ cut = prepare_targets(job)
21
+ if not cut['ready']:
22
+ raise ValueError(f'Mix-Full references not ready: {cut}')
23
+ resolved = dict(job, pixel_spec=asdict(spec), bc_identity=dict(family=job['family'], framework=job['framework']), split_hash=canonical_hash(dataset.split), normalizer_hash=canonical_hash(dataset.normalizer), training_samples=len(dataset.indices['train']), validation_samples=len(dataset.indices['validation']), ema_buffers='copy_live_running_statistics', checkpoint_encoding='lossless_gzip_ema_only')
24
+ if cut['contract'] is not None:
25
+ resolved['full_cut_stop'] = cut['contract']
26
+ import importlib.metadata
27
+ resolved['runtime'] = dict(python=sys.executable, requested_device=device, framework_version=importlib.metadata.version('jax'))
28
+ resolved_hash = canonical_hash(resolved)
29
+ if old.exists() and canonical_hash(read_json(old)) != resolved_hash:
30
+ raise ValueError('Resume model/framework/data/training contract changed')
31
+ trainer = make_trainer(job['framework'], spec, dataset.normalizer, config, job['seed'], device)
32
+ if not job.get('validation_only'):
33
+ import jax
34
+ if jax.default_backend() != 'gpu':
35
+ raise ValueError('Formal JAX BC requires the assigned CUDA GPU')
36
+ start_epoch = 0
37
+ resume_path = output / 'resume.pkl'
38
+ if resume:
39
+ if sha256_file(resume_path) != read_json(resume_path.with_suffix('.json'))['sha256']:
40
+ raise ValueError('BC resume checksum changed')
41
+ saved = read_state(resume_path)
42
+ if saved['resolved_hash'] != resolved_hash:
43
+ raise ValueError('BC resume belongs to another contract')
44
+ trainer.restore(saved['trainer'])
45
+ start_epoch = saved['epoch']
46
+ archive_after_resume(output, start_epoch)
47
+ write_json(old, resolved)
48
+ write_json(output / 'split.json', dataset.split)
49
+ write_json(output / 'normalizer.json', dataset.normalizer)
50
+ val = dataset.sample(job['validation']['batch_size'], 'validation', np.random.default_rng(job['validation']['batch_seed']))
51
+ epochs, batch_size = (config['epochs'], config['batch_size'])
52
+ status = dict(status='running', epoch=start_epoch, step=trainer.step, epochs=epochs, family=job['family'], framework=job['framework'], identity_hash=job['identity_hash'])
53
+ write_json(output / 'status.json', status)
54
+ tracker = None
55
+ wb = job.get('wandb', {})
56
+ mode = os.environ.get('WANDB_MODE', wb.get('mode', 'offline'))
57
+ if wb.get('project') and mode != 'disabled':
58
+ import wandb
59
+ tracker = wandb.init(project=wb['project'], mode=mode, dir=str(output), id=job['identity_hash'][:12], name=f'{job['family']}-{job['framework']}-{job['experiment']}-{job['regime']}-s{job['seed']}', group=output.parents[4].name, config=resolved, tags=[job['family'], job['framework'], 'bc_pretraining'], job_type='bc_pretraining')
60
+ tracker.define_metric('epoch')
61
+ for prefix in ('train/*', 'val/*', 'eval/*'):
62
+ tracker.define_metric(prefix, step_metric='epoch')
63
+ write_json(output / 'wandb_run.json', dict(id=tracker.id, mode=mode, project=wb['project'], url=tracker.url))
64
+ try:
65
+ if resume and cut['contract'] is not None:
66
+ crossing = record_crossings(output, cut['targets'])
67
+ if crossing['all_found']:
68
+ status.update(status='full_cut_found', full_cut_found_count=len(cut['targets']['references']), full_cut_targets_hash=cut['targets']['hash'])
69
+ return status
70
+ for epoch in range(start_epoch + 1, epochs + 1):
71
+ started, loss_sum, grad_sum, samples, updates = (time.monotonic(), 0.0, 0.0, 0, 0)
72
+ for batch in dataset.epoch_batches(batch_size, epoch, job['seed']):
73
+ info = trainer.update(batch)
74
+ if not all((np.isfinite(v) for v in info.values())):
75
+ raise FloatingPointError(f'Nonfinite BC update {trainer.step}')
76
+ n = len(batch['actions'])
77
+ loss_sum += info['loss'] * n
78
+ grad_sum += info['grad_norm'] * n
79
+ samples += n
80
+ updates += 1
81
+ if job.get('validation_only') and updates >= job.get('max_batches_per_epoch', 2):
82
+ break
83
+ if not job.get('validation_only') and samples != len(dataset.indices['train']):
84
+ raise RuntimeError('Incomplete full training epoch')
85
+ row = dict(epoch=epoch, step=trainer.step, loss=loss_sum / samples, grad_norm=grad_sum / samples, train_seconds=time.monotonic() - started, samples=samples, updates=updates, full_epoch=samples == len(dataset.indices['train']))
86
+ append(output / 'train_metrics.jsonl', row)
87
+ validation_loss = trainer.validation(val, job['validation']['loss_seed'])
88
+ if not np.isfinite(validation_loss):
89
+ raise FloatingPointError('Nonfinite validation loss')
90
+ append(output / 'validation_metrics.jsonl', dict(epoch=epoch, step=trainer.step, validation_loss=validation_loss))
91
+ metrics = {'epoch': epoch, 'gradient_step': trainer.step, 'train/loss': row['loss'], 'train/grad_norm': row['grad_norm'], 'train/seconds': row['train_seconds'], 'val/loss': validation_loss}
92
+ saving = epoch % job['checkpoint_interval_epochs'] == 0 or epoch == epochs
93
+ evaluating = epoch % 50 == 0 or epoch == epochs
94
+ if saving or evaluating:
95
+ if shutil.disk_usage(output).free < 20 * 1024 ** 3:
96
+ raise OSError('Less than 20 GiB free; preserving completed checkpoints and pausing writes')
97
+ path = output / ('checkpoints' if saving else '.eval') / f'epoch_{epoch}.pkl'
98
+ weights, stats = trainer.weights()
99
+ meta = export_checkpoint(path, spec, job['framework'], weights, stats, dataset, resolved, epoch, trainer.step)
100
+ if evaluating and (not job.get('skip_eval', False)):
101
+ eval_started = time.monotonic()
102
+ result = evaluate(path, job, epoch)
103
+ metrics.update({'eval/success_rate': result['success'], 'eval/ci_low': result['ci_low'], 'eval/ci_high': result['ci_high']})
104
+ append(output / 'evaluations.jsonl', dict(epoch=epoch, step=trainer.step, checkpoint=str(path) if saving else None, checkpoint_sha256=meta['sha256'], eval_seconds=time.monotonic() - eval_started, **{k: v for k, v in result.items() if k not in ('rows', 'episodes')}))
105
+ if saving:
106
+ write_state(resume_path, dict(resolved_hash=resolved_hash, epoch=epoch, trainer=trainer.save_state()), compress=False)
107
+ write_json(resume_path.with_suffix('.json'), dict(sha256=sha256_file(resume_path), epoch=epoch, step=trainer.step))
108
+ elif evaluating:
109
+ path.unlink()
110
+ path.with_suffix('.json').unlink()
111
+ status.update(epoch=epoch, step=trainer.step, last_loss=row['loss'], validation_loss=validation_loss)
112
+ if cut['contract'] is not None and evaluating:
113
+ crossing = record_crossings(output, cut['targets'])
114
+ status.update(full_cut_found_count=len(crossing['crossings']), full_cut_targets_hash=cut['targets']['hash'])
115
+ if crossing['all_found']:
116
+ status.update(status='full_cut_found', stopping_reason=cut['targets'].get('stopping_reason', 'all_three_first_strict_crossings'))
117
+ write_json(output / 'status.json', status)
118
+ if tracker is not None:
119
+ tracker.log(metrics)
120
+ print(f'{job['framework']} {job['experiment']} {job['regime']} epoch={epoch}/{epochs} loss={row['loss']:.6g} val={validation_loss:.6g} seconds={row['train_seconds']:.1f}', flush=True)
121
+ if status['status'] == 'full_cut_found':
122
+ break
123
+ if status['status'] != 'full_cut_found':
124
+ status['status'] = 'validated_short_run' if job.get('validation_only') else 'no_full_cut_before_cap' if cut['contract'] is not None else 'completed'
125
+ except BaseException as error:
126
+ status.update(status='interrupted' if isinstance(error, KeyboardInterrupt) else 'failed', error=repr(error))
127
+ raise
128
+ finally:
129
+ write_json(output / 'status.json', status)
130
+ if tracker is not None:
131
+ tracker.finish(exit_code=0 if status['status'] in ('completed', 'validated_short_run', 'full_cut_found', 'no_full_cut_before_cap') else 1)
132
+ return status
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/evaluation_schedule_active.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "resume_epoch": 1850,
3
+ "effective_after_epoch": 1350,
4
+ "interval_epochs": 50,
5
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
6
+ "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
7
+ "effective_train_sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
8
+ "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9"
9
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/native_training_status.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2001,
