Caesarrr commited on
Commit
80869c8
·
verified ·
1 Parent(s): fef67c6

Publish corrected issued-action walker2d pi05 profile model

Browse files
Files changed (18) hide show
  1. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/README.md +6 -28
  2. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/REPORT.md +18 -0
  3. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/checkpoints/model.pt +1 -1
  4. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/comparison.json +553 -0
  5. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.full.yaml +17 -16
  6. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.yaml +4 -3
  7. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/dataset_statistics.json +117 -117
  8. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/evaluations/issued_action_iid_20261010/episode_metrics.jsonl +0 -0
  9. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/evaluations/issued_action_iid_20261010/eval_config.yaml +166 -0
  10. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/manifest.json +27 -27
  11. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/normalization_binding.json +6 -0
  12. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/original_training_statistics.json +180 -0
  13. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/provenance.json +0 -34
  14. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/README.md +0 -5
  15. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/provenance.json +0 -96
  16. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/task_contract.json +0 -80
  17. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/training.complete.json +14 -0
  18. latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/validation.json +0 -10
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/README.md CHANGED
@@ -1,31 +1,9 @@
1
- # walker2d / qwenpi_v3
2
 
3
- Training condition: `latency-aware`. Run: `walker2d_pi05_profile_h1_5fps_g128_20260922_m32`.
4
 
5
- Copied from [Standard-Pipeline](https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/Pi05) at `2208875f92b21e3f8ec2363a30a6705dfe56cc04`.
 
6
 
7
- - Checkpoint: `checkpoints/model.pt`
8
- - Selection: only published VLA checkpoint
9
- - Checkpoint SHA256: `248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84`
10
- - Configuration, FPS, actions, normalization and latency semantics retain their source values.
11
- - This import does not run evaluations or change training/evaluation/acceptance status.
12
-
13
- Action horizon is 1. Use `config.yaml` and `dataset_statistics.json` with the matching
14
- prepared backbone. `task_contract.json` is extracted from the original training configuration.
15
- Original machine paths in historical configurations are provenance: bind checkpoint,
16
- backbone, manifest and profile paths to downloaded local files when running inference.
17
-
18
- See [the original model notes](source/README.md) for experiment results and limitations.
19
-
20
- [Original provenance](source/provenance.json) retains dataset references and recorded experiment evidence.
21
-
22
- ## Download
23
-
24
- ```python
25
- from huggingface_hub import snapshot_download
26
- prefix = "latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32"
27
- snapshot_download("latency-sensitive-bench/benchmark-models", revision="<commit>",
28
- allow_patterns=[prefix + "/**"])
29
- ```
30
-
31
- Use the verified import commit listed in the repository's import report.
 
1
+ # walker2d Pi0.5 profile model
2
 
3
+ Corrected source-observation teacher-issued action supervision.
4
 
5
+ Step 5000, seed 42, global batch 128, gradient checkpointing enabled.
6
+ Evaluation: 100 episodes, seeds 42–141, simulated IID, latency seed 0, 5 FPS, H1, cap 1000, parallel 32, simulated capacity 1.
7
 
