MarxistLeninist commited on
Commit
de48eb0
·
1 Parent(s): d3b4906

retention: prune 2 old checkpoints dirs

Browse files
checkpoints/step2847537_20260911/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.provenance.json DELETED
@@ -1,590 +0,0 @@
1
- {
2
- "batch_size": 56,
3
- "block_size": 2048,
4
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
5
- "checkpoint_type": "full",
6
- "created_at_iso": "2026-09-11T08:25:29Z",
7
- "created_at_unix": 1789115129.1128435,
8
- "created_from_scratch": false,
9
- "dataset_provenance": {
10
- "chat": false,
11
- "dataset_field_text": "text",
12
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
13
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
14
- "dataset_hotload_config_stamp": [
15
- 50,
16
- 1124990893,
17
- 1789091829464680808,
18
- 7210
19
- ],
20
- "dataset_hotload_config_used": true,
21
- "dataset_hotload_enabled": true,
22
- "dataset_hotload_generation": 13,
23
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
24
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
25
- "dataset_revisions": {
26
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
27
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
28
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
29
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
30
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
31
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
32
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
33
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
34
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
35
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
36
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
37
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
38
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
39
- },
40
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
41
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
42
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
43
- "has_language_mix": true,
44
- "has_numeracy": false,
45
- "hot_config_expected_sha256": "",
46
- "hot_config_has_append": true,
47
- "hot_config_has_datasets": false,
48
- "hot_config_mtime": 1789092253.2119184,
49
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
50
- "hot_config_used": true,
51
- "hot_reload_capable": true,
52
- "phase": "pretrain",
53
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
54
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
55
- "schema": "agillm.dataset_provenance.v1",
56
- "sequence_prefetch_depth": 256,
57
- "source_count": 23,
58
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
59
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
60
- "source_specs": [
61
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
62
- "wikimedia/wikipedia:20231101.en|1.63",
63
- "allenai/c4:en|0.36",
64
- "Skylion007/openwebtext|0.18",
65
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
66
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
67
- "EleutherAI/proof-pile-2:all|0.67",
68
- "allenai/dolma:v1_6-sample|0.5",
69
- "codeparrot/codeparrot-clean|0.24",
70
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
71
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
72
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
73
- "allenai/dolmino-mix-1124|0.78",
74
- "mlfoundations/dclm-baseline-1.0|1.05",
75
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
76
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
77
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
78
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
79
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
80
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