3
+ "epochs": 3000,
4
+ "error": "KeyboardInterrupt()",
5
+ "family": "StandardBaseBC",
6
+ "framework": "jax",
7
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
8
+ "last_loss": 0.056489796539936236,
9
+ "status": "interrupted",
10
+ "step": 106053,
11
+ "validation_loss": 0.1636236011981964
12
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/provenance/train_bc_noleak.py ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """scripts/pretrain_bc.py with one runtime fix: release evaluation policies after each evaluation.
2
+
3
+ PixelPolicy.encode/sample (and subclasses) are jax.jit methods with `self` static, so every periodic
4
+ evaluation's freshly loaded policy compiles new executables with its weights baked in as constants
5
+ and the jit cache keeps them forever: a BC's GPU memory grows by roughly one model per evaluation
6
+ until it, or a neighbour on the card, runs out (2026-10-25: Transport50 13.7 -> 23.8 GiB in ~1 h).
7
+ Clearing those methods' caches after each evaluation returns the memory to JAX's pool. Training math,
8
+ checkpoints and the hashed sources are untouched (this file is outside bc_pretrain.SOURCE_FILES).
9
+ Run from the checkout root: python train_bc_noleak.py train --job ... [--resume]
10
+ """
11
+ import gc
12
+ import sys
13
+ from pathlib import Path
14
+
15
+ # Evaluation workers are spawned and re-import this file as __mp_main__; they must not import jax
16
+ # before pixel_vec sets their environment, so everything happens under the __main__ guard.
17
+
18
+
19
+ def main():
20
+ sys.path.insert(0, str(Path.cwd() / "src"))
21
+ from rl_pt_analyze import bc_pretrain
22
+ from rl_pt_analyze.pixel import PixelPolicy
23
+ import rl_pt_analyze.bc_standard_policy # noqa: F401 -- register the Standard subclass
24
+
25
+ def policy_classes(cls=PixelPolicy):
26
+ yield cls
27
+ for sub in cls.__subclasses__():
28
+ yield from policy_classes(sub)
29
+
30
+ def release_policy_caches():
31
+ for cls in policy_classes():
32
+ for value in vars(cls).values():
33
+ if hasattr(value, "clear_cache"):
34
+ value.clear_cache()
35
+ gc.collect()
36
+
37
+ original = bc_pretrain._evaluate
38
+
39
+ def evaluate(checkpoint, job, epoch):
40
+ try:
41
+ return original(checkpoint, job, epoch)
42
+ finally:
43
+ release_policy_caches()
44
+
45
+ bc_pretrain.evaluate = evaluate
46
+ sys.path.insert(0, "/home/nhy/RL-postTrain/repo_visual/scripts")
47
+ from bc_eval_schedule import install
48
+ if "train" in sys.argv and "--job" in sys.argv:
49
+ install(bc_pretrain, sys.argv[sys.argv.index("--job") + 1], "--resume" in sys.argv)
50
+ bc_pretrain.main()
51
+
52
+
53
+ if __name__ == "__main__":
54
+ main()
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/release.json ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "task": "DrawerCleanup",
3
+ "suite": "DexMimicGen",
4
+ "experiment": "dmg_drawer_cleanup",
5
+ "regime": "diverse50",
6
+ "seed": 1,
7
+ "epoch": 1050,
8
+ "trained_epochs_at_selection": 2001,
9
+ "planned_epochs": 3000,
10
+ "first_phase_epochs": 2000,
11
+ "training_status_at_selection": "selected_after_joint_review",
12
+ "family": "StandardBaseBC",
13
+ "framework": "jax",
14
+ "weights": "complete_model_EMA",
15
+ "horizon": 8,
16
+ "collision_geoms": false,
17
+ "selection_rule": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
18
+ "window_success_rate": 0.6066666666666667,
19
+ "window_epochs": [
20
+ 1020,
21
+ 1050,
22
+ 1080
23
+ ],
24
+ "window_episodes": 150,
25
+ "selection_note": "Neighborhood score combines different checkpoints on reused periodic seeds; use final200 for this frozen checkpoint performance.",
26
+ "final200": {
27
+ "success": 0.61,
28
+ "successes": 122,
29
+ "total": 200,
30
+ "ci_low": 0.5409374595716976,
31
+ "ci_high": 0.6749165686253038
32
+ },
33
+ "final200_panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
34
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
35
+ "checkpoint_bytes": 138495554,
36
+ "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
37
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
38
+ "path": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/checkpoints/epoch_1050.pkl",
39
+ "directory": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1",
40
+ "supervision": {
41
+ "status": "selected_after_joint_review",
42
+ "training": "stopped",
43
+ "joint_curve_review_status": "finalized_after_phase2000_review",
44
+ "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
45
+ "review_evidence": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json",
46
+ "joint_curve_review_reason": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
47
+ "final200_role": "report_only_not_used_for_selection",
48
+ "zero_overfitting_claim": false,
49
+ "completed_train_epoch": 2001,
50
+ "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
51
+ },
52
+ "evaluation_interval_at_selected_epoch": 30,
53
+ "evaluation_schedule": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/evaluation_schedule.json",
54
+ "supersedes_or_reconfirms": {
55
+ "epoch": 1050,
56
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
57
+ "archive": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
58
+ }
59
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/resolved_config.json ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bc_identity": {
3
+ "family": "StandardBaseBC",
4
+ "framework": "jax"
5
+ },
6
+ "checkpoint_encoding": "lossless_gzip_ema_only",
7
+ "checkpoint_interval_epochs": 50,
8
+ "ema_buffers": "copy_live_running_statistics",
9
+ "evaluation": {
10
+ "episodes": 50,
11
+ "final": {
12
+ "episodes": 200,
13
+ "seed_panel_seed": 20261021,
14
+ "seed_range": [
15
+ 10000000,
16
+ 90000000
17
+ ],
18
+ "seed_sampling": "uniform_without_replacement"
19
+ },
20
+ "interval_epochs": 30,
21
+ "policy_seed": 700001,
22
+ "reset_seed_start": 880000,
23
+ "workers": 8
24
+ },
25
+ "experiment": "dmg_drawer_cleanup",
26
+ "family": "StandardBaseBC",
27
+ "framework": "jax",
28
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
29
+ "index": 0,
30
+ "manifest": "/home/nhy/RL-postTrain/repo_visual/data/bc/versions/dmg_distribution_v1_local_20261025/manifests/dmg_drawer_cleanup.json",
31
+ "manifest_sha256": "110e1f8373c49813ff466b7f24ac25347ae7564ff970efb01556e4fa05e57bba",
32
+ "model_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
33
+ "normalizer_hash": "e224f78a15bbee112e82d8544f735d9253c42dff6c5f5f1f5730100fe3a0b2d2",
34
+ "observation_mode": "pixel_state",
35
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
36
+ "pixel_cache": "/home/nhy/RL-postTrain/repo_visual/data/pixels/dmg_drawer_cleanup/visual_20261025",
37
+ "pixel_spec": {
38
+ "action_dim": 24,
39
+ "architecture": "dice_vision_resnet_flow_unet_v1",
40
+ "backbone_initialization": "random",
41
+ "backbone_norm": "batch_norm",
42
+ "cameras": [
43
+ "agentview",
44
+ "robot0_eye_in_hand",
45
+ "robot1_eye_in_hand"
46
+ ],
47
+ "collision_geoms": false,
48
+ "crop_size": 84,
49
+ "family": "StandardBaseBC",
50
+ "feature_dim": 166,
51
+ "features_per_camera": 64,
52
+ "flow_steps": 10,
53
+ "horizon": 8,
54
+ "image_size": 96,
55
+ "proprio_dim": 38,
56
+ "proprio_keys": [
57
+ "robot0_eef_pos",
58
+ "robot0_eef_quat",
59
+ "robot0_gripper_qpos",
60
+ "robot1_eef_pos",
61
+ "robot1_eef_quat",
62
+ "robot1_gripper_qpos"
63
+ ],
64
+ "task": "dmg_drawer_cleanup",
65
+ "time_embedding_dim": 64
66
+ },
67
+ "regime": "diverse50",
68
+ "runtime": {
69
+ "framework_version": "0.10.1",
70
+ "python": "/home/nhy/RL-postTrain/repo/.venvs/ogpo/bin/python",
71
+ "requested_device": "cuda:0"
72
+ },
73
+ "seed": 1,
74
+ "sources_sha256": {
75
+ "configs/bc_models.yaml": "ce58a903c4270328985ab8677babdaa12748ad18efc3749faed72425024e61c1",
76
+ "configs/bc_pretraining.yaml": "7d1b9be73f25aeb40cd342a3e3ca2543ea865b38b2abe8fb28a7da6588d419a6",
77
+ "configs/bc_pretraining_ph_full.yaml": "a93c7d4a50e9afa836d98a6193ba1bc2fb3827a259cf5de39f5dc7a2504d6728",
78
+ "configs/bc_standard_init_schema.json": "e2c70eee2a1ce8d1a812ec427d24ba3cc45519434331354318d97415fd559bc8",
79
+ "configs/study1.yaml": "35d065e242b205a4c99f8add2eab35bb85f80401f4e07d49f063a0eddf4de608",
80
+ "configs/study1_ph_full.yaml": "f9cebc4b66279d4ea7492f7a80a09f1adca7c1cae7b242374305be4f701e76ef",
81
+ "scripts/pretrain_bc.py": "9972b77f7b8ed964e7e04fa3ba3821b767f0b7a5d697625ed1c6f8fcdf990890",
82
+ "src/rl_pt_analyze/bc_evaluation.py": "7319c346f453263a99cdc5d6fd39e6a0a88783d2c42f68a312ee36ab6d656c6f",