8
+ See comparison.json, training.complete.json and evaluations/issued_action_iid_20261010 for checkpoint and episode evidence.
9
+ Inference state statistics use the source manifest raw-state extrema; original training statistics are retained.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/REPORT.md ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # walker2d Pi0.5 IID rerun
2
+
3
+ Training and evaluation completed. Publication was not performed.
4
+
5
+ Evaluation: simulated IID, 100 episodes, seeds 42–141, latency seed 0, 5 FPS, H1, cap 1000.
6
+ Parallel environments: 32. Simulated worker capacity remains 1 per environment.
7
+
8
+ | Checkpoint | Mean return | Population SD | Mean length |
9
+ | --- | ---: | ---: | ---: |
10
+ | old | 405.6097 | 88.3381 | 183.43 |
11
+ | new | 3812.6454 | 1016.7950 | 913.76 |
12
+
13
+ Mean paired gain: 3407.0357. Improved seeds: 100/100.
14
+
15
+ Changes include corrected labels, current runtime/loader code, gradient checkpointing and the selected micro batch.
16
+ This comparison does not isolate label correction from these other changes.
17
+
18
+ Checkpoint identities, source versions, profile, configurations and episode evidence are in comparison.json.
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/checkpoints/model.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84
3
  size 10922653853
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4834728df5aa192b59664546e6caeae2941d0a0aee2c49e63275cc8742560358
3
  size 10922653853
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/comparison.json ADDED
@@ -0,0 +1,553 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "task": "walker2d",
3
+ "method": "iid",
4
+ "parallel_envs": 32,
5
+ "seeds": [
6
+ 42,
7
+ 43,
8
+ 44,
9
+ 45,
10
+ 46,
11
+ 47,
12
+ 48,
13
+ 49,
14
+ 50,
15
+ 51,
16
+ 52,
17
+ 53,
18
+ 54,
19
+ 55,
20
+ 56,
21
+ 57,
22
+ 58,
23
+ 59,
24
+ 60,
25
+ 61,
26
+ 62,
27
+ 63,
28
+ 64,
29
+ 65,
30
+ 66,
31
+ 67,
32
+ 68,
33
+ 69,
34
+ 70,
35
+ 71,
36
+ 72,
37
+ 73,
38
+ 74,
39
+ 75,
40
+ 76,
41
+ 77,
42
+ 78,
43
+ 79,
44
+ 80,
45
+ 81,
46
+ 82,
47
+ 83,
48
+ 84,
49
+ 85,
50
+ 86,
51
+ 87,
52
+ 88,
53
+ 89,
54
+ 90,
55
+ 91,
56
+ 92,
57
+ 93,
58
+ 94,
59
+ 95,
60
+ 96,
61
+ 97,
62
+ 98,
63
+ 99,
64
+ 100,
65
+ 101,
66
+ 102,
67
+ 103,
68
+ 104,
69
+ 105,
70
+ 106,
71
+ 107,
72
+ 108,
73
+ 109,
74
+ 110,
75
+ 111,
76
+ 112,
77
+ 113,
78
+ 114,
79
+ 115,
80
+ 116,
81
+ 117,
82
+ 118,
83
+ 119,
84
+ 120,
85
+ 121,
86
+ 122,
87
+ 123,
88
+ 124,
89
+ 125,
90
+ 126,
91
+ 127,
92
+ 128,
93
+ 129,
94
+ 130,
95
+ 131,
96
+ 132,
97
+ 133,
98
+ 134,
99
+ 135,
100
+ 136,
101
+ 137,
102
+ 138,
103
+ 139,
104
+ 140,
105
+ 141
106
+ ],
107
+ "latency_seed": 0,
108
+ "eval_max_steps": 1000,
109
+ "old": {
110
+ "mean_return": 405.60973120507776,
111
+ "population_std_return": 88.33809684182947,
112
+ "mean_length": 183.43,
113
+ "checkpoint": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/old_bundle/checkpoints/model.pt",
114
+ "config": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/eval_old_parallel32.yaml",
115
+ "episodes": 100,
116
+ "primary_evidence": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/evaluation/old_parallel32/episode_metrics.jsonl",
117
+ "returns": [
118
+ 349.5234759373331,
119
+ 336.40765715341433,
120
+ 378.1645037152372,
121
+ 422.63699905416433,
122
+ 517.9973329060393,
123
+ 373.26849124553956,
124
+ 533.695807400339,
125
+ 394.2994652090446,
126
+ 374.8355493063823,
127
+ 424.34860940575214,
128
+ 494.0881235756625,
129
+ 336.5511857735024,
130
+ 549.3262227736218,
131
+ 440.30025722124395,
132
+ 398.28413102053287,
133
+ 438.4898870437722,
134
+ 410.0560286595904,
135
+ 548.298113573577,
136
+ 426.1093063157936,
137
+ 447.32086245031894,
138
+ 426.4455274361643,
139
+ 414.2808240820888,
140
+ 451.4343247804819,
141
+ 731.5208201062378,
142
+ 443.9623219955346,
143
+ 366.3996024361343,
144
+ 354.799419275636,
145
+ 406.83017580887594,
146
+ 363.5807974398917,
147
+ 374.6919053594528,
148
+ 396.40696155600455,
149
+ 403.2192852793021,
150
+ 440.03761468708905,
151
+ 387.40907797713356,
152
+ 426.2472898735999,
153
+ 547.4367614934459,
154
+ 603.6783892233947,
155
+ 365.4729650186947,
156
+ 375.017648916159,
157
+ 537.7504862556481,
158
+ 344.09872673708406,
159
+ 443.59261263988293,
160
+ 451.8709659727219,
161
+ 381.71539080169543,
162
+ 355.34996872837,
163
+ 402.192006282284,
164
+ 428.4075565079848,
165
+ 557.7823797651452,
166
+ 393.6722156021898,
167
+ -11.366919102524454,
168
+ 391.9755724932713,
169
+ 484.0013314315968,
170
+ 335.60015194960476,
171
+ 329.5117996655172,
172
+ 387.17141228047484,
173
+ 359.5402873229976,
174
+ 283.236441958444,
175
+ 331.8034503869983,
176
+ 485.1829735364345,
177
+ 410.5884452954014,
178
+ 398.57940954467574,
179
+ 289.7134880658964,
180
+ 319.02499215370796,
181
+ 487.4007641151117,
182
+ 383.52841554938794,
183
+ 367.775931930128,
184
+ 361.57128361690576,
185
+ 333.8801777616892,
186
+ 405.0017230831131,
187
+ 139.71282678377153,
188
+ 400.79802566787214,
189
+ 415.5373948838387,
190
+ 488.6213195321705,
191
+ 401.0653400152562,
192
+ 401.22677006684296,
193
+ 380.27919914135856,
194
+ 448.24005140909895,
195
+ 402.33414466142955,
196
+ 403.4279921879888,
197
+ 379.16211446554826,
198
+ 381.65094736577043,
199
+ 360.7231422614541,
200
+ 468.01649576343857,
201
+ 378.6331413744438,
202
+ 354.5586516669707,
203
+ 383.82181120013627,
204
+ 435.24854171532354,
205
+ 530.524555456558,
206
+ 332.70773640700725,
207
+ 406.4891560532332,
208
+ 613.5874432647438,
209
+ 390.30310286611393,
210
+ 446.55380253634075,
211
+ 285.0095646308312,
212
+ 419.66018507893085,
213
+ 380.0662194672008,
214
+ 368.49812233233246,
215
+ 276.1424860728806,
216
+ 398.4140572855555,
217
+ 386.931614075315
218
+ ]
219
+ },
220
+ "new": {
221
+ "mean_return": 3812.6454174263054,
222
+ "population_std_return": 1016.794969772278,
223
+ "mean_length": 913.76,
224
+ "checkpoint": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/checkpoints/model.pt",
225
+ "config": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/eval_new_parallel32.yaml",
226
+ "episodes": 100,
227
+ "primary_evidence": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/evaluation/new_parallel32/episode_metrics.jsonl",
228
+ "returns": [
229
+ 4569.236923310628,
230
+ 4416.164969963065,
231
+ 4544.930478042572,
232
+ 1151.5068522031675,
233
+ 3836.1139218934163,
234
+ 4343.70793956925,
235
+ 4329.026191906756,
236
+ 1315.6254878010327,
237
+ 4472.293033300189,
238
+ 4470.1136693127855,
239
+ 3776.2399406329014,
240
+ 1040.485673323373,
241
+ 4331.40648234694,
242
+ 4414.753067012341,
243
+ 4410.864050363345,
244
+ 4442.895190750697,
245
+ 4387.163922874358,
246
+ 3957.736737691899,
247
+ 4453.843815891176,
248
+ 1141.8949190545252,
249
+ 4329.885297261079,
250
+ 4162.299126852293,
251
+ 1971.4218382658128,
252
+ 3718.6847991694544,
253
+ 882.7887685649018,
254
+ 4402.295081136063,
255
+ 4309.420930292155,
256
+ 4546.888400610404,
257
+ 4268.126263435571,
258
+ 4577.9866625969635,
259
+ 4488.907081955325,
260
+ 4459.297046816433,
261
+ 4267.847082198069,
262
+ 4374.213645300379,
263
+ 4301.136232707875,
264
+ 3417.9887175974623,
265
+ 4044.73823644874,
266
+ 4431.394224689892,
267
+ 3951.265878364092,
268
+ 3655.2406046150977,
269
+ 4009.8119500575517,
270
+ 4379.700491637848,
271
+ 4302.773029713501,
272
+ 4342.243547233506,
273
+ 4396.533395988757,
274
+ 3847.487296288443,
275
+ 4279.877276975001,
276
+ 1204.49738486329,
277
+ 4453.521923980218,
278
+ 4045.3431099382624,
279
+ 1946.347707106949,
280
+ 4215.868783085659,
281
+ 1745.575152374404,
282
+ 2498.1575274034517,
283
+ 4316.773450504653,
284
+ 4398.539350595546,
285
+ 4379.578572389666,
286
+ 4319.321677682342,
287
+ 4476.2449162534895,
288
+ 4291.672201693065,
289
+ 4020.470779915537,
290
+ 3941.977069554052,
291
+ 3987.1175274217135,
292
+ 4004.153473815914,
293
+ 4070.0945914278514,
294
+ 4423.3193472724,
295
+ 3669.602085862326,
296
+ 4452.112258288449,
297
+ 4143.752304744489,
298
+ 4323.333354163676,
299
+ 4345.942379977469,
300
+ 4478.588549518908,
301
+ 4548.868052433143,
302
+ 4202.103427001851,
303
+ 1721.0446979797837,
304
+ 3956.204649600708,
305
+ 4185.779955720576,
306
+ 4070.682565189042,
307
+ 4186.093279396205,
308
+ 4414.998249801816,
309
+ 4279.482269961291,
310
+ 1746.6772425292113,
311
+ 4121.424358099625,
312
+ 4410.70504114393,
313
+ 3924.1791746996355,
314
+ 1829.073828471807,
315
+ 4201.457753671447,
316
+ 3915.279403115198,
317
+ 1193.8727735614484,
318
+ 4164.986164674403,
319
+ 4236.37797692242,
320
+ 3711.7560253163015,
321
+ 4585.098675848439,
322
+ 1874.8619685913798,
323
+ 4417.989284521721,
324
+ 4377.188463991608,
325
+ 4509.388332042925,
326
+ 1959.9181946186059,
327
+ 4076.9988721082223,
328
+ 4363.8894077669265
329
+ ]
330
+ },
331
+ "paired_gain": [
332
+ 4219.713447373295,
333
+ 4079.757312809651,
334
+ 4166.765974327334,
335
+ 728.8698531490031,
336
+ 3318.116588987377,
337
+ 3970.43944832371,
338
+ 3795.330384506417,
339
+ 921.326022591988,
340
+ 4097.457483993807,
341
+ 4045.7650599070334,
342
+ 3282.151817057239,
343
+ 703.9344875498706,
344
+ 3782.080259573318,
345
+ 3974.4528097910966,
346
+ 4012.579919342812,
347
+ 4004.4053037069248,
348
+ 3977.1078942147674,
349
+ 3409.438624118322,
350
+ 4027.734509575382,
351
+ 694.5740566042062,
352
+ 3903.439769824915,
353
+ 3748.018302770204,
354
+ 1519.987513485331,
355
+ 2987.1639790632166,
356
+ 438.82644656936725,
357
+ 4035.8954786999293,
358
+ 3954.6215110165194,
359
+ 4140.058224801528,
360
+ 3904.545465995679,
361
+ 4203.294757237511,
362
+ 4092.5001203993206,
363
+ 4056.0777615371308,
364
+ 3827.80946751098,
365
+ 3986.804567323245,
366
+ 3874.888942834275,
367
+ 2870.5519561040164,
368
+ 3441.059847225345,
369
+ 4065.9212596711977,
370
+ 3576.2482294479328,
371
+ 3117.4901183594498,
372
+ 3665.7132233204675,
373
+ 3936.107878997965,
374
+ 3850.9020637407793,
375
+ 3960.528156431811,
376
+ 4041.1834272603865,
377
+ 3445.2952900061587,
378
+ 3851.4697204670165,
379
+ 646.715005098145,
380
+ 4059.8497083780285,
381
+ 4056.710029040787,
382
+ 1554.3721346136776,
383
+ 3731.8674516540623,
384
+ 1409.9750004247994,
385
+ 2168.6457277379345,
386
+ 3929.602038224178,
387
+ 4038.9990632725485,
388
+ 4096.342130431222,
389
+ 3987.5182272953434,
390
+ 3991.061942717055,
391
+ 3881.083756397664,
392
+ 3621.8913703708613,
393
+ 3652.2635814881555,
394
+ 3668.0925352680056,
395
+ 3516.7527097008024,
396
+ 3686.5661758784636,
397
+ 4055.5434153422716,
398
+ 3308.03080224542,
399
+ 4118.23208052676,
400
+ 3738.750581661376,
401
+ 4183.620527379904,
402
+ 3945.1443543095966,
403
+ 4063.051154635069,
404
+ 4060.2467329009723,
405
+ 3801.038086986595,
406
+ 1319.8179279129408,
407
+ 3575.925450459349,
408
+ 3737.539904311477,
409
+ 3668.3484205276127,
410
+ 3782.665287208216,
411
+ 4035.836135336268,
412
+ 3897.8313225955208,
413
+ 1385.9541002677572,
414
+ 3653.4078623361866,
415
+ 4032.0718997694867,
416
+ 3569.620523032665,
417
+ 1445.2520172716709,
418
+ 3766.2092119561235,
419
+ 3384.75484765864,
420
+ 861.1650371544412,
421
+ 3758.4970086211697,
422
+ 3622.7905336576764,
423
+ 3321.4529224501875,
424
+ 4138.5448733120975,
425
+ 1589.8524039605486,
426
+ 3998.3290994427903,
427
+ 3997.122244524407,
428
+ 4140.890209710593,
429
+ 1683.7757085457251,
430
+ 3678.584814822667,
431
+ 3976.9577936916116
432
+ ],
433
+ "mean_paired_gain": 3407.035686221227,
434
+ "wins": 100,
435
+ "source_identity": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/source_identity.json",
436
+ "asset_revisions": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/asset_revisions.json",
437
+ "profile": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/original_profile/profiling/profile/profile.json",
438
+ "label_audit": {
439
+ "decisions": 99169,
440
+ "issued_differs_from_applied": 98023,
441
+ "clipped_issued_differs_from_applied": 98023,
442
+ "issued_requires_clip": 0,
443
+ "max_abs_issued_applied_difference": 2.0,
444
+ "examples": [
445
+ {
446
+ "episode_id": 0,
447
+ "decision_step": 0,
448
+ "issued": [
449
+ 1.0,
450
+ 1.0,
451
+ 0.5466500520706177,
452
+ 1.0,
453
+ 1.0,
454
+ 1.0
455
+ ],
456
+ "applied": [
457
+ 0.0,
458
+ 0.0,
459
+ 0.0,
460
+ 0.0,
461
+ 0.0,
462
+ 0.0
463
+ ]
464
+ },
465
+ {
466
+ "episode_id": 0,
467
+ "decision_step": 1,
468
+ "issued": [
469
+ 1.0,
470
+ 1.0,
471
+ 0.576922595500946,
472
+ 1.0,
473
+ 1.0,
474
+ 1.0
475
+ ],
476
+ "applied": [
477
+ 1.0,
478
+ 1.0,
479
+ 0.5466500520706177,
480
+ 1.0,
481
+ 1.0,
482
+ 1.0
483
+ ]
484
+ },
485
+ {
486
+ "episode_id": 0,
487
+ "decision_step": 2,
488
+ "issued": [
489
+ 1.0,
490
+ 1.0,
491
+ -1.0,
492
+ 1.0,
493
+ 1.0,
494
+ 1.0
495
+ ],
496
+ "applied": [
497
+ 1.0,
498
+ 1.0,
499
+ 0.5466500520706177,
500
+ 1.0,
501
+ 1.0,
502
+ 1.0
503
+ ]
504
+ },
505
+ {
506
+ "episode_id": 0,
507
+ "decision_step": 3,
508
+ "issued": [
509
+ 1.0,
510
+ 1.0,
511
+ -1.0,
512
+ 1.0,
513
+ 1.0,
514
+ 1.0
515
+ ],
516
+ "applied": [
517
+ 1.0,
518
+ 1.0,
519
+ 0.576922595500946,
520
+ 1.0,
521
+ 1.0,
522
+ 1.0
523
+ ]
524
+ },
525
+ {
526
+ "episode_id": 0,
527
+ "decision_step": 5,
528
+ "issued": [
529
+ 1.0,
530
+ 1.0,
531
+ -1.0,
532
+ 0.7121425271034241,
533
+ 1.0,
534
+ 1.0
535
+ ],
536
+ "applied": [
537
+ 1.0,
538
+ 1.0,
539
+ -1.0,
540
+ 1.0,
541
+ 1.0,
542
+ 1.0
543
+ ]
544
+ }
545
+ ]
546
+ },
547
+ "training_changes": [
548
+ "issued teacher action labels",
549
+ "current code including loader contract fixes",
550
+ "gradient checkpointing",
551
+ "micro batch selected by OOM"
552
+ ]
553
+ }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.full.yaml CHANGED
@@ -1,17 +1,16 @@
1
  framework:
2
  name: QwenPI_v3
3
  qwenvl:
4
- base_vlm: ${oc.env:PI05_BACKBONE_DIR}
5
  attn_implementation: flash_attention_2
6
  vl_hidden_dim: 2560
7
  num_vl_layers: 36
8
- enable_gradient_checkpointing: false
9
  action_model:
10
  action_model_type: LayerwiseFM
11
  action_dim: 6
12
  state_dim: 17
13
  state_encoding: continuous_projector
14
- task_objective: null
15
  action_horizon: 1
16
  repeated_diffusion_steps: 2
17
  num_inference_timesteps: 4
@@ -35,6 +34,7 @@ framework:
35
  cross_attention_dim: 1024
36
  output_dim: 1024
37
  num_attention_heads: 16
 