81
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
82
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
83
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
84
- ],
85
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
86
- "val_source_requested": ""
87
- },
88
- "effective_global_step": 5692827,
89
- "effective_seen_tok": 283463718912,
90
- "global_origin_seen_tok": 0,
91
- "global_origin_step": 2845290,
92
- "gpu": {
93
- "allocated_gb": 7.13,
94
- "peak_allocated_gb": 12.0,
95
- "reserved_gb": 13.34
96
- },
97
- "hostname": "ead6505f92c4",
98
- "lane": "pretrain",
99
- "launch": {
100
- "argv": [
101
- "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py",
102
- "train",
103
- "--preset",
104
- "agillm4_floor",
105
- "--block",
106
- "2048",
107
- "--source",
108
- "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
109
- "--nat_mask_token_id",
110
- "2",
111
- "--tie_weights",
112
- "--tie_kv",
113
- "--moe_ffn",
114
- "--moe_experts",
115
- "2",
116
- "--moe_top_k",
117
- "1",
118
- "--moe_mlp_mult",
119
- "4",
120
- "--moe_shared_experts",
121
- "1",
122
- "--moe_shared_mlp_mult",
123
- "2",
124
- "--moe_aux_coef",
125
- "0.01",
126
- "--moe_z_coef",
127
- "0.001",
128
- "--amp",
129
- "--grad_checkpoint",
130
- "--attn_backend",
131
- "sdpa",
132
- "--sublinear_window",
133
- "128",
134
- "--sublinear_stride",
135
- "128",
136
- "--sublinear_max_anchors",
137
- "128",
138
- "--sublinear_chunk",
139
- "128",
140
- "--sublinear_sinks",
141
- "4",
142
- "--sublinear_recent_anchors",
143
- "64",
144
- "--no-sublinear_pooled_landmarks",
145
- "--anchor_stride",
146
- "256",
147
- "--anchor_max",
148
- "2048",
149
- "--anchor_position",
150
- "-1",
151
- "--alibi_mode",
152
- "corrected",
153
- "--alibi_scale",
154
- "0.0",
155
- "--optimizer",
156
- "adamw8bit",
157
- "--weight_decay",
158
- "0.0",
159
- "--lr_decay",
160
- "cosine",
161
- "--lr_decay_tokens",
162
- "49000954880",
163
- "--lr_min_mult",
164
- "0.2",
165
- "--lr_schedule_reset_on_resume",
166
- "--lr_warmup_tokens",
167
- "10000000",
168
- "--lr_warmup_min_mult",
169
- "0.1",
170
- "--dblock",
171
- "--dblock_schedule",
172
- "roundrobin",
173
- "--dblock_router_ramp_steps",
174
- "64",
175
- "--dblock_warmup_steps",
176
- "14",
177
- "--dblock_checkpoint_skip_tail",
178
- "0",
179
- "--dblock_loop_layers",
180
- "0",
181
- "--dblock_loop_start",
182
- "0",
183
- "--dblock_loop_cond_scale",
184
- "1.0",
185
- "--dblock_sigma_curriculum_steps",
186
- "2000",
187
- "--dblock_sigma_sampling",
188
- "lognormal",
189
- "--dblock_sigma_stratified",
190
- "--dblock_sigma_min",
191
- "0.002",
192
- "--dblock_sigma_max",
193
- "2.0",
194
- "--dblock_sigma_pmean",
195
- "-1.2",
196
- "--dblock_sigma_pstd",
197
- "1.2",
198
- "--dblock_edm_wmax",
199
- "5.0",
200
- "--dblock_ar_prob",
201
- "0.8",
202
- "--dblock_sat_prob",
203
- "0.1",
204
- "--dblock_nat_prob",
205
- "0.1",
206
- "--dblock_ar_loss_tokens",
207
- "8192",
208
- "--dblock_sat_loss_tokens",
209
- "4096",
210
- "--dblock_nat_loss_tokens",
211
- "4096",
212
- "--dblock_nat_embed_noise_mode",
213
- "off",
214
- "--dblock_nat_embed_noise_scale",
215
- "1.0",
216
- "--nat_loss_weight",
217
- "1.0",
218
- "--nat_mask_ratio",
219
- "0.5",
220
- "--nat_max_tokens",
221
- "0",
222
- "--nat_span_mask_prob",
223
- "0.35",
224
- "--nat_suffix_mask_prob",
225
- "0.20",
226
- "--nat_span_max_tokens",
227
- "0",
228
- "--loss_spike_skip",
229
- "0.0",
230
- "--sat_every",
231
- "1",
232
- "--nat_every",
233
- "1",
234
- "--dblock_fullstack_anchor_spike_skip",
235
- "0.0",
236
- "--dblock_fullstack_anchor_max_ce",
237
- "0.0",
238
- "--dblock_anchor_softcap_mult",
239