83
+ "src/rl_pt_analyze/bc_fleet.py": "a49c24c66ec0bb1a5548d384deefda429ff1ca693cb19f1c99dbd661148d7ffc",
84
+ "src/rl_pt_analyze/bc_full_cut.py": "a8170de58bc0a16dd5c6358a6429d96afb5227e5701602806d68e42727f671bc",
85
+ "src/rl_pt_analyze/bc_identity.py": "f94c284c71df7ed06eaf4553297de53f2b801a5fac743beb3b98d1bc41314fe0",
86
+ "src/rl_pt_analyze/bc_policy.py": "aff3c4613a0dfc46d6491bdba6478261049c2c9631eca446fa8939a1e498fc30",
87
+ "src/rl_pt_analyze/bc_pretrain.py": "703b2e50fe53d62b23910b907d75c09cec49463e5e9e2c08d8f77d2b6b77a5bd",
88
+ "src/rl_pt_analyze/bc_runtime.py": "29d85733684176bd7328c01921ed1eb45857a914d8ee998e6df4fb0a5be5315d",
89
+ "src/rl_pt_analyze/bc_standard_jax.py": "9063d0d4160e972bb9f7b1acd9e508a2ea1118ceeed7647614c787abcb7b32ce",
90
+ "src/rl_pt_analyze/bc_standard_policy.py": "54d733549ec571f709d49698520f2461def5a20fc4146d155b3d56fbbcf5deb1",
91
+ "src/rl_pt_analyze/bc_storage.py": "83b79d5beaaaa46310d756476f5da90c9227a656533dc9414334d395a6749e54",
92
+ "src/rl_pt_analyze/bc_training.py": "bf20475ceeef114e3444c2610ea6d490389b02d147f22c962e2f63fd94b9dafa",
93
+ "src/rl_pt_analyze/pixel.py": "ec46a0b0e3c40bca7ea2d0c76b086b482fedaf98293b06f7ab719b86b81141e6",
94
+ "src/rl_pt_analyze/pixel_data.py": "74ed81ff034d0782671ff9e48157a3be2d274e7db3735a955705b8e8fcf42b34",
95
+ "src/rl_pt_analyze/pixel_vec.py": "f247a6d9ed1791b82880ecef7505275959835fc3b69bfab2e85dedd096416bc8",
96
+ "src/rl_pt_analyze/r0.py": "e0a13e9c02d935dbab6a0e091fd925f08ce2d00e661b6cf09aff0ee95925249d",
97
+ "third_party/dice/model/diffusion/modules.py": "3b6b2efae0bb67cc30a061a8dd7c6b2d8d6905dac7cecac92b3dbfa5da0067cd",
98
+ "third_party/dice/model/flow_matching/vision_resnet_mlp_flow_matching.py": "2f0d3aa0b6e1624e32a540e66a5803e613481f35bb190926e51560856249981d"
99
+ },
100
+ "spec": {
101
+ "action_dim": 24,
102
+ "architecture": "dice_vision_resnet_flow_unet_v1",
103
+ "backbone_initialization": "random",
104
+ "backbone_norm": "batch_norm",
105
+ "cameras": [
106
+ "agentview",
107
+ "robot0_eye_in_hand",
108
+ "robot1_eye_in_hand"
109
+ ],
110
+ "collision_geoms": false,
111
+ "crop_size": 84,
112
+ "family": "StandardBaseBC",
113
+ "feature_dim": 166,
114
+ "features_per_camera": 64,
115
+ "flow_steps": 10,
116
+ "horizon": 8,
117
+ "image_size": 96,
118
+ "proprio_dim": 38,
119
+ "proprio_keys": [
120
+ "robot0_eef_pos",
121
+ "robot0_eef_quat",
122
+ "robot0_gripper_qpos",
123
+ "robot1_eef_pos",
124
+ "robot1_eef_quat",
125
+ "robot1_gripper_qpos"
126
+ ],
127
+ "task": "dmg_drawer_cleanup",
128
+ "time_embedding_dim": 64
129
+ },
130
+ "split_hash": "21ccaea9a04cc7084ab9f8d1b6c1e75a6c266d5c66fa77120c78ce32915ea59b",
131
+ "task": "dmg_drawer_cleanup",
132
+ "training": {
133
+ "batch_size": 256,
134
+ "bc_only": true,
135
+ "drop_last": false,
136
+ "early_stopping": false,
137
+ "ema_decay": 0.999,
138
+ "epoch_definition": "shuffled_full_training_split_without_replacement",
139
+ "epochs": 3000,
140
+ "learning_rate": 0.0001,
141
+ "online_steps": 0,
142
+ "optimizer": "adamw",
143
+ "padding_loss_mask": true,
144
+ "stopping": "none_fixed_epochs",
145
+ "weight_decay": 1e-06
146
+ },
147
+ "training_samples": 13354,
148
+ "validation": {
149
+ "batch_seed": 771,
150
+ "batch_size": 256,
151
+ "interval_epochs": 1,
152
+ "loss_seed": 882
153
+ },
154
+ "validation_only": false,
155
+ "validation_samples": 1501,
156
+ "wandb": {
157
+ "mode": "offline",
158
+ "project": "standard_bc_pretrain"
159
+ }
160
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/selection_complete.json ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 2,
3
+ "job": "dmg_drawer_cleanup/diverse50",
4
+ "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
5
+ "status": "selected_after_joint_review",
6
+ "decision": "retain_existing_candidate",
7
+ "selected_epoch": 1050,
8
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
9
+ "stopped_epoch": 2001,
10
+ "first_phase": {
11
+ "target_epoch": 2000,
12
+ "stopped": "2026-10-26T08:31:50Z",
13
+ "status": "awaiting_joint_curve_review",
14
+ "native_epoch": 2001,
15
+ "historical_selection_preserved": true,
16
+ "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_2000.pkl",
17
+ "sha256": "bd73a82ecfb7aa8157292f8f819a95c9921a5d8caebd5dbdc6470f8b0cb116d9",
18
+ "resume_epoch": 2000
19
+ },
20
+ "criterion": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
21
+ "train_loss_evidence": "Train mean e951-1000 0.0677247 -> e1001-1050 0.0664753 -> e1051-1100 0.0655113; improvement has slowed to about1.5-1.8% per50 epochs.",
22
+ "validation_loss_evidence": "Validation mean e951-1000 0.143344 -> e1001-1050 0.145048 -> e1051-1100 0.146439; e1801-1950 0.162890. Prefer e1050 to later more overfit weights.",
23
+ "success_evidence": "Periodic success e1020/1050/1080=62/64/56%, window60.67%. Later150-epoch blocks55.6/54/58/55.33/56.67/56.67%, endpoint56%. No sustained later gain.",
24
+ "alternatives": "e1000 neighborhood60%; e1550 neighborhood59.33% with higher validation loss. Retain e1050 without consulting final200.",
25
+ "conclusion": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
26
+ "limitations": [
27
+ "Validation already exceeds its minimum at this point; it is not accurate to claim completely absent overfitting.",
28
+ "The train curve has slowed, not reached a mathematically zero slope.",
29
+ "Periodic windows reuse seeds and sometimes differ in temporal span; final200 is reporting-only and does not change this selection."
30
+ ],
31
+ "final200_result": "/home/nhy/RL-postTrain/repo_visual/runs/operations/20261025_visual_bc_queue/review_20261025/final200/dmg_drawer_cleanup_diverse50/result.json",
32
+ "final200_role": "report_only_selection_frozen_before_evaluation",
33
+ "assessment": {
34
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
35
+ "window": 50,
36
+ "evaluation_schedule": {
37
+ "schema_version": 1,
38
+ "job": "dmg_drawer_cleanup/diverse50",
39
+ "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
40
+ "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
41
+ "recorded_at": "2026-10-26T04:57:55.268092+00:00",
42
+ "original_interval_epochs": 30,
43
+ "interval_epochs": 50,
44
+ "effective_after_epoch": 1350,
45
+ "resume_sha256_at_change": "e64fef31c951965087a4ea6118ff08623fce179d1b4aa5ea8a362c0fbf7391fa",
46
+ "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
47
+ "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
48
+ "validation_loss_interval_epochs": 1,
49
+ "periodic_episodes": 50,
50
+ "original_job_preserved": true,
51
+ "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
52
+ },
53
+ "evaluated_epoch": 2000,
54
+ "saved": 40,
55
+ "best": {
56
+ "epoch": 1050,
57
+ "window_success": 0.6066666666666667,
58
+ "episodes": 150,
59
+ "window_epochs": [
60
+ 1020,
61
+ 1050,
62
+ 1080
63
+ ],
64
+ "interval_epochs": 30,
65
+ "window_radius": 45,
66
+ "success_at_epoch": 0.64
67
+ },
68
+ "blocks": [
69
+ {
70
+ "epochs": [
71
+ 1,
72
+ 150
73
+ ],
74
+ "success": 0.228
75
+ },
76
+ {
77
+ "epochs": [
78
+ 151,
79
+ 300
80
+ ],
81
+ "success": 0.404
82
+ },
83
+ {
84
+ "epochs": [
85
+ 301,
86
+ 450
87
+ ],
88
+ "success": 0.496
89
+ },
90
+ {
91
+ "epochs": [
92
+ 451,
93
+ 600
94
+ ],
95
+ "success": 0.484
96
+ },
97
+ {
98
+ "epochs": [
99
+ 601,
100
+ 750
101
+ ],
102
+ "success": 0.564
103
+ },
104
+ {
105
+ "epochs": [
106
+ 751,
107
+ 900
108
+ ],
109
+ "success": 0.536
110
+ },
111
+ {
112
+ "epochs": [
113
+ 901,
114
+ 1050
115
+ ],
116
+ "success": 0.588
117
+ },
118
+ {
119
+ "epochs": [
120
+ 1051,
121
+ 1200
122
+ ],
123
+ "success": 0.556
124
+ },
125
+ {
126
+ "epochs": [
127
+ 1201,
128
+ 1350
129
+ ],
130
+ "success": 0.54
131
+ },
132
+ {
133
+ "epochs": [
134
+ 1351,
135
+ 1500
136
+ ],
137
+ "success": 0.58
138
+ },
139
+ {
140
+ "epochs": [
141
+ 1501,
142
+ 1650
143
+ ],
144
+ "success": 0.5533333333333333
145
+ },
146
+ {
147
+ "epochs": [
148
+ 1651,
149
+ 1800
150
+ ],
151
+ "success": 0.5666666666666667
152
+ },
153
+ {
154
+ "epochs": [
155
+ 1801,
156
+ 1950
157
+ ],
158
+ "success": 0.5666666666666667
159
+ }
160
+ ],
161
+ "plateau": true,
162
+ "loss_trends": {
163
+ "train": {
164
+ "last_epoch": 2001,
165
+ "preceding_50_mean": 0.0558663970049623,
166
+ "recent_50_mean": 0.05525940129480686,
167
+ "relative_change": -0.010865130788755284,
168
+ "complete_100_epoch_window": true
169
+ },
170
+ "validation": {
171
+ "last_epoch": 2001,