38
  action_env_dim: 6
39
  future_action_window_size: 0
40
  past_action_window_size: 0
@@ -45,7 +45,7 @@ datasets:
45
  data_root_dir: /
46
  data_mix: train
47
  eval_data_mix: validation
48
- custom_mixtures_path: ${PI05_RUN_DIR}/profile_latency/vla/mixture.json
49
  action_type: continuous
50
  sequential_step_sampling: true
51
  eval_sequential_step_sampling: null
@@ -76,7 +76,7 @@ datasets:
76
  eval_at_phase_end: false
77
  save_at_phase_end: false
78
  computed_plan: null
79
- per_device_batch_size: 32
80
  load_all_data_for_training: true
81
  num_obs_frames: 1
82
  image_mode: single
@@ -139,7 +139,7 @@ datasets:
139
  - 0.0
140
  obs_fps: 5.0
141
  registration_imports:
142
- - latency_bench.envs.gymnasium_walker2d
143
  state_space:
144
  labels:
145
  - torso_height
@@ -221,7 +221,7 @@ trainer:
221
  enabled: false
222
  log_interval: 10
223
  gradient_clipping: 1.0
224
- gradient_accumulation_steps: 2
225
  distributed_backend: deepspeed
226
  is_resume: false
227
  pretrained_checkpoint: null
@@ -236,20 +236,21 @@ trainer:
236
  weight_decay: 1.0e-08
237
  fused: true
238
  save_format: pt
239
- run_id: walker2d_pi05_profile_h1_5fps_g128_20260922_m32
240
- run_root_dir: ${PI05_RUN_DIR}/profile_latency/vla/training
241
  seed: 42
242
  is_debug: false
243
  version_id: '0.21'
244
  wandb_project: latency-sensitive-bench
245
- wandb_entity: dongqianyu99-zhejiang-university
246
- wandb_name: walker2d_pi05_profile_h1_5fps_g128_20260922_m32
247
- wandb_group: pi05-seven-env
248
  wandb_tags:
249
  - walker2d
250
- - profile_latency
251
  - Pi05
252
- - h1
 
 
253
  training_latency_condition: profile_latency
254
  backbone:
255
  repo_id: Qwen/Qwen3-VL-4B-Instruct
@@ -262,5 +263,5 @@ checkpoint:
262
  local:
263
  keep_last_n: 1
264
  global_batch: 128
265
- config_yaml: ${PI05_RUN_DIR}/profile_latency/train-m32.yaml
266
- output_dir: ${PI05_RUN_DIR}/profile_latency/vla/training/walker2d_pi05_profile_h1_5fps_g128_20260922_m32
 
