farguney commited on
Commit
271e217
·
verified ·
1 Parent(s): d4db3f7

Upload language_breakthrough_v1/b34_reasoning_a100_audit_smoke/b34-reasoning-rung2-a100-audit-20260809T051110Z-5fd7cd0f.result.signed.json with huggingface_hub

Browse files
language_breakthrough_v1/b34_reasoning_a100_audit_smoke/b34-reasoning-rung2-a100-audit-20260809T051110Z-5fd7cd0f.result.signed.json ADDED
@@ -0,0 +1,402 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "payload": {
3
+ "abi": "TRIDENT-B34-REASONING-CURRICULUM-A100X4-GATE-v1-A100X4-AUDIT-SMOKE-RESULT",
4
+ "approved_main_run_id": "b34-reasoning-rung2-20260809T051500Z-a4c19e72",
5
+ "audit_run_id": "b34-reasoning-rung2-a100-audit-20260809T051110Z-5fd7cd0f",
6
+ "claim_state": "REAL-A100X4-AUDIT-DDP-FIRST-STEP-MEASURED",
7
+ "cost": {
8
+ "billed_estimate_usd": 1.0756341915390093,
9
+ "elapsed_seconds": 387.2283089540433,
10
+ "hard_ceiling_usd": 2.0,
11
+ "hourly_usd": 10.0,
12
+ "timeout_minutes": 12
13
+ },
14
+ "created_at": "2026-08-09T05:18:21.100598+00:00",
15
+ "eval_decontamination": {
16
+ "algorithm": "lowercase alphanumeric/underscore 13-word SHA-256 shingles",
17
+ "base_documents_after": 25623,
18
+ "base_documents_before": 25623,
19
+ "base_documents_rejected_by_source": {
20
+ "algebraic_stack_permissive": 0,
21
+ "codeparrot_permissive": 0,
22
+ "gutenberg_en": 0,
23
+ "wikipedia_en": 0
24
+ },
25
+ "comparator_outputs_loaded": false,
26
+ "fingerprints_entered_neural_path": false,
27
+ "forbidden_shingle_count": 8822,
28
+ "passed": true,
29
+ "reasoning_blocks_after": 1,
30
+ "reasoning_blocks_before": 1,
31
+ "reasoning_blocks_rejected": 0,
32
+ "section_7_5_raw_prompts_entered_neural_path": false
33
+ },
34
+ "evaluation_artifact_staging": {
35
+ "contents_opened_before_training": false,
36
+ "manifest": {
37
+ "file_sha256": "d3f2da346429a60af19ec470026888a3d4770c337dc70e534eebac4822e35190",
38
+ "path": "results/capability_yardstick_v1/dataset_manifest.json"
39
+ },
40
+ "neural_path_access": false,
41
+ "passed": true,
42
+ "preregistration": {
43
+ "file_sha256": "e017026c98bc64e5bb05928a3c7dc91dbffa6e6cbd78a3c0528119cf0a4fe970",
44
+ "path": "results/capability_yardstick_v1/preregistration.signed.json"
45
+ }
46
+ },
47
+ "failed_bars": [],
48
+ "finite_optimizer_steps": true,
49
+ "full_OSS_data_staging": {
50
+ "all_nonprimary_ranks_reopened_same_cache": true,
51
+ "passed": true,
52
+ "primary_wrote_one_cache": true,
53
+ "rank_private_mkdtemp_used": false,
54
+ "shared_workdir": "/tmp/trident-b34-full-data-audit-b34-reasoning-rung2-a100-audit-20260809T051110Z-5fd7cd0f",
55
+ "source_count": 4,
56
+ "source_names": [
57
+ "algebraic_stack_permissive",
58
+ "codeparrot_permissive",
59
+ "gutenberg_en",
60
+ "wikipedia_en"
61
+ ]
62
+ },
63
+ "hardware": {
64
+ "a100x4_exact": true,
65
+ "eight_GPU_forbidden": true,
66
+ "gather_validated_on_primary_then_broadcast": true,