172
+ "preceding_50_mean": 0.1629074254631996,
173
+ "recent_50_mean": 0.16240808933973314,
174
+ "relative_change": -0.0030651526291493036,
175
+ "complete_100_epoch_window": true
176
+ }
177
+ },
178
+ "validation_min": {
179
+ "epoch": 284,
180
+ "loss": 0.12474054843187332
181
+ },
182
+ "validation_at_best": 0.14491789042949677,
183
+ "validation_last": 0.1636236011981964
184
+ },
185
+ "reuse_prior_final200": true
186
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/selection_supervision.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "selected_after_joint_review",
3
+ "training": "stopped",
4
+ "joint_curve_review_status": "finalized_after_phase2000_review",
5
+ "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
6
+ "review_evidence": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json",
7
+ "joint_curve_review_reason": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
8
+ "final200_role": "report_only_not_used_for_selection",
9
+ "zero_overfitting_claim": false,
10
+ "completed_train_epoch": 2001,
11
+ "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
12
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/split.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "duplicates_removed": [],
3
+ "manifest_hash": "57937c542aa6790d0dacfb9e3fbae03147553d3ecea84c4b61692944aa942199",
4
+ "provisional": false,
5
+ "seed": 73119,
6
+ "training": [
7
+ "mg/demo_923",
8
+ "mg/demo_53",
9
+ "mg/demo_649",
10
+ "mg/demo_49",
11
+ "mg/demo_850",
12
+ "mg/demo_1014",
13
+ "mg/demo_461",
14
+ "mg/demo_252",
15
+ "mg/demo_465",
16
+ "mg/demo_586",
17
+ "mg/demo_551",
18
+ "mg/demo_844",
19
+ "mg/demo_364",
20
+ "mg/demo_637",
21
+ "mg/demo_376",
22
+ "mg/demo_475",
23
+ "mg/demo_255",
24
+ "mg/demo_5",
25
+ "mg/demo_742",
26
+ "mg/demo_241",
27
+ "mg/demo_847",
28
+ "mg/demo_105",
29
+ "mg/demo_356",
30
+ "mg/demo_674",
31
+ "mg/demo_580",
32
+ "mg/demo_879",
33
+ "mg/demo_515",
34
+ "mg/demo_643",
35
+ "mg/demo_320",
36
+ "mg/demo_145",
37
+ "mg/demo_758",
38
+ "mg/demo_640",
39
+ "mg/demo_481",
40
+ "mg/demo_696",
41
+ "mg/demo_140",
42
+ "mg/demo_880",
43
+ "mg/demo_102",
44
+ "mg/demo_896",
45
+ "mg/demo_636",
46
+ "mg/demo_992",
47
+ "mg/demo_797",
48
+ "mg/demo_942",
49
+ "mg/demo_664",
50
+ "mg/demo_791",
51
+ "mg/demo_1001"
52
+ ],
53
+ "unit": "whole_trajectory",
54
+ "validation": [
55
+ "mg/demo_85",
56
+ "mg/demo_556",
57
+ "mg/demo_407",
58
+ "mg/demo_521",
59
+ "mg/demo_912"
60
+ ]
61
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/train_metrics.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/training_status.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "status": "selected_after_joint_review",
3
+ "selected_epoch": 1050,
4
+ "trained_epoch": 2001,
5
+ "phase_complete": true
6
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400/validation_metrics.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json CHANGED
@@ -1,186 +1,43 @@
1
  {
2
- "schema_version": 2,
3
  "job": "dmg_drawer_cleanup/diverse50",
4
- "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
5
  "status": "selected_after_joint_review",
6
- "decision": "retain_existing_candidate",
7
- "selected_epoch": 1050,
8
- "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
9
- "stopped_epoch": 2001,
10
- "first_phase": {
11
- "target_epoch": 2000,
12
- "stopped": "2026-10-26T08:31:50Z",
13
- "status": "awaiting_joint_curve_review",
14
- "native_epoch": 2001,
15
- "historical_selection_preserved": true,
16
- "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1/checkpoints/epoch_2000.pkl",
17
- "sha256": "bd73a82ecfb7aa8157292f8f819a95c9921a5d8caebd5dbdc6470f8b0cb116d9",
18
- "resume_epoch": 2000
19
  },
20
- "criterion": "Joint practical review: training-loss gains substantially slowed, high periodic success before sustained lower performance, and avoidance of substantially worse later validation. No claim of zero overfitting or a global optimum.",
21
- "train_loss_evidence": "Train mean e951-1000 0.0677247 -> e1001-1050 0.0664753 -> e1051-1100 0.0655113; improvement has slowed to about1.5-1.8% per50 epochs.",
22
- "validation_loss_evidence": "Validation mean e951-1000 0.143344 -> e1001-1050 0.145048 -> e1051-1100 0.146439; e1801-1950 0.162890. Prefer e1050 to later more overfit weights.",
23
- "success_evidence": "Periodic success e1020/1050/1080=62/64/56%, window60.67%. Later150-epoch blocks55.6/54/58/55.33/56.67/56.67%, endpoint56%. No sustained later gain.",
24
- "alternatives": "e1000 neighborhood60%; e1550 neighborhood59.33% with higher validation loss. Retain e1050 without consulting final200.",
25
- "conclusion": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
26
- "limitations": [
27
- "Validation already exceeds its minimum at this point; it is not accurate to claim completely absent overfitting.",
28
- "The train curve has slowed, not reached a mathematically zero slope.",
29
- "Periodic windows reuse seeds and sometimes differ in temporal span; final200 is reporting-only and does not change this selection."
30
- ],
31
- "final200_result": "/home/nhy/RL-postTrain/repo_visual/runs/operations/20261025_visual_bc_queue/review_20261025/final200/dmg_drawer_cleanup_diverse50/result.json",
32
- "final200_role": "report_only_selection_frozen_before_evaluation",
33
- "assessment": {
34
- "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/diverse50/seed_1",
35
- "window": 50,
36
- "evaluation_schedule": {
37
- "schema_version": 1,
38
- "job": "dmg_drawer_cleanup/diverse50",
39
- "identity_hash": "ede9860b9ac43a4d914535839be74a84c2a7288c9f07ee64f893f65d1bc11b97",
40
- "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
41
- "recorded_at": "2026-10-26T04:57:55.268092+00:00",
42
- "original_interval_epochs": 30,
43
- "interval_epochs": 50,
44
- "effective_after_epoch": 1350,
45
- "resume_sha256_at_change": "e64fef31c951965087a4ea6118ff08623fce179d1b4aa5ea8a362c0fbf7391fa",
46
- "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
47
- "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
48
- "validation_loss_interval_epochs": 1,
49
- "periodic_episodes": 50,
50
- "original_job_preserved": true,
51
- "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
52
- },
53
- "evaluated_epoch": 2000,
54
- "saved": 40,
55
- "best": {
56
- "epoch": 1050,
57
- "window_success": 0.6066666666666667,
58
- "episodes": 150,
59
- "window_epochs": [
60
- 1020,
61
- 1050,
62
- 1080
63
- ],
64
- "interval_epochs": 30,
65
- "window_radius": 45,
66
- "success_at_epoch": 0.64
67
- },
68
- "blocks": [
69
- {
70
- "epochs": [
71
- 1,
72
- 150
73
- ],
74
- "success": 0.228
75
- },
76
- {
77
- "epochs": [
78
- 151,
79
- 300
80
- ],
81
- "success": 0.404
82
- },
83
- {
84
- "epochs": [
85
- 301,
86
- 450
87
- ],
88
- "success": 0.496
89
- },
90
- {
91
- "epochs": [
92
- 451,
93
- 600
94
- ],
95
- "success": 0.484
96
- },
97
- {
98
- "epochs": [
99
- 601,
100
- 750
101
- ],
102
- "success": 0.564
103
- },
104
- {
105
- "epochs": [
106
- 751,
107
- 900
108
- ],
109
- "success": 0.536
110
- },
111
- {
112
- "epochs": [
113
- 901,
114
- 1050
115
- ],
116
- "success": 0.588
117
- },
118
- {
119
- "epochs": [
120
- 1051,
121
- 1200
122
- ],
123
- "success": 0.556
124
- },
125
- {
126
- "epochs": [
127
- 1201,
128
- 1350
129
- ],
130
- "success": 0.54
131
- },
132
- {
133
- "epochs": [
134
- 1351,
135
- 1500
136
- ],
137
- "success": 0.58
138
- },
139
- {
140
- "epochs": [
141
- 1501,
142
- 1650
143
- ],
144
- "success": 0.5533333333333333
145
- },
146
- {
147
- "epochs": [
148
- 1651,
149
- 1800
150
- ],
151
- "success": 0.5666666666666667
152
- },
153
- {
154
- "epochs": [
155
- 1801,
156
- 1950
157
- ],
158
- "success": 0.5666666666666667
159
- }
160
- ],
161
- "plateau": true,
162
- "loss_trends": {
163
- "train": {
164
- "last_epoch": 2001,
165
- "preceding_50_mean": 0.0558663970049623,
166
- "recent_50_mean": 0.05525940129480686,
167
- "relative_change": -0.010865130788755284,
168
- "complete_100_epoch_window": true
169
- },
170
- "validation": {
171
- "last_epoch": 2001,
172
- "preceding_50_mean": 0.1629074254631996,
173
- "recent_50_mean": 0.16240808933973314,
174
- "relative_change": -0.0030651526291493036,
175
- "complete_100_epoch_window": true
176
- }
177
- },
178
- "validation_min": {
179
- "epoch": 284,
180
- "loss": 0.12474054843187332
181
  },
182
- "validation_at_best": 0.14491789042949677,
183
- "validation_last": 0.1636236011981964
184
  },
185
- "reuse_prior_final200": true
 