1
  framework:
2
  name: QwenPI_v3
3
  qwenvl:
4
+ base_vlm: /home/ubuntu/.cache/huggingface/hub/models--Qwen--Qwen3-VL-4B-Instruct/snapshots/ebb281ec70b05090aa6165b016eac8ec08e71b17
5
  attn_implementation: flash_attention_2
6
  vl_hidden_dim: 2560
7
  num_vl_layers: 36
8
+ enable_gradient_checkpointing: true
9
  action_model:
10
  action_model_type: LayerwiseFM
11
  action_dim: 6
12
  state_dim: 17
13
  state_encoding: continuous_projector
 
14
  action_horizon: 1
15
  repeated_diffusion_steps: 2
16
  num_inference_timesteps: 4
 
34
  cross_attention_dim: 1024
35
  output_dim: 1024
36
  num_attention_heads: 16
37
+ task_objective: null
38
  action_env_dim: 6
39
  future_action_window_size: 0
40
  past_action_window_size: 0
 
45
  data_root_dir: /
46
  data_mix: train
47
  eval_data_mix: validation
48
+ custom_mixtures_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/mixture.json
49
  action_type: continuous
50
  sequential_step_sampling: true
51
  eval_sequential_step_sampling: null
 
76
  eval_at_phase_end: false
77
  save_at_phase_end: false
78
  computed_plan: null
79
+ per_device_batch_size: 64
80
  load_all_data_for_training: true
81
  num_obs_frames: 1
82
  image_mode: single
 
139
  - 0.0
140
  obs_fps: 5.0
141
  registration_imports:
142
+ - latency_bench.envs.tasks.walker2d
143
  state_space:
144
  labels:
145
  - torso_height
 
221
  enabled: false
222
  log_interval: 10
223
  gradient_clipping: 1.0
224
+ gradient_accumulation_steps: 1
225
  distributed_backend: deepspeed
226
  is_resume: false
227
  pretrained_checkpoint: null
 
236
  weight_decay: 1.0e-08
237
  fused: true
238
  save_format: pt
239
+ run_id: walker2d_pi05_issued_action_iid_s42_20261010
240
+ run_root_dir: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/training
241
  seed: 42
242
  is_debug: false
243
  version_id: '0.21'
244
  wandb_project: latency-sensitive-bench
245
+ wandb_entity: saberrr-zju
246
+ wandb_name: walker2d_pi05_issued_action_iid_s42_20261010
247
+ wandb_group: pi05-issued-action-iid-20261010
248
  wandb_tags:
249
  - walker2d
 
250
  - Pi05
251
+ - H1
252
+ - IID
253
+ - teacher-issued-action
254
  training_latency_condition: profile_latency
255
  backbone:
256
  repo_id: Qwen/Qwen3-VL-4B-Instruct
 
263
  local:
264
  keep_last_n: 1
265
  global_batch: 128
266
+ config_yaml: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/train.yaml
267
+ output_dir: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/training/walker2d_pi05_issued_action_iid_s42_20261010
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/config.yaml CHANGED
@@ -1,11 +1,11 @@
1
  framework:
2
  name: QwenPI_v3
3
  qwenvl:
4
- base_vlm: ${oc.env:PI05_BACKBONE_DIR}
5
  attn_implementation: flash_attention_2
6
  vl_hidden_dim: 2560
7
  num_vl_layers: 36
8
- enable_gradient_checkpointing: false
9
  action_model:
10
  action_model_type: LayerwiseFM
11
  action_dim: 6
@@ -100,7 +100,7 @@ datasets:
100
  - 0.0
101
  obs_fps: 5.0
102
  registration_imports:
103
- - latency_bench.envs.gymnasium_walker2d
104
  state_space:
105
  labels:
106
  - torso_height
@@ -121,6 +121,7 @@ datasets:
121
  - left_leg_angular_velocity
122
  - left_foot_angular_velocity
123
  task_name: walker2d
 
124
  trainer:
125
  pretrained_checkpoint: null
126
  profile_timing:
 
1
  framework:
2
  name: QwenPI_v3
3
  qwenvl:
4
+ base_vlm: Qwen/Qwen3-VL-4B-Instruct
5
  attn_implementation: flash_attention_2
6
  vl_hidden_dim: 2560
7
  num_vl_layers: 36
8
+ enable_gradient_checkpointing: true
9
  action_model:
10
  action_model_type: LayerwiseFM
11
  action_dim: 6
 
100
  - 0.0
101
  obs_fps: 5.0
102
  registration_imports:
103
+ - latency_bench.envs.tasks.walker2d
104
  state_space:
105
  labels:
106
  - torso_height
 
121
  - left_leg_angular_velocity
122
  - left_foot_angular_velocity
123
  task_name: walker2d
124
+ robot_type: rl_games_gymnasium
125
  trainer:
126
  pretrained_checkpoint: null
127
  profile_timing:
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/dataset_statistics.json CHANGED
@@ -2,20 +2,20 @@
2
  "new_embodiment": {
3
  "action": {
4
  "mean": [
5
- 0.4979535639286041,
6
- 0.46174800395965576,
7
- -0.29137447476387024,
8
- 0.9985677003860474,
9
- 0.7979836463928223,
10
- 0.38867390155792236
11
  ],
12
  "std": [
13
- 0.581741213798523,
14
- 0.4915325939655304,
15
- 0.7721808552742004,
16
- 0.035041093826293945,
17
- 0.5022090673446655,
18
- 0.6507065296173096
19
  ],
20
  "max": [
21
  1.0,
@@ -29,7 +29,7 @@
29
  -1.0,
30
  -1.0,
31
  -1.0,
32
- 0.0,
33
  -1.0,
34
  -1.0
35
  ],
@@ -60,121 +60,121 @@
60
  },
61
  "state": {
62
  "mean": [
63
- 0.23815855383872986,
64
- -0.1610857993364334,
65
- 0.5441378951072693,
66
- 0.672118604183197,
67
- -0.11596863716840744,
68
- 0.6917569041252136,
69
- 0.637482762336731,
70
- 0.4345299303531647,
71
- -0.18043601512908936,
72
- 0.1474768966436386,
73
- -0.007371156010776758,
74
- -0.01192600466310978,
75
- -0.0426424965262413,
76
- 0.011722072958946228,
77
- -0.002667834050953388,
78
- -0.02254018746316433,
79
- -0.048787184059619904
80
  ],
81
  "std": [
82
- 0.37967589497566223,
83
- 0.21708180010318756,
84
- 0.318278968334198,
85
- 0.24248674511909485,
86
- 0.5238001942634583,
87
- 0.1172376424074173,
88
- 0.2069370299577713,
89
- 0.458295077085495,
90
- 0.2840738296508789,
91
- 0.3721705675125122,
92
- 0.3089592158794403,
93
- 0.28156015276908875,
94
- 0.4521273672580719,
95
- 0.5419289469718933,
96
- 0.1749182939529419,
97
- 0.4066046178340912,
98
- 0.5292556285858154
99
  ],
100
  "max": [
101
- 1.0,
102
- 1.0,
103
- 1.0,
104
- 1.0,
105
- 1.0,
106
- 1.0,
107
- 1.0,
108
- 1.0,
109
- 1.0,
110
- 1.0,
111
- 1.0,
112
- 1.0,
113
- 1.0,
114
- 1.0,
115
- 1.0,
116
- 1.0,
117
- 1.0
118
  ],
119
  "min": [
120
- -1.0,
121
- -1.0,
122
- -1.0,
123
- -1.0,
124
- -1.0,
125
- -1.0,
126
- -1.0,
127
- -1.0,
128
- -1.0,
129
- -1.0,
130
- -1.0,
131
- -1.0,
132
- -1.0,
133
- -1.0,
134
- -1.0,
135
- -1.0,
136
- -1.0
137
  ],
138
  "q01": [
139
- -0.40312816202640533,
140
- -0.7265578722953796,
141
- -0.3082904410362243,
142
- -0.060336984992027276,
143
- -0.7253159177303314,
144
- 0.18186532378196718,
145
- -0.03887396693229675,
146
- -0.7812737548351287,
147
- -0.8779666405916214,
148
- -0.5537199997901916,
149
- -0.578927686214447,
150
- -0.5143137556314469,
151
- -1.0,
152
- -1.0,
153
- -0.3972254878282547,
154
- -1.0,
155
- -1.0
156
  ],
157
  "q99": [
158
- 0.8957907927036282,
159
- 0.27771593093872016,
160
- 0.946374636888504,
161
- 0.9333751380443571,
162
- 0.8444677841663356,
163
- 0.859323478937149,
164
- 0.8583487355709075,
165
- 0.9217252075672145,
166
- 0.6287772941589355,
167
- 0.9185322749614713,
168
- 1.0,
169
- 1.0,
170
- 1.0,
171
- 1.0,
172
- 1.0,
173
- 1.0,
174
- 1.0
175
  ]
176
  },
177
- "num_transitions": 88708,
178
  "num_trajectories": 90
179
  }
180
- }
 