67
+ "local_world_size": 4,
68
+ "maximum_gpu_count": 4,
69
+ "passed": true,
70
+ "rank_records": [
71
+ {
72
+ "cuda_visible_devices": null,
73
+ "device_name": "NVIDIA A100-SXM4-80GB",
74
+ "hostname": "j-deepadapt-6a780befda2af92a634efdaa-irlns637-3c173-zqpc6",
75
+ "local_rank": 0,
76
+ "local_world_size": 4,
77
+ "nvidia_visible_devices": "void",
78
+ "rank": 0,
79
+ "selected_device_index": 0,
80
+ "visible_cuda_devices": 4,
81
+ "world_size": 4
82
+ },
83
+ {
84
+ "cuda_visible_devices": null,
85
+ "device_name": "NVIDIA A100-SXM4-80GB",
86
+ "hostname": "j-deepadapt-6a780befda2af92a634efdaa-irlns637-3c173-zqpc6",
87
+ "local_rank": 1,
88
+ "local_world_size": 4,
89
+ "nvidia_visible_devices": "void",
90
+ "rank": 1,
91
+ "selected_device_index": 1,
92
+ "visible_cuda_devices": 4,
93
+ "world_size": 4
94
+ },
95
+ {
96
+ "cuda_visible_devices": null,
97
+ "device_name": "NVIDIA A100-SXM4-80GB",
98
+ "hostname": "j-deepadapt-6a780befda2af92a634efdaa-irlns637-3c173-zqpc6",
99
+ "local_rank": 2,
100
+ "local_world_size": 4,
101
+ "nvidia_visible_devices": "void",
102
+ "rank": 2,
103
+ "selected_device_index": 2,
104
+ "visible_cuda_devices": 4,
105
+ "world_size": 4
106
+ },
107
+ {
108
+ "cuda_visible_devices": null,
109
+ "device_name": "NVIDIA A100-SXM4-80GB",
110
+ "hostname": "j-deepadapt-6a780befda2af92a634efdaa-irlns637-3c173-zqpc6",
111
+ "local_rank": 3,
112
+ "local_world_size": 4,
113
+ "nvidia_visible_devices": "void",
114
+ "rank": 3,
115
+ "selected_device_index": 3,
116
+ "visible_cuda_devices": 4,
117
+ "world_size": 4
118
+ }
119
+ ],
120
+ "single_node": true,
121
+ "visibility_mode": "all-four-visible-per-rank",
122
+ "world_size": 4
123
+ },
124
+ "launch_authorization": {
125
+ "GPU_job_namespace": "deepadapt",
126
+ "abi": "TRIDENT-OWNER-TEXTUAL-AUTHORIZATION-v1",
127
+ "authorization_path": "OWNER-TEXTUAL-AUTHORIZATION-OF-RECORD",
128
+ "billing_namespace": "deepadapt",
129
+ "created_at": "2026-08-09T05:10:41.183792+00:00",
130
+ "decision": "LAUNCH_APPROVED",
131
+ "owner_record": {
132
+ "abi": "TRIDENT-OWNER-TEXTUAL-AUTHORIZATION-v1",
133
+ "authority": "owner-directive-authorization-of-record",
134
+ "cryptographic_signature_claimed": false,
135
+ "deadline": "2026-08-12T00:58:00+00:00",
136
+ "pinned_owner_key_forged_or_reused": false,
137
+ "registry_capabilities_sha256": "d76cb6adc92c260b83486839222df37a7a259d9cc81ed00fc6027592fa81ed89",
138
+ "registry_key": "owner_binding_spend_authorization_72h_countdown_v1",
139
+ "rung_caps_usd": {
140
+ "rung1": 50.0,
141
+ "rung2": 300.0,
142
+ "rung3": 5000.0
143
+ },
144
+ "scheme": "TRIDENT-GIT-RECORDED-OWNER-TEXT-v1",
145
+ "source_commit": "2f1b13d15f4aa0e8b05f5deb2936009b7b30ea95",
146
+ "t0": "2026-08-09T00:58:00+00:00",
147
+ "trident_md_sha256": "72a843a78568dcc6ed42d79ecbe978c62bffc7477de6f02638969c333afc0e7a",
148
+ "update_marker": "2026-08-09T01:15:00+00:00"