 
 
 
 
186
  }
 
1
  {
2
+ "schema_version": 4,
3
  "job": "dmg_drawer_cleanup/diverse50",
4
+ "reviewed_at": "2026-10-27T08:29:29.361939+00:00",
5
  "status": "selected_after_joint_review",
6
+ "decision": "user_uniform_e1400_20261027",
7
+ "selected_epoch": 1400,
8
+ "checkpoint_sha256": "06dad7e3f2ab09f5ee38125e4debd18900425dbf1429d2d98e8911d3c3fba383",
9
+ "trained_epoch_at_freeze": 2001,
10
+ "final200": {
11
+ "success": 0.57,
12
+ "successes": 114,
13
+ "total": 200,
14
+ "ci_low": 0.5007048776415268,
15
+ "ci_high": 0.6366567766656557
 
 
 
16
  },
17
+ "paired_vs_prior_on_final200": {
18
+ "prior_epoch": 1050,
19
+ "prior_successes": 122,
20
+ "e1400_successes": 114,
21
+ "prior_only": 38,
22
+ "e1400_only": 30,
23
+ "exact_mcnemar_p": 0.39612484029702977
24
+ },
25
+ "prior": {
26
+ "epoch": 1050,
27
+ "checkpoint_sha256": "bcc2a60da8af3d34c6ee495984784710a69a55ae2c186ea8cca4d31b56fc56d5",
28
+ "final200": {
29
+ "success": 0.61,
30
+ "successes": 122,
31
+ "total": 200,
32
+ "ci_low": 0.5409374595716976,
33
+ "ci_high": 0.6749165686253038
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
34
  },
35
+ "archive": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400"
 
36
  },
37
+ "note": "Trained to the e2000 first-phase boundary; e1400 is an intermediate saved checkpoint.",
38
+ "limitations": [
39
+ "Single training seed.",
40
+ "Final200 reused the prior selection panel and was seen before the decision.",
41
+ "Zero overfitting not established."
42
+ ]
43
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_supervision.json CHANGED
@@ -1,12 +1,13 @@
1
  {
2
  "status": "selected_after_joint_review",
3
- "training": "stopped",
4
- "joint_curve_review_status": "finalized_after_phase2000_review",
5
- "reviewed_at": "2026-10-26T20:18:15.212073+00:00",
6
- "review_evidence": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/selection_complete.json",
7
- "joint_curve_review_reason": "Select saved e1050 under the trained/high-success/before-later-degradation criterion; keep all earlier releases as historical evidence.",
8
- "final200_role": "report_only_not_used_for_selection",
9
  "zero_overfitting_claim": false,
10
  "completed_train_epoch": 2001,
11
- "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261026_e1050"
 
 
12
  }
 
1
  {
2
  "status": "selected_after_joint_review",
3
+ "decision": "user_uniform_e1400_20261027",
4
+ "reviewed_at": "2026-10-27T08:29:29.361939+00:00",
5
+ "joint_curve_review_status": "replaced_by_user_uniform_epoch",
6
+ "joint_curve_review_reason": "User 2026-10-27: use epoch 1400 for every locally owned visual BC, after seeing e1400 vs prior-selection final200 on the same 200 seeds.",
7
+ "final200_role": "decision_evidence_same_panel_as_prior_selection_not_independent",
 
8
  "zero_overfitting_claim": false,
9
  "completed_train_epoch": 2001,
10
+ "training_note": "Trained to the e2000 first-phase boundary; e1400 is an intermediate saved checkpoint.",
11
+ "review_evidence": "visual_no_collision/reviews/20261027_uniform_e1400/comparison.json",
12
+ "historical_release": "visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/releases/before_20261027_e1400"
13
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/diverse50/seed_1/training_status.json CHANGED
@@ -1,6 +1,5 @@
1
  {
2
  "status": "selected_after_joint_review",
3
- "selected_epoch": 1050,
4
- "trained_epoch": 2001,
5
- "phase_complete": true
6
  }
 
1
  {
2
  "status": "selected_after_joint_review",
3
+ "selected_epoch": 1400,
4
+ "trained_epoch": 2001
 
5
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/README.md CHANGED
@@ -1,19 +1,7 @@
1
- # DrawerCleanup mixed50: reviewed visual-only BC
2
 
3
- Selected **epoch 1100**, complete EMA encoder and flow policy; collision_geoms=false. The 2000-epoch training phase is complete.
4
 
5
- At e1100 the trailing50 training loss is0.0621518 versus0.0630162 in the preceding50, a1.37% decrease; the next50 improves by about1%. Training has clearly slowed, though it has not reached a zero slope. e1800/e2000 trailing50 means are0.0550184/0.0536351 with further decreases of0.71%/0.93%.
6
 
7
- Trailing50 validation loss is0.141649 at e1100 versus0.150305 at e1800 and0.152317 at e2000 (+6.11%/+7.53%). All exceed the early minimum0.124569 at e284. Select the earlier mature point to avoid this additional validation deterioration; zero overfitting is not claimed.
8
-
9
- On the predeclared common100 candidate-validation seeds, e1100/e1800/e2000 score65/69/71%. Paired e1100 vs e1800 has9 earlier-only and13 later-only successes; vs e2000 has6 earlier-only and12 later-only. Holm-adjusted p values are1.0 and0.71365. These differences do not establish a clear later advantage, but absence of significance does not prove equivalent performance. The e1100 periodic neighborhood is67.33%; late periodic success remains high and improves modestly, with no demonstrated sustained post-peak decline.
10
-
11
- Use the predeclared preference for earlier, lower-validation-loss mature weights unless a later point establishes a clear paired success advantage. e1800/e2000 have higher point estimates and may truly perform better; choosing e1100 explicitly prioritizes the user's avoidance of unnecessary overfitting over an uncertain4-6 point validation-panel gain. No universal optimum or later success collapse is asserted.
12
-
13
- This is a practical choice among the reviewed checkpoints. Validation loss already exceeds its historical minimum; zero overfitting and global optimality are not established. Nonsignificant paired differences do not prove equivalence.
14
-
15
- Independent final200: **135/200 = 67.5%**, Wilson 95% CI [60.7%, 73.6%]. This panel was run only after freezing the choice, and never drives reselection. Candidate-validation100 is a separate selection panel and is not an unbiased final performance report.
16
-
17
- SHA256: `8546111eaa19cf058caa0f5d8e52c690fa7ff2e623664a4f0975737ebe78ab1b`.
18
-
19
- See `selection_complete.json`, `selection_validation/` (all three candidates), and `final200/`. Original data/split/config contracts and the recorded evaluation schedule are preserved. `SHA256SUMS.json` covers this release.
 
1
+ # DrawerCleanup mixed50: visual-only BC, epoch 1400
2
 
3
+ Selected **epoch 1400** by user decision (2026-10-27): every locally trained visual BC uses e1400. Complete EMA encoder and flow policy; collision_geoms=false. `release.json` names the canonical checkpoint; any other checkpoint left in `checkpoints/` belongs to an archived release.
4
 
5
+ Final200: **141/200 = 70.5%**, Wilson 95% CI [63.8%, 76.4%]. Same panel, prior selection e1100: 135/200 vs e1400 141/200 (only prior 14, only e1400 20, exact McNemar p=0.39). This panel was seen before the decision, so it is not an independent test. Trained to the e2000 first-phase boundary; e1400 is an intermediate saved checkpoint. Single training seed; zero overfitting is not established.
6
 
7
+ SHA256: `647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d`. The prior release is preserved under [releases/before_20261027_e1400](releases/before_20261027_e1400/README.md).
 
 
 
 
 
 
 
 
 
 
 