2
  "new_embodiment": {
3
  "action": {
4
  "mean": [
5
+ 0.4981546700000763,
6
+ 0.4653775095939636,
7
+ -0.2947416603565216,
8
+ 0.9996105432510376,
9
+ 0.8025897741317749,
10
+ 0.38950979709625244
11
  ],
12
  "std": [
13
+ 0.5788458585739136,
14
+ 0.48928776383399963,
15
+ 0.7716024518013,
16
+ 0.01672535389661623,
17
+ 0.4969850182533264,
18
+ 0.6478453278541565
19
  ],
20
  "max": [
21
  1.0,
 
29
  -1.0,
30
  -1.0,
31
  -1.0,
32
+ -1.0,
33
  -1.0,
34
  -1.0
35
  ],
 
60
  },
61
  "state": {
62
  "mean": [
63
+ 1.3917548859970976,
64
+ 0.19466248989659896,
65
+ -0.1781303635711442,
66
+ -0.18586760076124165,
67
+ -0.2629807176695049,
68
+ 0.013511480994872649,
69
+ -0.08032302205876363,
70
+ 0.506051526284315,
71
+ 3.6700938667588288,
72
+ 0.023221609554759937,
73
+ -0.07593836635351181,
74
+ -0.12343162670731544,
75
+ -0.43039456009864807,
76
+ 0.10517130605876446,
77
+ -0.028043393976986408,
78
+ -0.2231612242758274,
79
+ -0.4842042177915573
80
  ],
81
  "std": [
82
+ 0.18056249015691783,
83
+ 0.15079242568025464,
84
+ 0.23092414607450706,
85
+ 0.37573537800816403,
86
+ 0.7524627893619709,
87
+ 0.059269492051136984,
88
+ 0.2615950920515493,
89
+ 0.6095452098155665,
90
+ 1.4138589207407506,
91
+ 1.4833367206713035,
92
+ 3.0759069323539734,
93
+ 2.79937207698822,
94
+ 4.478988349437714,
95
+ 5.409497022628784,
96
+ 1.7595402896404266,
97
+ 4.052764475345612,
98
+ 5.279714465141296
99
  ],
100
  "max": [
101
+ 1.7504061460494995,
102
+ 0.9881704449653625,
103
+ 0.16799136996269226,
104
+ 0.33612364530563354,
105
+ 1.3760323524475098,
106
+ 0.14901067316532135,
107
+ 0.335942804813385,
108
+ 1.3045134544372559,
109
+ 9.560564041137695,
110
+ 3.40214204788208,
111
+ 10.0,
112
+ 10.0,
113
+ 10.0,
114
+ 10.0,
115
+ 10.0,
116
+ 10.0,
117
+ 10.0
118
  ],
119
  "min": [
120
+ 0.8091598153114319,
121
+ -0.958375871181488,
122
+ -1.471215009689331,
123
+ -2.750671863555908,
124
+ -1.5357930660247803,
125
+ -1.3458030223846436,
126
+ -1.5723819732666016,
127
+ -1.5042258501052856,
128
+ -0.7429636120796204,
129
+ -5.120609760284424,
130
+ -10.0,
131
+ -10.0,
132
+ -10.0,
133
+ -10.0,
134
+ -10.0,
135
+ -10.0,
136
+ -10.0
137
  ],
138
  "q01": [
139
+ 1.0862045369709916,
140
+ -0.19665443707486105,
141
+ -0.8015672890735462,
142
+ -1.3186654967195892,
143
+ -1.1315221272412699,
144
+ -0.2412459852531751,
145
+ -0.9344754521468955,
146
+ -1.1165430815337853,
147
+ 0.18895018455511892,
148
+ -2.7601140095799543,
149
+ -5.72732138633728,
150
+ -5.157614970207214,
151
+ -10.0,
152
+ -10.0,
153
+ -3.9685154438018797,
154
+ -10.0,
155
+ -10.0
156
  ],
157
  "q99": [
158
+ 1.699588914840302,
159
+ 0.49903253564570815,
160
+ 0.11458354220421241,
161
+ 0.21274757932750177,
162
+ 1.1216048321801582,
163
+ 0.0975775162835184,
164
+ 0.1993145519733367,
165
+ 1.1535549134602952,
166
+ 7.675136129807546,
167
+ 3.0821373759979585,
168
+ 10.0,
169
+ 10.0,
170
+ 10.0,
171
+ 10.0,
172
+ 10.0,
173
+ 10.0,
174
+ 10.0
175
  ]
176
  },
177
+ "num_transitions": 89169,
178
  "num_trajectories": 90
179
  }
180
+ }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/evaluations/issued_action_iid_20261010/episode_metrics.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/evaluations/issued_action_iid_20261010/eval_config.yaml ADDED
@@ -0,0 +1,166 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment:
2
+ name: walker2d_pi05_new_iid_100ep
3
+ seed: 42
4
+ backend:
5
+ type: sample_factory
6
+ algo: APPO
7
+ device: cuda
8
+ train_dir: /mnt/checkpoints/latency-sensitive-bench/small_models/walker2d
9
+ restart_behavior: overwrite
10
+ run_mode: eval
11
+ executor:
12
+ mode: simulated
13
+ inference_devices:
14
+ - cuda:0
15
+ - cuda:1
16
+ inference_batch_size: 8
17
+ simulated_inference_pool: true
18
+ simulated_worker_capacity: 1
19
+ env:
20
+ name: gymnasium
21
+ task_name: walker2d
22
+ env_id: LatencyBench/Walker2dContinuous-v0
23
+ registration_imports:
24
+ - latency_bench.envs.tasks.walker2d
25
+ make_kwargs:
26
+ base_env_id: Walker2d-v4
27
+ base_make_kwargs:
28
+ ctrl_cost_weight: 0.001
29
+ exclude_current_positions_from_observation: true
30
+ forward_reward_weight: 1.0
31
+ healthy_reward: 1.0
32
+ reset_noise_scale: 0.005
33
+ terminate_when_unhealthy: true
34
+ render_mode: rgb_array
35
+ env_fps: 5
36
+ obs_fps: 5
37
+ frame_stack: 1
38
+ action_space:
39
+ dtype: float32
40
+ high:
41
+ - 1.0
42
+ - 1.0
43
+ - 1.0
44
+ - 1.0
45
+ - 1.0
46
+ - 1.0
47
+ labels:
48
+ - right_thigh_torque
49
+ - right_leg_torque
50
+ - right_foot_torque
51
+ - left_thigh_torque
52
+ - left_leg_torque
53
+ - left_foot_torque
54
+ low:
55
+ - -1.0
56
+ - -1.0
57
+ - -1.0
58
+ - -1.0
59
+ - -1.0
60
+ - -1.0
61
+ type: box
62
+ noop_action:
63
+ - 0.0
64
+ - 0.0
65
+ - 0.0
66
+ - 0.0
67
+ - 0.0
68
+ - 0.0
69
+ base_prompt: Move the Walker2d robot forward while keeping its torso upright. Predict
70
+ six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot,
71
+ left thigh, left leg, and left foot.
72
+ state_space:
73
+ labels:
74
+ - torso_height
75
+ - torso_angle
76
+ - right_thigh_angle
77
+ - right_leg_angle
78
+ - right_foot_angle
79
+ - left_thigh_angle
80
+ - left_leg_angle
81
+ - left_foot_angle
82
+ - torso_x_velocity
83
+ - torso_z_velocity
84
+ - torso_angular_velocity
85
+ - right_thigh_angular_velocity
86
+ - right_leg_angular_velocity
87
+ - right_foot_angular_velocity
88
+ - left_thigh_angular_velocity
89
+ - left_leg_angular_velocity
90
+ - left_foot_angular_velocity
91
+ obs_resize:
92
+ - 224
93
+ - 224
94
+ latency:
95
+ method: iid
96
+ sync_cuda: false
97
+ profile_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/original_profile/profiling/profile/profile.json
98
+ profile_worker_slot: 0
99
+ seed: 0
100
+ scheduler:
101
+ hold_policy: hold
102
+ ordering_policy: issue_order_fifo
103
+ policy:
104
+ type: starvla
105
+ checkpoint_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/checkpoints/model.pt
106
+ model_config_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/config.yaml
107
+ backbone_path: /home/ubuntu/.cache/huggingface/hub/models--Qwen--Qwen3-VL-4B-Instruct/snapshots/ebb281ec70b05090aa6165b016eac8ec08e71b17
108
+ device: cuda:0
109
+ task_manifest_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/manifest.json
110
+ latency_prompt_map_path: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/latency_prompt_map.json
111
+ latency_prompt_key: 1
112
+ training:
113
+ train_for_env_steps: 10000000
114
+ num_workers: 8
115
+ num_envs_per_worker: 8
116
+ worker_num_splits: 2
117
+ num_policies: 1
118
+ batch_size: 1024
119
+ rollout: 64
120
+ recurrence: 1
121
+ num_epochs: 2
122
+ num_batches_per_epoch: 4
123
+ max_policy_lag: 300
124
+ learning_rate: 0.00295
125
+ gamma: 0.99
126
+ gae_lambda: 0.95
127
+ ppo_clip_ratio: 0.2
128
+ ppo_clip_value: 0.2
129
+ value_loss_coeff: 1.3
130
+ max_grad_norm: 3.5
131
+ exploration_loss_coeff: 0.0
132
+ async_rl: false
133
+ batched_sampling: false
134
+ use_rnn: false
135
+ encoder_mlp_layers:
136
+ - 64
137
+ - 64
138
+ normalize_input: true
139
+ normalize_returns: true
140
+ save_every_sec: 600
141
+ keep_checkpoints: 5
142
+ lr_schedule: linear_decay
143
+ kl_loss_coeff: 0.1
144
+ serial_mode: false
145
+ nonlinearity: tanh
146
+ adaptive_stddev: false
147
+ policy_initialization: torch_default
148
+ initial_stddev: 1.0
149
+ shuffle_minibatches: false
150
+ value_bootstrap: true
151
+ evaluation:
152
+ eval_episodes: 100
153
+ eval_parallel_envs: 32
154
+ eval_max_steps: 1000
155
+ eval_deterministic: true
156
+ eval_latency_values: null
157
+ episode_seed_offset: 0
158
+ logging:
159
+ output_dir: /home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/evaluation/new_parallel32
160
+ video:
161
+ enabled: false
162
+ num_bins: 1
163
+ save_step_records: true
164
+ save_action_records: true
165
+ save_latency_records: true
166
+ realtime_pipeline_profile: false
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/manifest.json CHANGED
@@ -1,15 +1,15 @@
1
  {
2
- "dataset_name": "walker2d_h1_5fps_profile",
3
  "env_name": "walker2d",
4
  "episodes": 90,
5
- "frames": 88708,
6
  "task_prompts": [
7
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.",
8
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (400.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.",
9
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 3 raw frames (600.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action."
10
  ],
11
  "format": "lerobot_v2_image_parquet",
12
- "source": "${PI05_RUN_DIR}/profile_latency/data/raw_profile",
13
  "integration_name": "gymnasium",
14
  "task_name": "walker2d",
15
  "action_layout": "gymnasium_continuous_v1",
@@ -59,16 +59,16 @@
59
  "state_normalization": {
60
  "type": "min_max",
61
  "min": [
62
- 0.801384687423706,
63
- -0.3846876919269562,
64
- -1.3016910552978516,
65
- -2.7476444244384766,
66
- -1.5346914529800415,
67
- -0.8290755152702332,
68
- -2.1717329025268555,
69
- -1.4086319208145142,
70
- -0.415105402469635,
71
- -4.531341552734375,
72
  -10.0,
73
  -10.0,
74
  -10.0,
@@ -78,16 +78,16 @@
78
  -10.0
79
  ],
80
  "max": [
81
- 1.7503005266189575,
82
- 0.9993473291397095,
83
- 0.15303508937358856,
84
- 0.31470051407814026,
85
- 1.3454551696777344,
86
- 0.16739727556705475,
87
- 0.3809681832790375,
88
- 1.2573986053466797,
89
- 9.562085151672363,
90
- 3.402139902114868,
91
  10.0,
92
  10.0,
93
  10.0,
@@ -98,8 +98,8 @@
98
  ]
99
  },
100
  "robot_type": "rl_games_gymnasium",
101
- "latency_prompt_map_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/walker2d_h1_5fps_profile/latency_prompt_map.json",
102
- "custom_mixtures_path": "${PI05_RUN_DIR}/profile_latency/data/lerobot/_generated_mixtures/walker2d_h1_5fps_profile.json",
103
  "gymnasium_task": {
104
  "action_space": {
105
  "dtype": "float32",
@@ -155,7 +155,7 @@
155
  ],
156
  "obs_fps": 5.0,
157
  "registration_imports": [
158
- "latency_bench.envs.gymnasium_walker2d"
159
  ],
160
  "state_space": {
161
  "labels": [
@@ -180,7 +180,7 @@
180
  },
181
  "task_name": "walker2d"
182
  },
183
- "validation_dataset_name": "walker2d_h1_5fps_profile__val",
184
  "validation_episodes": 10,
185
  "validation_frames": 10000
186
  }
 