149
+ },
150
+ "owner_record_git_verification": {
151
+ "registry_capabilities_sha256": "d76cb6adc92c260b83486839222df37a7a259d9cc81ed00fc6027592fa81ed89",
152
+ "source_commit": "2f1b13d15f4aa0e8b05f5deb2936009b7b30ea95",
153
+ "trident_md_sha256": "72a843a78568dcc6ed42d79ecbe978c62bffc7477de6f02638969c333afc0e7a"
154
+ },
155
+ "package_file_sha256": "c0a9ab4e4307a3f49b069d066775b78b3e99e10983671164cd1ba251a935a712",
156
+ "package_hard_abort_usd": 250.0,
157
+ "package_payload_sha256": "8195035d135e6600faffb8a677f4fba9b95c077c915a7275fdad9f58f868bf75",
158
+ "pinned_owner_key_forged_or_reused": false,
159
+ "private_key_used": false,
160
+ "run_id": "b34-reasoning-rung2-20260809T051500Z-a4c19e72",
161
+ "rung1": {
162
+ "flavor": "a10g-small",
163
+ "gpu_count": 1,
164
+ "maximum_spend_usd": 50.0
165
+ },
166
+ "rung2": {
167
+ "flavor": "a100x4",
168
+ "gpu_count": 4,
169
+ "maximum_spend_usd": 300.0,
170
+ "updates": 3000
171
+ },
172
+ "rung3": {
173
+ "authorized_by_owner_record": true,
174
+ "flavor": "h200x4",
175
+ "gpu_count": 4,
176
+ "maximum_spend_usd": 5000.0,
177
+ "not_exercised_by_this_gate_token": true
178
+ },
179
+ "verified_authorization_path": "TRIDENT-GIT-RECORDED-OWNER-TEXT-v1"
180
+ },
181
+ "package": {
182
+ "file_sha256": "c0a9ab4e4307a3f49b069d066775b78b3e99e10983671164cd1ba251a935a712",
183
+ "path": "results/b34_task_accuracy_pivot_v1/gate_package.signed.json",
184
+ "payload_sha256": "8195035d135e6600faffb8a677f4fba9b95c077c915a7275fdad9f58f868bf75",
185
+ "signature_verified": true
186
+ },
187
+ "proofwriter": {
188
+ "bytes": 15965941,
189
+ "license": "Apache-2.0",
190
+ "local_sha256": "a37758ca145e681935ec371538b6b36ed97710260651a43ce3825256ed7a84d9",
191
+ "path": "data/train-00000-of-00002-6176cf3d78f48858.parquet",
192
+ "rendered_block_bytes": 128,
193
+ "rendered_block_root_sha256": "8548444c280d20cc9374a9a429a58a0385fb516b4b8605e5e15b7908dc2b78b0",
194
+ "repo": "tasksource/proofwriter",
195
+ "revision": "ee4d22d49b8486786cbc01c0b6f6f6266e56f3bd",
196
+ "rows": 292776,
197
+ "section_7_5_or_comparator_data": false,
198
+ "selected_rows": 1,
199
+ "selection": "lowest SHA256(TRIDENT-B34-REASONING-CURRICULUM-A100X4-GATE-v1|id|question)"
200
+ },
201
+ "root": {
202
+ "parameters": 927458306,
203
+ "sha256": "f18f36417f678f0d47092a4d7d3d26bb85a65eb8a39c557c8dc3b386699eb8a1",
204
+ "strict_load_required": true
205
+ },
206
+ "section_7_5_or_comparator_data_used": false,
207
+ "task_accuracy_claimed": false,
208
+ "terminal_train_authorized": false,
209
+ "train_core_prelude": {
210
+ "chunked_delta_parity": {
211
+ "chunk": 64,
212
+ "enabled": true,
213
+ "live_state": {
214
+ "deployment_state_detach_unchanged": true,
215
+ "maximum_absolute_delta": 0.0,
216
+ "passed": true,
217
+ "tolerance": 2e-05
218
+ },
219
+ "n_delta_blocks": 1,
220
+ "reasoning_curriculum": true,
221