 
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/SHA256SUMS.json CHANGED
@@ -1,20 +1,12 @@
1
  {
2
- "checkpoints/epoch_1100.pkl": {
3
- "sha256": "8546111eaa19cf058caa0f5d8e52c690fa7ff2e623664a4f0975737ebe78ab1b",
4
- "bytes": 138495801
5
  },
6
- "checkpoints/epoch_1100.json": {
7
- "sha256": "4bfef458b25e04f6ef3aa3654adde9dff96dc870674a6c551b3ed9354f5389db",
8
  "bytes": 7960
9
  },
10
- "best.json": {
11
- "sha256": "3668cd0c0562388c4842c14338172ab4051713c64b5520a10d8b23199309e1fa",
12
- "bytes": 4038
13
- },
14
- "selection_complete.json": {
15
- "sha256": "b21eb0fc670b7c3e8ad58028c64afcfc0d7838ee6c36155256675127f494e1ac",
16
- "bytes": 11953
17
- },
18
  "resolved_config.json": {
19
  "sha256": "21c94c3b961e3352462926a4fa8b0699be832f3f41601aa497eb198e8b788334",
20
  "bytes": 6613
@@ -43,112 +35,64 @@
43
  "sha256": "968e3132b063687bfc8c8db4e5f31df3cf28ea474b317a4d7e5daf1cb7015939",
44
  "bytes": 84712
45
  },
46
- "first_phase_complete.json": {
47
- "sha256": "2ea5bed7c24ce51e4fb08df1e86f5687292cbb3f7e577345f95a15b5789f7fda",
48
- "bytes": 521
49
- },
50
  "provenance/native_training_status.json": {
51
  "sha256": "a408ce110aeaf0cdf45664b24f5d2bbda6db12aca67c41b43298e3d1ff9faf41",
52
  "bytes": 333
53
  },
54
- "evaluation_schedule.json": {
55
- "sha256": "c532a7a18c452b9e1e1fa71a7a25662b2ecc635fa49d0cdf48139f508ed3f133",
56
- "bytes": 1007
57
- },
58
- "provenance/evaluation_schedule_active.json": {
59
- "sha256": "b8489cee19b0995edeb03d9b611a472b3daef2b774e88abdf3a3b3da26205c31",
60
- "bytes": 455
61
- },
62
- "provenance/effective_train_eval50.py": {
63
- "sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
64
- "bytes": 9987
65
- },
66
- "provenance/bc_eval_schedule.py": {
67
- "sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
68
- "bytes": 3539
69
- },
70
  "provenance/train_bc_noleak.py": {
71
  "sha256": "587b66eca4c82c8ee519ca7df23874ca9a9acdbb8cdabd8a36afb68ebcf891d5",
72
  "bytes": 2186
73
  },
74
- "provenance/evaluate_panel.py": {
75
- "sha256": "051ff2a6fdb3b5a11a97b3bc6fbf9aa2f2ef5203b7f68a37a4cc75ee7f0484f4",
76
- "bytes": 4288
77
  },
78
  "final200/result.json": {
79
- "sha256": "b338fb41ecc2fb4e43e3d71ca58011231e6156d424683a3272f0b180512f60f6",
80
- "bytes": 119176
81
  },
82
  "final200/protocol.json": {
83
- "sha256": "b720bbbdc798dd7a3699c48b6978ffb9bc1fe55b0fad1a05fbc63e4ec8e42429",
84
- "bytes": 4104
85
- },
86
- "final200/epoch_1100.parquet": {
87
- "sha256": "47e1b6437509a8c548315e3a134757bf4504864defc10af1c80be43ebc9e4b48",
88
- "bytes": 6499
89
- },
90
- "selection_validation/plan.json": {
91
- "sha256": "131bf4cbd04ed3cfd77b23b5aedd4b52a0d68815bc242a8e24054b2cab0e8c76",
92
- "bytes": 8433
93
  },
94
- "selection_validation/comparison.json": {
95
- "sha256": "000130473ee5a5751f11756e12a4e2e12f6f8ac2a92b41f6786b5d5665c3b5aa",
96
- "bytes": 9813
97
  },
98
- "selection_validation/frozen_selection.json": {
99
- "sha256": "aab9b2adffe217c850afd08c0c2ce773db4884cac4ffe589f71a81195085b1b0",
100
- "bytes": 16333
101
- },
102
- "selection_validation/epoch_1100/result.json": {
103
- "sha256": "509b6c7ce1b945b8c9db76a75e8b5e49f575924cbaac16d57349354fb3145959",
104
- "bytes": 61803
105
- },
106
- "selection_validation/epoch_1100/protocol.json": {
107
- "sha256": "b09f74b8acb2404a728cb295b3804bc340d4389efcfa8b007cf11a116b59d65f",
108
- "bytes": 3986
109
- },
110
- "selection_validation/epoch_1100/episodes.parquet": {
111
- "sha256": "fe830118532e34d4e711d09f3ac387858835c2f7cbf37960bd3f312f0ccc64f8",
112
- "bytes": 5639
113
- },
114
- "selection_validation/epoch_1800/result.json": {
115
- "sha256": "3a90d1fb1410f0a102c5ce2709399db8bb861ac9ca2d033b8dedba93ddbed1e0",
116
- "bytes": 61795
117
- },
118
- "selection_validation/epoch_1800/protocol.json": {
119
- "sha256": "d8da4459361bed46d898418f4f6a9ac6e140f475ab4b7de3be1c477f7e307454",
120
- "bytes": 3986
121
- },
122
- "selection_validation/epoch_1800/episodes.parquet": {
123
- "sha256": "7c26aeddeb75c5a8e7c827ac5c9ef71c3a20706325353752cace7b846cd50b54",
124
- "bytes": 5651
125
- },
126
- "selection_validation/epoch_2000/result.json": {
127
- "sha256": "038441fc606ffdb761fa085997a86646d66bb9540464a255dd63f26b323b87d5",
128
- "bytes": 61791
129
  },
130
- "selection_validation/epoch_2000/protocol.json": {
131
- "sha256": "70bad58ae5acd146578da403d29ed34085951fff2eae96d70cbf2b1c5606c90e",
132
- "bytes": 3986
133
  },
134
- "selection_validation/epoch_2000/episodes.parquet": {
135
- "sha256": "dee49cd5e4b5781fb19f1d9f9bfa148030031357c40391cd7f15bf13bf4cdff0",
136
- "bytes": 5547
137
  },
138
  "release.json": {
139
- "sha256": "e9cd549ac18856aeefc5e0e69817272baded412e2ce141932365001858846155",
140
- "bytes": 2808
141
  },
142
  "selection_supervision.json": {
143
- "sha256": "2907e1ae1d42378388dfe93e93e92a08b196e098f17dbd38da95d0d3f2c8028d",
144
- "bytes": 679
 
 
 
 
145
  },
146
  "training_status.json": {
147
- "sha256": "a00279282476a5894b0d9a6ef7c7b5a988a7d733639d7575787e6f4d463a7a40",
148
- "bytes": 123
 
 
 
 
149
  },
150
  "README.md": {
151
- "sha256": "4e35c3205400891ddf6224c20e2fee6be6404320635dab0243a2372943b823ed",
152
- "bytes": 2481
153
  }
154
  }
 
1
  {
2
+ "checkpoints/epoch_1400.pkl": {
3
+ "sha256": "647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d",
4
+ "bytes": 138484087
5
  },
6
+ "checkpoints/epoch_1400.json": {
7
+ "sha256": "fa2095016ee253f7443ae4a55439748496774959b4a228ddfbab4453d8dd0554",
8
  "bytes": 7960
9
  },
 
 
 
 
 
 
 
 
10
  "resolved_config.json": {
11
  "sha256": "21c94c3b961e3352462926a4fa8b0699be832f3f41601aa497eb198e8b788334",
12
  "bytes": 6613
 
35
  "sha256": "968e3132b063687bfc8c8db4e5f31df3cf28ea474b317a4d7e5daf1cb7015939",
36
  "bytes": 84712
37
  },
 
 
 
 
38
  "provenance/native_training_status.json": {
39
  "sha256": "a408ce110aeaf0cdf45664b24f5d2bbda6db12aca67c41b43298e3d1ff9faf41",
40
  "bytes": 333
41
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
42
  "provenance/train_bc_noleak.py": {
43
  "sha256": "587b66eca4c82c8ee519ca7df23874ca9a9acdbb8cdabd8a36afb68ebcf891d5",
44
  "bytes": 2186
45
  },
46
+ "provenance/evaluate_best.py": {
47
+ "sha256": "7be8b5440b354a862228f5a3d3855ec0ba7267c78c4e04158fa028e9633e8d32",
48
+ "bytes": 3370
49
  },
50
  "final200/result.json": {
51
+ "sha256": "759679cad67cd7467eadfa6e1490804c6e0fec538d98dc336087286c83bc0196",
52
+ "bytes": 119040
53
  },
54
  "final200/protocol.json": {
55
+ "sha256": "08d3e8d489b401a4751f8ede5ab0ede5e31e4c0c3ce24373af819290048cc5c1",
56
+ "bytes": 3991
 
 
 
 
 
 
 