1
  {
2
+ "dataset_name": "lerobot",
3
  "env_name": "walker2d",
4
  "episodes": 90,
5
+ "frames": 89169,
6
  "task_prompts": [
7
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 1 raw frames (200.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.",
8
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 2 raw frames (400.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action.",
9
  "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot. Current action latency is 3 raw frames (600.00 ms). The environment runs at 5 FPS and observations are emitted at 5 FPS. Choose the best next action."
10
  ],
11
  "format": "lerobot_v2_image_parquet",
12
+ "source": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/raw",
13
  "integration_name": "gymnasium",
14
  "task_name": "walker2d",
15
  "action_layout": "gymnasium_continuous_v1",
 
59
  "state_normalization": {
60
  "type": "min_max",
61
  "min": [
62
+ 0.8091598153114319,
63
+ -0.958375871181488,
64
+ -1.471215009689331,
65
+ -2.750671863555908,
66
+ -1.5357930660247803,
67
+ -1.3458030223846436,
68
+ -1.5723819732666016,
69
+ -1.5042258501052856,
70
+ -0.7429636120796204,
71
+ -5.120609760284424,
72
  -10.0,
73
  -10.0,
74
  -10.0,
 
78
  -10.0
79
  ],
80
  "max": [
81
+ 1.7504061460494995,
82
+ 0.9881704449653625,
83
+ 0.16799136996269226,
84
+ 0.33612364530563354,
85
+ 1.3760323524475098,
86
+ 0.14901067316532135,
87
+ 0.335942804813385,
88
+ 1.3045134544372559,
89
+ 9.560564041137695,
90
+ 3.40214204788208,
91
  10.0,
92
  10.0,
93
  10.0,
 
98
  ]
99
  },
100
  "robot_type": "rl_games_gymnasium",
101
+ "latency_prompt_map_path": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/lerobot/latency_prompt_map.json",
102
+ "custom_mixtures_path": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/_generated_mixtures/lerobot.json",
103
  "gymnasium_task": {
104
  "action_space": {
105
  "dtype": "float32",
 
155
  ],
156
  "obs_fps": 5.0,
157
  "registration_imports": [
158
+ "latency_bench.envs.tasks.walker2d"
159
  ],
160
  "state_space": {
161
  "labels": [
 