+ "worst_rel_err": 5.451602191897109e-07
222
+ },
223
+ "cuda_device_count": 4,
224
+ "cuda_peak_memory_bytes": 20615501824,
225
+ "device": "cuda:0",
226
+ "family": "Core5 parameter-neutral PCOF",
227
+ "first_measurement_micro_step_reached": true,
228
+ "first_training_step": {
229
+ "gradient_ancestry": {
230
+ "disconnected": [],
231
+ "loss_ancestor": "causal raw-byte codelength terms only",
232
+ "nonfinite": [],
233
+ "passed": true,
234
+ "trainable_tensor_count": 94,
235
+ "zero_gradient": []
236
+ },
237
+ "gradient_clip_norm": 0.5,
238
+ "gradient_norm_before_clip": 26.847230911254883,
239
+ "learning_rate": 1.25e-05,
240
+ "loss_bpb": 2.5060086250305176,
241
+ "optimizer_step_completed": true,
242
+ "training_objective_audit": {
243
+ "accuracy_verifier_reward_gradient": false,
244
+ "all_targets_are_ordinary_future_bytes": true,
245
+ "applied_global_rows": [
246
+ 0
247
+ ],
248
+ "event_codelength_bits": [
249
+ 4.295586109161377
250
+ ],
251
+ "objective": "uniform raw next-byte L_PC + retrieved OSS reasoning trajectory L_PC + PCOF action/compute rate",
252
+ "reasoning_event_weight": 0.25,
253
+ "task_or_relation_id_neural_input": false,
254
+ "transport_direction_constructed": false,
255
+ "update": 1
256
+ },
257
+ "update": 1
258
+ },
259
+ "gate_enable_passed": true,
260
+ "geometry": {
261
+ "block": 128,
262
+ "global_batch": 4,
263
+ "updates": 2
264
+ },
265
+ "initial_measurement": {
266
+ "block_bpb": [
267
+ 1.554079532623291
268
+ ],
269
+ "by_source": {
270
+ "a100_audit_smoke": {
271
+ "blocks": 1,
272
+ "mean_bpb": 1.554079532623291
273
+ }
274
+ },
275
+ "mean_bpb": 1.554079532623291
276
+ },
277
+ "last_training_objective_audit": null,
278
+ "last_training_step": {
279
+ "gradient_ancestry": {
280
+ "disconnected": [],
281
+ "loss_ancestor": "causal raw-byte codelength terms only",
282
+ "nonfinite": [],
283
+ "passed": true,
284
+ "trainable_tensor_count": 94,
285
+ "zero_gradient": []
286
+ },
287
+ "gradient_clip_norm": 0.5,
288
+ "gradient_norm_before_clip": 16.25172233581543,
289
+ "learning_rate": 2.5e-05,
290
+ "loss_bpb": 1.7890183925628662,
291
+ "optimizer_step_completed": true,
292
+ "training_objective_audit": {
293
+ "accuracy_verifier_reward_gradient": false,
294
+ "all_targets_are_ordinary_future_bytes": true,
295
+ "applied_global_rows": [
296
+ 0
297
+ ],
298
+ "event_codelength_bits": [
299
+ 2.909604072570801
300
+ ],
301
+ "objective": "uniform raw next-byte L_PC + retrieved OSS reasoning trajectory L_PC + PCOF action/compute rate",
302
+ "reasoning_event_weight": 0.25,
303
+ "task_or_relation_id_neural_input": false,
304
+ "transport_direction_constructed": false,
305
+ "update": 2
306
+ },
307
+ "update": 2
308
+ },
309
+ "mode": "real-CUDA full-width train_core prelude smoke",
310
+ "parameter_contract": {
311
+ "parameters": 927458306,
312
+ "tensor_count": 94,
313
+ "trainable_parameters": 927458306