 
57
  },
58
+ "final200/epoch_1400.parquet": {
59
+ "sha256": "42bb62ca863eae57ca06cb337cd83963c77d619949bebbb682fd8d6f30c7b862",
60
+ "bytes": 6543
61
  },
62
+ "evaluation_schedule.json": {
63
+ "sha256": "c532a7a18c452b9e1e1fa71a7a25662b2ecc635fa49d0cdf48139f508ed3f133",
64
+ "bytes": 1007
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65
  },
66
+ "provenance/evaluation_schedule_active.json": {
67
+ "sha256": "b8489cee19b0995edeb03d9b611a472b3daef2b774e88abdf3a3b3da26205c31",
68
+ "bytes": 455
69
  },
70
+ "provenance/effective_train_eval50.py": {
71
+ "sha256": "71544fc66ca6e1a0258ab078f6dcf4567677b8a1e0d3e16cdddedb02cac23d9f",
72
+ "bytes": 9987
73
  },
74
  "release.json": {
75
+ "sha256": "a79aee3605ea43a7542b5798559cbff6a9a7963825f0d10559bb0adbf6206c60",
76
+ "bytes": 3057
77
  },
78
  "selection_supervision.json": {
79
+ "sha256": "cc23800b0b26700a75ffaf7271ed5e28236c6405963fb437d11e258d3469d005",
80
+ "bytes": 859
81
+ },
82
+ "selection_complete.json": {
83
+ "sha256": "6dc81d331fdefcd954954c2705083c38e391bb1050dbf8b5942767258cf554e5",
84
+ "bytes": 1403
85
  },
86
  "training_status.json": {
87
+ "sha256": "e9ab55fb532bf907660519a68e88da8d260ae394b0b27894caca201e8fac1897",
88
+ "bytes": 97
89
+ },
90
+ "best.json": {
91
+ "sha256": "6e2f733ad727ec5db7c5d4cbb38caa73492a6dfe847623fc9c60312d93c85ae8",
92
+ "bytes": 303
93
  },
94
  "README.md": {
95
+ "sha256": "f0f621c8c5c888b3ab4682868908c7970fbbd58d8639f38fa653c8213a375e26",
96
+ "bytes": 922
97
  }
98
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/best.json CHANGED
@@ -1,157 +1,7 @@
1
  {
2
  "job": "dmg_drawer_cleanup/mixed50",
3
- "stopped": "2026-10-26T21:20:57.382679+00:00",
4
  "reason": "selected_after_joint_review",
5
- "review_note": "Choose saved e1100 as the conservative mature/less-overfit practical checkpoint among the three reviewed candidates. Freeze it before independent final200; report that test without reselection.",
6
- "rule": "Joint practical choice among predeclared checkpoints: slowed training-loss improvement, candidate-validation success, and less unnecessary validation-loss deterioration. Not a guarantee of zero overfitting or globally optimal success.",
7
- "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/mixed50/seed_1",
8
- "window": 50,
9
- "evaluation_schedule": {
10
- "schema_version": 1,
11
- "job": "dmg_drawer_cleanup/mixed50",
12
- "identity_hash": "29fdfa802c2fbd37332749469930ecb2f211b0779cc747f6f99327ef6593bf0d",
13
- "authorization": "User 2026-10-26: pause remaining training, evaluate every50 epochs, then resume.",
14
- "recorded_at": "2026-10-26T04:57:55.268092+00:00",
15
- "original_interval_epochs": 30,
16
- "interval_epochs": 50,
17
- "effective_after_epoch": 1500,
18
- "resume_sha256_at_change": "fd1cf668367e3a192d8eace5e26e8b7ee9792d6c9ed14a128c12c4d3051ade11",
19
- "helper_sha256": "dcc816174f3539365f6d9b572336b649cf3b28e8c915b18e06c761b578843ed9",
20
- "native_train_source_sha256": "a8b7b0b35ca2d73dd1c05bf2a59c4e9bd1c80c64fd9f382885c372c4a70c9b1c",
21
- "validation_loss_interval_epochs": 1,
22
- "periodic_episodes": 50,
23
- "original_job_preserved": true,
24
- "note": "Explicit runtime evaluation trigger override; include this sidecar with future checkpoints. Training, optimizer/RNG, validation, checkpoint schedule, render settings and periodic reset/policy seeds unchanged."
25
- },
26
- "evaluated_epoch": 2000,
27
- "saved": 40,
28
- "best": {
29
- "epoch": 1100,
30
- "window_success": 0.6733333333333333,
31
- "episodes": 150,
32
- "window_epochs": [
33
- 1080,
34
- 1110,
35
- 1140
36
- ],
37
- "interval_epochs": 30,
38
- "window_radius": 45,
39
- "success_at_epoch": null
40
- },
41
- "blocks": [
42
- {
43
- "epochs": [
44
- 1,
45
- 150
46
- ],
47
- "success": 0.192
48
- },
49
- {
50
- "epochs": [
51
- 151,
52
- 300
53
- ],
54
- "success": 0.444
55
- },
56
- {
57
- "epochs": [
58
- 301,
59
- 450
60
- ],
61
- "success": 0.484
62
- },
63
- {
64
- "epochs": [
65
- 451,
66
- 600
67
- ],
68
- "success": 0.532
69
- },
70
- {
71
- "epochs": [
72
- 601,
73
- 750
74
- ],
75
- "success": 0.592
76
- },
77
- {
78
- "epochs": [
79
- 751,
80
- 900
81
- ],
82
- "success": 0.58
83
- },
84
- {
85
- "epochs": [
86
- 901,
87
- 1050
88
- ],
89
- "success": 0.628
90
- },
91
- {
92
- "epochs": [
93
- 1051,
94
- 1200
95
- ],
96
- "success": 0.632
97
- },
98
- {
99
- "epochs": [
100
- 1201,
101
- 1350
102
- ],
103
- "success": 0.636
104
- },
105
- {
106
- "epochs": [
107
- 1351,
108
- 1500
109
- ],
110
- "success": 0.628
111
- },
112
- {
113
- "epochs": [
114
- 1501,
115
- 1650
116
- ],
117
- "success": 0.6133333333333333
118
- },
119
- {
120
- "epochs": [
121
- 1651,
122
- 1800
123
- ],
124
- "success": 0.66
125
- },
126
- {
127
- "epochs": [
128
- 1801,
129
- 1950
130
- ],
131
- "success": 0.6666666666666666
132
- }
133
- ],
134
- "plateau": false,
135
- "loss_trends": {
136
- "train": {
137
- "last_epoch": 2013,
138
- "preceding_50_mean": 0.05399720574060297,
139
- "recent_50_mean": 0.0535615592600809,
140
- "relative_change": -0.008067944897276136,
141
- "complete_100_epoch_window": true
142
- },
143
- "validation": {
144
- "last_epoch": 2013,
145
- "preceding_50_mean": 0.15144177079200744,
146
- "recent_50_mean": 0.1523429837822914,
147
- "relative_change": 0.005950887826857931,
148
- "complete_100_epoch_window": true
149
- }
150
- },
151
- "validation_min": {
152
- "epoch": 284,
153
- "loss": 0.12456896901130676
154
- },
155
- "validation_at_best": 0.141687273979187,
156
- "validation_last": 0.1517714262008667
157
  }
 