180
  },
181
  "task_name": "walker2d"
182
  },
183
+ "validation_dataset_name": "lerobot__val",
184
  "validation_episodes": 10,
185
  "validation_frames": 10000
186
  }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/normalization_binding.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "weights": "unchanged",
3
+ "original_statistics": "original_training_statistics.json",
4
+ "state_transform": "Raw-state min/max from the source manifest reproduces converter-normalized training inputs.",
5
+ "manifest": "/home/ubuntu/lzj/latency-sensitive-bench/outputs/pi05_issued_action_iid_s42_20261010/walker2d/new_bundle/manifest.json"
6
+ }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/original_training_statistics.json ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "action": {
4
+ "mean": [
5
+ 0.4981546700000763,
6
+ 0.4653775095939636,
7
+ -0.2947416603565216,
8
+ 0.9996105432510376,
9
+ 0.8025897741317749,
10
+ 0.38950979709625244
11
+ ],
12
+ "std": [
13
+ 0.5788458585739136,
14
+ 0.48928776383399963,
15
+ 0.7716024518013,
16
+ 0.01672535389661623,
17
+ 0.4969850182533264,
18
+ 0.6478453278541565
19
+ ],
20
+ "max": [
21
+ 1.0,
22
+ 1.0,
23
+ 1.0,
24
+ 1.0,
25
+ 1.0,
26
+ 1.0
27
+ ],
28
+ "min": [
29
+ -1.0,
30
+ -1.0,
31
+ -1.0,
32
+ -1.0,
33
+ -1.0,
34
+ -1.0
35
+ ],
36
+ "q01": [
37
+ -1.0,
38
+ -1.0,
39
+ -1.0,
40
+ 1.0,
41
+ -1.0,
42
+ -1.0
43
+ ],
44
+ "q99": [
45
+ 1.0,
46
+ 1.0,
47
+ 1.0,
48
+ 1.0,
49
+ 1.0,
50
+ 1.0
51
+ ],
52
+ "mask": [
53
+ true,
54
+ true,
55
+ true,
56
+ true,
57
+ true,
58
+ true
59
+ ]
60
+ },
61
+ "state": {
62
+ "mean": [
63
+ 0.23792263865470886,
64
+ 0.18470169603824615,
65
+ 0.5776959657669067,
66
+ 0.6617908477783203,
67
+ -0.12576328217983246,
68
+ 0.8187075853347778,
69
+ 0.5637369155883789,
70
+ 0.43144461512565613,
71
+ -0.1433890163898468,
72
+ 0.20708228647708893,
73
+ -0.007593836635351181,
74
+ -0.012343162670731544,
75
+ -0.04303945600986481,
76
+ 0.010517130605876446,
77
+ -0.002804339397698641,
78
+ -0.02231612242758274,
79
+ -0.04842042177915573
80
+ ],
81
+ "std": [
82
+ 0.3836668133735657,
83
+ 0.15493330359458923,
84
+ 0.2817511558532715,
85
+ 0.24344688653945923,
86
+ 0.5168323516845703,
87
+ 0.07930017262697185,
88
+ 0.27416202425956726,
89
+ 0.4340347349643707,
90
+ 0.2744417190551758,
91
+ 0.34808868169784546,
92
+ 0.30759069323539734,
93
+ 0.279937207698822,
94
+ 0.44789883494377136,
95
+ 0.5409497022628784,
96
+ 0.17595402896404266,
97
+ 0.40527644753456116,
98
+ 0.5279714465141296
99
+ ],
100
+ "max": [
101
+ 1.0,
102
+ 1.0,
103
+ 1.0,
104
+ 1.0,
105
+ 1.0,
106
+ 1.0,
107
+ 1.0,
108
+ 1.0,
109
+ 1.0,
110
+ 1.0,
111
+ 1.0,
112
+ 1.0,
113
+ 1.0,
114
+ 1.0,
115
+ 1.0,
116
+ 1.0,
117
+ 1.0
118
+ ],
119
+ "min": [
120
+ -1.0,
121
+ -1.0,
122
+ -1.0,
123
+ -1.0,
124
+ -1.0,
125
+ -1.0,
126
+ -1.0,
127
+ -1.0,
128
+ -1.0,
129
+ -1.0,
130
+ -1.0,
131
+ -1.0,
132
+ -1.0,
133
+ -1.0,
134
+ -1.0,
135
+ -1.0,
136
+ -1.0
137
+ ],
138
+ "q01": [
139
+ -0.4113236618041992,
140
+ -0.21736109972000123,
141
+ -0.18296106100082396,
142
+ -0.07217283248901366,
143
+ -0.7223247408866882,
144
+ 0.47785244464874266,
145
+ -0.3314486837387085,
146
+ -0.7239453530311585,
147
+ -0.8191078186035157,
148
+ -0.44607192516326905,
149
+ -0.5727321386337281,
150
+ -0.5157614970207214,
151
+ -1.0,
152
+ -1.0,
153
+ -0.396851544380188,
154
+ -1.0,
155
+ -1.0
156
+ ],
157
+ "q99": [
158
+ 0.8920213985443114,
159
+ 0.4974299812316893,
160
+ 0.9348369693756103,
161
+ 0.9200620412826538,
162
+ 0.8252453470230101,
163
+ 0.9311845254898071,
164
+ 0.8568081760406492,
165
+ 0.8925079727172848,
166
+ 0.6340228366851779,
167
+ 0.9249057865142821,
168
+ 1.0,
169
+ 1.0,
170
+ 1.0,
171
+ 1.0,
172
+ 1.0,
173
+ 1.0,
174
+ 1.0
175
+ ]
176
+ },
177
+ "num_transitions": 89169,
178
+ "num_trajectories": 90
179
+ }
180
+ }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/provenance.json DELETED
@@ -1,34 +0,0 @@
1
- {
2
- "task": "walker2d",
3
- "model": "qwenpi_v3",
4
- "training_condition": "latency-aware",
5
- "training_run_id": "walker2d_pi05_profile_h1_5fps_g128_20260922_m32",
6
- "source": {
7
- "repo_id": "latency-sensitive-bench/Standard-Pipeline",
8
- "revision": "2208875f92b21e3f8ec2363a30a6705dfe56cc04",
9
- "path": "walker2d/profile_latency/Pi05",
10
- "url": "https://huggingface.co/latency-sensitive-bench/Standard-Pipeline/tree/2208875f92b21e3f8ec2363a30a6705dfe56cc04/walker2d/profile_latency/Pi05"
11
- },
12
- "hf_repo_id": "latency-sensitive-bench/benchmark-models",
13
- "hf_path": "latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32",
14
- "checkpoint": {
15
- "source_file": "walker2d/profile_latency/Pi05/checkpoints/model.pt",
16
- "source_sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84",
17
- "source_bytes": 10922653853,
18
- "file": "checkpoints/model.pt",
19
- "selection_rule": "only published VLA checkpoint",
20
- "method": "copy",
21
- "sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84",
22
- "bytes": 10922653853
23
- },
24
- "config_source": "walker2d/profile_latency/Pi05/config.yaml",
25
- "task_contract_file": "task_contract.json",
26
- "missing_source_files": [],
27
- "migration": {
28
- "operation": "copy_selected_inference_bundle",
29
- "source_modified": false,
30
- "evaluation_performed": false,
31
- "experiment_status": "preserved_in_source_evidence"
32
- },
33
- "action_horizon": 1
34
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/README.md DELETED
@@ -1,5 +0,0 @@
1
- # walker2d Pi0.5 profile-latency VLA
2
-
3
- Fresh QwenPI_v3,5000 updates,seed42,global128 (micro32 × accumulation2 (explicitly approved after original micro64 OOM) ×2 GPUs),GC off,ZeRO2,H1,RGB224,5/5FPS,native state/action and train-only state min-max once. Dataset retained the >3000 gate: 100 accepted episodes in 118 total attempts including the original probe.
4
-
5
- Set PI05_BACKBONE_DIR to Qwen3-VL-4B-Instruct revision ebb281ec70b05090aa6165b016eac8ec08e71b17. Use config.yaml/checkpoints/model.pt and latency_prompt_map.json. Weights are byte-identical to final training export. CPU tensor-finiteness/hash validation passed; actual realtime F evaluation is a separate stage. No optimizer, training state, W&B cache or credentials are included.
 