314
+ },
315
+ "passed": true,
316
+ "post_step_measurement": {
317
+ "block_bpb": [
318
+ 1.5892614126205444
319
+ ],
320
+ "by_source": {
321
+ "a100_audit_smoke": {
322
+ "blocks": 1,
323
+ "mean_bpb": 1.5892614126205444
324
+ }
325
+ },
326
+ "mean_bpb": 1.5892614126205444
327
+ },
328
+ "stability_rows": [
329
+ {
330
+ "gradient_ancestry": {
331
+ "disconnected": [],
332
+ "loss_ancestor": "causal raw-byte codelength terms only",
333
+ "nonfinite": [],
334
+ "passed": true,
335
+ "trainable_tensor_count": 94,
336
+ "zero_gradient": []
337
+ },
338
+ "gradient_clip_norm": 0.5,
339
+ "gradient_norm_before_clip": 26.847230911254883,
340
+ "learning_rate": 1.25e-05,
341
+ "loss_bpb": 2.5060086250305176,
342
+ "optimizer_step_completed": true,
343
+ "training_objective_audit": {
344
+ "accuracy_verifier_reward_gradient": false,
345
+ "all_targets_are_ordinary_future_bytes": true,
346
+ "applied_global_rows": [
347
+ 0
348
+ ],
349
+ "event_codelength_bits": [
350
+ 4.295586109161377
351
+ ],
352
+ "objective": "uniform raw next-byte L_PC + retrieved OSS reasoning trajectory L_PC + PCOF action/compute rate",
353
+ "reasoning_event_weight": 0.25,
354
+ "task_or_relation_id_neural_input": false,
355
+ "transport_direction_constructed": false,
356
+ "update": 1
357
+ },
358
+ "update": 1
359
+ },
360
+ {
361
+ "gradient_ancestry": {
362
+ "disconnected": [],
363
+ "loss_ancestor": "causal raw-byte codelength terms only",
364
+ "nonfinite": [],
365
+ "passed": true,
366
+ "trainable_tensor_count": 94,
367
+ "zero_gradient": []
368
+ },
369
+ "gradient_clip_norm": 0.5,
370
+ "gradient_norm_before_clip": 16.25172233581543,
371
+ "learning_rate": 2.5e-05,
372
+ "loss_bpb": 1.7890183925628662,
373
+ "optimizer_step_completed": true,
374
+ "training_objective_audit": {
375
+ "accuracy_verifier_reward_gradient": false,
376
+ "all_targets_are_ordinary_future_bytes": true,
377
+ "applied_global_rows": [
378
+ 0
379
+ ],
380
+ "event_codelength_bits": [
381
+ 2.909604072570801
382
+ ],
383
+ "objective": "uniform raw next-byte L_PC + retrieved OSS reasoning trajectory L_PC + PCOF action/compute rate",
384
+ "reasoning_event_weight": 0.25,
385
+ "task_or_relation_id_neural_input": false,
386
+ "transport_direction_constructed": false,
387
+ "update": 2
388
+ },
389
+ "update": 2
390
+ }
391
+ ]
392
+ },
393
+ "verdict": "B34-REASONING-A100X4-AUDIT-DDP-FIRST-STEP-PASS"
394
+ },
395
+ "payload_sha256": "19326e9e05b3edb0d46ba17afd6ff0b0a2d58fa85742c7957f4786109b381d36",
396
+ "signature": {
397
+ "authority": "b34-reasoning-curriculum-gate-local",
398
+ "public_key_hex": "d66151f167757c664125db0548f830f3a57472f80cef141464d67368e0af9dbb",
399
+ "scheme": "TRIDENT-SIG-ED25519-v1",
400
+ "signature_hex": "0bacbeb5f9269b3af1a3d7c1c7e3d195d49944afd4e4ec5954755b8f2a566b3d2f3f32dcf968cc4d61096459ef3294b3346347e540e1251ad35e2006a5cba40c"
401
+ }
402
+ }