1
  {
2
  "job": "dmg_drawer_cleanup/mixed50",
 
3
  "reason": "selected_after_joint_review",
4
+ "selected_epoch": 1400,
5
+ "checkpoint_sha256": "647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d",
6
+ "rule": "User decision 2026-10-27: a uniform epoch 1400 for all nine locally trained visual BCs."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/checkpoints/epoch_1400.json ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bc_identity": {
3
+ "architecture": "dice_vision_resnet_flow_unet_v1",
4
+ "family": "StandardBaseBC",
5
+ "framework": "jax"
6
+ },
7
+ "data_manifest_hash": "57937c542aa6790d0dacfb9e3fbae03147553d3ecea84c4b61692944aa942199",
8
+ "encoding": "gzip_pickle_numpy_float32_lossless",
9
+ "kind": "pixel_state_bc",
10
+ "metadata": {
11
+ "epoch": 1400,
12
+ "epochs": 3000,
13
+ "experiment": "dmg_drawer_cleanup",
14
+ "regime": "mixed50",
15
+ "seed": 1,
16
+ "smoke": false,
17
+ "stage": "fixed_epochs",
18
+ "step": 74200,
19
+ "task": "dmg_drawer_cleanup"
20
+ },
21
+ "normalizer_hash": "0d5d62bf8d055e444518ff18875b4265688f1daa9ae6892391572ddea8d83090",
22
+ "observation_mode": "pixel_state",
23
+ "pixel_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
24
+ "resolved_config": {
25
+ "bc_identity": {
26
+ "family": "StandardBaseBC",
27
+ "framework": "jax"
28
+ },
29
+ "checkpoint_encoding": "lossless_gzip_ema_only",
30
+ "checkpoint_interval_epochs": 50,
31
+ "ema_buffers": "copy_live_running_statistics",
32
+ "evaluation": {
33
+ "episodes": 50,
34
+ "final": {
35
+ "episodes": 200,
36
+ "seed_panel_seed": 20261021,
37
+ "seed_range": [
38
+ 10000000,
39
+ 90000000
40
+ ],
41
+ "seed_sampling": "uniform_without_replacement"
42
+ },
43
+ "interval_epochs": 30,
44
+ "policy_seed": 700001,
45
+ "reset_seed_start": 880000,
46
+ "workers": 8
47
+ },
48
+ "experiment": "dmg_drawer_cleanup",
49
+ "family": "StandardBaseBC",
50
+ "framework": "jax",
51
+ "identity_hash": "29fdfa802c2fbd37332749469930ecb2f211b0779cc747f6f99327ef6593bf0d",
52
+ "index": 1,
53
+ "manifest": "/home/nhy/RL-postTrain/repo_visual/data/bc/versions/dmg_distribution_v1_local_20261025/manifests/dmg_drawer_cleanup.json",
54
+ "manifest_sha256": "110e1f8373c49813ff466b7f24ac25347ae7564ff970efb01556e4fa05e57bba",
55
+ "model_spec_hash": "52d8261facbdebdad8d1e8a841c66b3241ccd3a723d78578b1f02a6661d4850b",
56
+ "normalizer_hash": "0d5d62bf8d055e444518ff18875b4265688f1daa9ae6892391572ddea8d83090",
57
+ "observation_mode": "pixel_state",
58
+ "output": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/mixed50/seed_1",
59
+ "pixel_cache": "/home/nhy/RL-postTrain/repo_visual/data/pixels/dmg_drawer_cleanup/visual_20261025",
60
+ "pixel_spec": {
61
+ "action_dim": 24,
62
+ "architecture": "dice_vision_resnet_flow_unet_v1",
63
+ "backbone_initialization": "random",
64
+ "backbone_norm": "batch_norm",
65
+ "cameras": [
66
+ "agentview",
67
+ "robot0_eye_in_hand",
68
+ "robot1_eye_in_hand"
69
+ ],
70
+ "collision_geoms": false,
71
+ "crop_size": 84,
72
+ "family": "StandardBaseBC",
73
+ "feature_dim": 166,
74
+ "features_per_camera": 64,
75
+ "flow_steps": 10,
76
+ "horizon": 8,
77
+ "image_size": 96,
78
+ "proprio_dim": 38,
79
+ "proprio_keys": [
80
+ "robot0_eef_pos",
81
+ "robot0_eef_quat",
82
+ "robot0_gripper_qpos",
83
+ "robot1_eef_pos",
84
+ "robot1_eef_quat",
85
+ "robot1_gripper_qpos"
86
+ ],
87
+ "task": "dmg_drawer_cleanup",
88
+ "time_embedding_dim": 64
89
+ },
90
+ "regime": "mixed50",
91
+ "runtime": {
92
+ "framework_version": "0.10.1",
93
+ "python": "/home/nhy/RL-postTrain/repo/.venvs/ogpo/bin/python",
94
+ "requested_device": "cuda:0"
95
+ },
96
+ "seed": 1,
97
+ "sources_sha256": {
98
+ "configs/bc_models.yaml": "ce58a903c4270328985ab8677babdaa12748ad18efc3749faed72425024e61c1",
99
+ "configs/bc_pretraining.yaml": "7d1b9be73f25aeb40cd342a3e3ca2543ea865b38b2abe8fb28a7da6588d419a6",
100
+ "configs/bc_pretraining_ph_full.yaml": "a93c7d4a50e9afa836d98a6193ba1bc2fb3827a259cf5de39f5dc7a2504d6728",
101
+ "configs/bc_standard_init_schema.json": "e2c70eee2a1ce8d1a812ec427d24ba3cc45519434331354318d97415fd559bc8",
102
+ "configs/study1.yaml": "35d065e242b205a4c99f8add2eab35bb85f80401f4e07d49f063a0eddf4de608",
103
+ "configs/study1_ph_full.yaml": "f9cebc4b66279d4ea7492f7a80a09f1adca7c1cae7b242374305be4f701e76ef",
104
+ "scripts/pretrain_bc.py": "9972b77f7b8ed964e7e04fa3ba3821b767f0b7a5d697625ed1c6f8fcdf990890",
105
+ "src/rl_pt_analyze/bc_evaluation.py": "7319c346f453263a99cdc5d6fd39e6a0a88783d2c42f68a312ee36ab6d656c6f",
106
+ "src/rl_pt_analyze/bc_fleet.py": "a49c24c66ec0bb1a5548d384deefda429ff1ca693cb19f1c99dbd661148d7ffc",
107
+ "src/rl_pt_analyze/bc_full_cut.py": "a8170de58bc0a16dd5c6358a6429d96afb5227e5701602806d68e42727f671bc",
108
+ "src/rl_pt_analyze/bc_identity.py": "f94c284c71df7ed06eaf4553297de53f2b801a5fac743beb3b98d1bc41314fe0",
109
+ "src/rl_pt_analyze/bc_policy.py": "aff3c4613a0dfc46d6491bdba6478261049c2c9631eca446fa8939a1e498fc30",
110
+ "src/rl_pt_analyze/bc_pretrain.py": "703b2e50fe53d62b23910b907d75c09cec49463e5e9e2c08d8f77d2b6b77a5bd",
111
+ "src/rl_pt_analyze/bc_runtime.py": "29d85733684176bd7328c01921ed1eb45857a914d8ee998e6df4fb0a5be5315d",
112
+ "src/rl_pt_analyze/bc_standard_jax.py": "9063d0d4160e972bb9f7b1acd9e508a2ea1118ceeed7647614c787abcb7b32ce",
113
+ "src/rl_pt_analyze/bc_standard_policy.py": "54d733549ec571f709d49698520f2461def5a20fc4146d155b3d56fbbcf5deb1",
114
+ "src/rl_pt_analyze/bc_storage.py": "83b79d5beaaaa46310d756476f5da90c9227a656533dc9414334d395a6749e54",
115
+ "src/rl_pt_analyze/bc_training.py": "bf20475ceeef114e3444c2610ea6d490389b02d147f22c962e2f63fd94b9dafa",
116
+ "src/rl_pt_analyze/pixel.py": "ec46a0b0e3c40bca7ea2d0c76b086b482fedaf98293b06f7ab719b86b81141e6",
117
+ "src/rl_pt_analyze/pixel_data.py": "74ed81ff034d0782671ff9e48157a3be2d274e7db3735a955705b8e8fcf42b34",
118
+ "src/rl_pt_analyze/pixel_vec.py": "f247a6d9ed1791b82880ecef7505275959835fc3b69bfab2e85dedd096416bc8",
119
+ "src/rl_pt_analyze/r0.py": "e0a13e9c02d935dbab6a0e091fd925f08ce2d00e661b6cf09aff0ee95925249d",
120
+ "third_party/dice/model/diffusion/modules.py": "3b6b2efae0bb67cc30a061a8dd7c6b2d8d6905dac7cecac92b3dbfa5da0067cd",
121
+ "third_party/dice/model/flow_matching/vision_resnet_mlp_flow_matching.py": "2f0d3aa0b6e1624e32a540e66a5803e613481f35bb190926e51560856249981d"
122
+ },
123
+ "spec": {
124
+ "action_dim": 24,
125
+ "architecture": "dice_vision_resnet_flow_unet_v1",
126
+ "backbone_initialization": "random",
127
+ "backbone_norm": "batch_norm",
128
+ "cameras": [
129
+ "agentview",
130
+ "robot0_eye_in_hand",
131
+ "robot1_eye_in_hand"
132
+ ],
133
+ "collision_geoms": false,
134
+ "crop_size": 84,
135
+ "family": "StandardBaseBC",
136
+ "feature_dim": 166,
137
+ "features_per_camera": 64,
138
+ "flow_steps": 10,
139
+ "horizon": 8,
140
+ "image_size": 96,
141
+ "proprio_dim": 38,
142
+ "proprio_keys": [
143
+ "robot0_eef_pos",
144
+ "robot0_eef_quat",
145
+ "robot0_gripper_qpos",
146
+ "robot1_eef_pos",
147
+ "robot1_eef_quat",
148
+ "robot1_gripper_qpos"
149
+ ],
150
+ "task": "dmg_drawer_cleanup",
151
+ "time_embedding_dim": 64
152
+ },
153
+ "split_hash": "0ed7b649ba1ccc81c5e2da4e661361cce3dd65c7311eed303d75a2677b21c9e4",
154
+ "task": "dmg_drawer_cleanup",
155
+ "training": {
156
+ "batch_size": 256,
157
+ "bc_only": true,
158
+ "drop_last": false,
159
+ "early_stopping": false,
160
+ "ema_decay": 0.999,
161
+ "epoch_definition": "shuffled_full_training_split_without_replacement",
162
+ "epochs": 3000,
163
+ "learning_rate": 0.0001,
164
+ "online_steps": 0,
165
+ "optimizer": "adamw",
166
+ "padding_loss_mask": true,
167
+ "stopping": "none_fixed_epochs",
168
+ "weight_decay": 1e-06
169
+ },
170
+ "training_samples": 13533,
171
+ "validation": {
172
+ "batch_seed": 771,
173
+ "batch_size": 256,
174
+ "interval_epochs": 1,
175
+ "loss_seed": 882
176
+ },
177
+ "validation_only": false,
178
+ "validation_samples": 1558,
179
+ "wandb": {
180
+ "mode": "offline",
181
+ "project": "standard_bc_pretrain"
182
+ }
183
+ },
184
+ "schema_version": 2,
185
+ "sha256": "647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d",
186
+ "split_hash": "0ed7b649ba1ccc81c5e2da4e661361cce3dd65c7311eed303d75a2677b21c9e4",
187
+ "weights_used_for_deployment": "complete_model_EMA"
188
+ }
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/checkpoints/epoch_1400.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d
3
+ size 138484087
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/epoch_1400.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42bb62ca863eae57ca06cb337cd83963c77d619949bebbb682fd8d6f30c7b862
3
+ size 6543
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/protocol.json CHANGED
@@ -1,13 +1,11 @@
1
  {
2
- "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/mixed50/seed_1/checkpoints/epoch_1100.pkl",
3
- "checkpoint_sha256": "8546111eaa19cf058caa0f5d8e52c690fa7ff2e623664a4f0975737ebe78ab1b",
4
  "collision_geoms": false,
5
- "epoch": 1100,
6
  "identity_hash": "29fdfa802c2fbd37332749469930ecb2f211b0779cc747f6f99327ef6593bf0d",
7
  "job": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/jobs/01.json",
8
  "panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
9
- "panel_role": "final200",
10
- "plan_sha256": "aab9b2adffe217c850afd08c0c2ce773db4884cac4ffe589f71a81195085b1b0",
11
  "reset_seeds": [
12
  70723122,
13
  66851989,
@@ -210,7 +208,7 @@
210
  10971071,
211
  17007174
212
  ],
213
- "script_sha256": "051ff2a6fdb3b5a11a97b3bc6fbf9aa2f2ef5203b7f68a37a4cc75ee7f0484f4",
214
  "selection": "frozen before this independent panel was evaluated",
215
  "settings": {
216
  "episodes": 200,
 
1
  {
2
+ "checkpoint": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/StandardBaseBC/jax/dmg_drawer_cleanup/mixed50/seed_1/checkpoints/epoch_1400.pkl",
3
+ "checkpoint_sha256": "647a8e3d94a62dbfe436a2574cf786b76e24547b80b5c10134e352e8d99deb1d",
4
  "collision_geoms": false,
5
+ "epoch": 1400,
6
  "identity_hash": "29fdfa802c2fbd37332749469930ecb2f211b0779cc747f6f99327ef6593bf0d",
7
  "job": "/home/nhy/RL-postTrain/repo_visual/.local/training_checkouts/20261025_dmg_visual_bc/runs/experiments/study1/pixel_state/20261025_dmg50_visual_bc/jobs/01.json",
8
  "panel_hash": "8b0a53b46a7c267937197ed4a003ecb33113b45f516bc79c9a9e9dfcf1a023d3",
 
 
9
  "reset_seeds": [
10
  70723122,
11
  66851989,
 
208
  10971071,
209
  17007174
210
  ],
211
+ "script_sha256": "7be8b5440b354a862228f5a3d3855ec0ba7267c78c4e04158fa028e9633e8d32",
212
  "selection": "frozen before this independent panel was evaluated",
213
  "settings": {
214
  "episodes": 200,
visual_no_collision/DexMimicGen/study1/DrawerCleanup/mixed50/seed_1/final200/result.json CHANGED
The diff for this file is too large to render. See raw diff