 
 
 
 
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/source/provenance.json DELETED
@@ -1,96 +0,0 @@
1
- {
2
- "task": "walker2d",
3
- "backbone": {
4
- "repo_id": "Qwen/Qwen3-VL-4B-Instruct",
5
- "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17"
6
- },
7
- "checkpoint": "steps_5000_pytorch_model.pt",
8
- "training_steps": 5000,
9
- "seed": 42,
10
- "global_batch": 128,
11
- "training_run_id": "walker2d_pi05_profile_h1_5fps_g128_20260922_m32",
12
- "condition": "profile_latency",
13
- "source": {
14
- "task": "walker2d",
15
- "teacher_verification_sha256": "20a4dde06a7ab677d8cb59116b3cb69d75204aa0fc463e8c48511a69288920e2",
16
- "profile_sha256": "b5409f8578bec62e2c5232c3916b3ddd2e0e9494031259265540ba6aef882e5e",
17
- "profile_publication_revision": "e1a9bdd566b96d5da4703f769d40eab656af491e",
18
- "selected_checkpoint": "${PI05_RUN_DIR}/profile_latency/teacher/checkpoints/pi05_walker2d_profile_appo_h1_5fps_g128_20260921/checkpoint_p0/best_000018928_9691136_reward_4148.139.pth",
19
- "selected_checkpoint_sha256": "3d51959c7d65d8e9d1ec34cfc4a8dc7741abf9726439512bc7a0442fd265044e",
20
- "selected_checkpoint_env_steps": 9691136,
21
- "probe12_state_sha256": "266f2ec22ca31030bcc2e2161543314b579d178a1f45ff428b3278aa0bdead58",
22
- "full_probe_state_sha256": "369efebe9c5abd237b02f5d368c4d1f21b0533fea1c3d9f59e501484d386f67f",
23
- "data_attempts": 118,
24
- "data_accepted": 100,
25
- "data_rejected": 18,
26
- "splits": {
27
- "train": {
28
- "rows": 88708,
29
- "episodes": 90,
30
- "return_min": 3049.0915127396584,
31
- "return_max": 4926.397267758846,
32
- "length_min": 707,
33
- "length_max": 1000
34
- },
35
- "val": {
36
- "rows": 10000,
37
- "episodes": 10,
38
- "return_min": 4344.816579580307,
39
- "return_max": 4727.460265636444,
40
- "length_min": 1000,
41
- "length_max": 1000
42
- }
43
- },
44
- "state_dim": 17,
45
- "native_action_dim": 6,
46
- "action_carrier": "native",
47
- "state_normalization": {
48
- "type": "min_max",
49
- "min": [
50
- 0.801384687423706,
51
- -0.3846876919269562,
52
- -1.3016910552978516,
53
- -2.7476444244384766,
54
- -1.5346914529800415,
55
- -0.8290755152702332,
56
- -2.1717329025268555,
57
- -1.4086319208145142,
58
- -0.415105402469635,
59
- -4.531341552734375,
60
- -10.0,
61
- -10.0,
62
- -10.0,
63
- -10.0,
64
- -10.0,
65
- -10.0,
66
- -10.0
67
- ],
68
- "max": [
69
- 1.7503005266189575,
70
- 0.9993473291397095,
71
- 0.15303508937358856,
72
- 0.31470051407814026,
73
- 1.3454551696777344,
74
- 0.16739727556705475,
75
- 0.3809681832790375,
76
- 1.2573986053466797,
77
- 9.562085151672363,
78
- 3.402139902114868,
79
- 10.0,
80
- 10.0,
81
- 10.0,
82
- 10.0,
83
- 10.0,
84
- 10.0,
85
- 10.0
86
- ]
87
- },
88
- "raw_metadata_sha256": "f01614b1df2395b2a4ae5f7bcaf7c215c42a029543ee7d31758ce49170fdbea2",
89
- "raw_train_parquet_sha256": "755e77d3f27a1e73ae2e01b720288b8ea261cbaaffb537b93794f3f8b2bae595",
90
- "raw_val_parquet_sha256": "eaff1190ab7ecb98aa95051e06e9e5efc8dcdcb2054e3464b73da2cc286e824e",
91
- "converted_manifest_sha256": "e850ecfe50be72061722c0a57d95efb797f705e5c0aa1049c56f9d0bec54fbdb",
92
- "latency_prompt_map_sha256": "ea71f353aa39bd39ca32d0fdda58cac6a0d0ae158f948241d8d494d261e1c3bc"
93
- },
94
- "training_config_sha256": "385d1a431384f36a79d4d6531ea26da7e4751a6924aac343ed685d268440f83a",
95
- "dataset_manifest_sha256": "e850ecfe50be72061722c0a57d95efb797f705e5c0aa1049c56f9d0bec54fbdb"
96
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/task_contract.json DELETED
@@ -1,80 +0,0 @@
1
- {
2
- "action_space": {
3
- "dtype": "float32",
4
- "high": [
5
- 1.0,
6
- 1.0,
7
- 1.0,
8
- 1.0,
9
- 1.0,
10
- 1.0
11
- ],
12
- "labels": [
13
- "right_thigh_torque",
14
- "right_leg_torque",
15
- "right_foot_torque",
16
- "left_thigh_torque",
17
- "left_leg_torque",
18
- "left_foot_torque"
19
- ],
20
- "low": [
21
- -1.0,
22
- -1.0,
23
- -1.0,
24
- -1.0,
25
- -1.0,
26
- -1.0
27
- ],
28
- "type": "box"
29
- },
30
- "base_prompt": "Move the Walker2d robot forward while keeping its torso upright. Predict six continuous torques in [-1, 1] ordered as right thigh, right leg, right foot, left thigh, left leg, and left foot.",
31
- "env_fps": 5.0,
32
- "env_id": "LatencyBench/Walker2dContinuous-v0",
33
- "frame_stack": 1,
34
- "make_kwargs": {
35
- "base_env_id": "Walker2d-v4",
36
- "base_make_kwargs": {
37
- "ctrl_cost_weight": 0.001,
38
- "exclude_current_positions_from_observation": true,
39
- "forward_reward_weight": 1.0,
40
- "healthy_reward": 1.0,
41
- "reset_noise_scale": 0.005,
42
- "terminate_when_unhealthy": true
43
- },
44
- "render_mode": "rgb_array"
45
- },
46
- "noop_action": [
47
- 0.0,
48
- 0.0,
49
- 0.0,
50
- 0.0,
51
- 0.0,
52
- 0.0
53
- ],
54
- "obs_fps": 5.0,
55
- "registration_imports": [
56
- "latency_bench.envs.gymnasium_walker2d"
57
- ],
58
- "state_space": {
59
- "labels": [
60
- "torso_height",
61
- "torso_angle",
62
- "right_thigh_angle",
63
- "right_leg_angle",
64
- "right_foot_angle",
65
- "left_thigh_angle",
66
- "left_leg_angle",
67
- "left_foot_angle",
68
- "torso_x_velocity",
69
- "torso_z_velocity",
70
- "torso_angular_velocity",
71
- "right_thigh_angular_velocity",
72
- "right_leg_angular_velocity",
73
- "right_foot_angular_velocity",
74
- "left_thigh_angular_velocity",
75
- "left_leg_angular_velocity",
76
- "left_foot_angular_velocity"
77
- ]
78
- },
79
- "task_name": "walker2d"
80
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/training.complete.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "steps": 5000,
3
+ "seed": 42,
4
+ "global_batch": 128,
5
+ "gradient_checkpointing": true,
6
+ "gpus": "2,3",
7
+ "attempts": [
8
+ {
9
+ "micro_batch": 64,
10
+ "accumulation": 1,
11
+ "state": "COMPLETED"
12
+ }
13
+ ]
14
+ }
latency-aware/walker2d/vla/starvla-qwenpi_v3-h1/walker2d_pi05_profile_h1_5fps_g128_20260922_m32/validation.json DELETED
@@ -1,10 +0,0 @@
1
- {
2
- "state": "CPU_STATE_DICT_AND_BUNDLE_HASH_VERIFIED",
3
- "checkpoint_sha256": "248d8959f1c4b4045b64fd0c4f7a8e086606eea5827f263b04ef703e26b4cd84",
4
- "tensor_count": 1386,
5
- "all_floating_tensors_finite": true,
6
- "training_steps": 5000,
7
- "global_batch": 128,
8
- "gpu_forward": "NOT_RUN_AT_PUBLICATION",
9
- "verified_at": "2026-09-22T01:32:01.661335+00:00"
10
- }