- "4.0",
240
- "--dblock_anchor_softcap_min_ce",
241
- "25.0",
242
- "--dblock_anchor_softcap_max_ce",
243
- "50.0",
244
- "--dblock_fullstack_ar_every",
245
- "3",
246
- "--dblock_fullstack_ar_offset",
247
- "0",
248
- "--dblock_fullstack_ar_tokens",
249
- "256",
250
- "--dblock_fullstack_sat_every",
251
- "3",
252
- "--dblock_fullstack_sat_offset",
253
- "1",
254
- "--dblock_fullstack_sat_tokens",
255
- "128",
256
- "--dblock_fullstack_nat_every",
257
- "3",
258
- "--dblock_fullstack_nat_offset",
259
- "2",
260
- "--dblock_fullstack_nat_tokens",
261
- "128",
262
- "--dblock_fullstack_nat_mask_id",
263
- "-1",
264
- "--ckpt_codec",
265
- "block-sharded-zstd",
266
- "--async_update_every_steps",
267
- "0",
268
- "--dblock_router_hidden",
269
- "64",
270
- "--dblock_router_heads",
271
- "4",
272
- "--dblock_router_layers",
273
- "2",
274
- "--dblock_router_lr",
275
- "0.0005",
276
- "--dblock_explore",
277
- "0.08",
278
- "--dblock_max_stale_steps",
279
- "64",
280
- "--dblock_max_count_skew",
281
- "1.35",
282
- "--dblock_stale_bonus",
283
- "0.35",
284
- "--dblock_undertrain_bonus",
285
- "0.25",
286
- "--dblock_activation_offload_min_mb",
287
- "1.0",
288
- "--nat_document_boundary_aware",
289
- "--dblock_fullstack_anchor_deterministic_eval",
290
- "--dblock_fullstack_anchor_spike_retry_limit",
291
- "3",
292
- "--dblock_local_spike_retry_limit",
293
- "3",
294
- "--dblock_fullstack_nat_no_valid_crop_limit",
295
- "8",
296
- "--target_tokens",
297
- "400000000000",
298
- "--lr_schedule_reanchor_on_resume",
299
- "--save_every_sec",
300
- "1800",
301
- "--delta_max_keep",
302
- "0",
303
- "--max_ckpts",
304
- "2",
305
- "--disk_free_floor_gb",
306
- "12.0",
307
- "--heartbeat_every_sec",
308
- "30",
309
- "--dblock_fullstack_ar_weight",
310
- "1.0",
311
- "--dblock_fullstack_sat_weight",
312
- "0.1",
313
- "--dblock_fullstack_nat_weight",
314
- "0.1",
315
- "--dblock_local_aux_warn_ce",
316
- "0.0",
317
- "--dblock_local_aux_max_ce",
318
- "0.0",
319
- "--lr_schedule_origin_tokens",
320
- "85002387456",
321
- "--dblock_stop_after_commits",
322
- "0",
323
- "--no-oom_auto_backoff",
324
- "--require_exact_resume_state",
325
- "--dblock_objective_mode",
326
- "committed_9_1_1",
327
- "--dblock_ar_weight",
328
- "0.25",
329
- "--dblock_sat_weight",
330
- "0.25",
331
- "--dblock_nat_weight",
332
- "0.25",
333
- "--dblock_aux_grad_ratio",
334
- "0.5",
335
- "--dblock_min_ar_share",
336
- "0.8",
337
- "--lr_core",
338
- "2e-7",
339
- "--lr_head",
340
- "2e-7",
341
- "--dblock_checkpoint_stride",
342
- "2",
343
- "--empty_cache_every_steps",
344
- "0",
345
- "--dblock_log_every",
346
- "25",
347
- "--cuda_max_reserved_mib",
348
- "0",
349
- "--cuda_min_free_mib",
350
- "0",
351
- "--dblock_clean_context",
352
- "e2e",
353
- "--dblock_local_rows",
354
- "2",
355
- "--dblock_clean_context_noise_scale",
356
- "0.0",
357
- "--dblock_depth_anchor_every",
358
- "50",
359
- "--dblock_fullstack_anchor_cap",
360
- "256",
361
- "--dblock_fullstack_ar_rows",
362
- "1",
363
- "--detachable_50m_upgrade_multimode",
364
- "--detachable_50m_upgrade_logit_distill",
365
- "--detachable_50m_upgrade_sat_gate_v2713",
366
- "--detachable_50m_create_from_legacy",
367
- "--detachable_50m_seed_from_mature",
368
- "--detachable_50m_lr",
369
- "0.0001",
370
- "--detachable_50m_weight_decay",
371
- "0.0",
372
- "--detachable_50m_update_mode",
373
- "sublayer",
374
- "--detachable_50m_batch",
375
- "1",
376
- "--detachable_50m_tokens",
377
- "64",
378
- "--detachable_50m_loss_tokens",
379
- "64",
380
- "--detachable_50m_every",
381
- "1",
382
- "--detachable_50m_teacher_every",
383
- "1",
384
- "--detachable_50m_distill_weight",
385
- "1.0",
386
- "--detachable_50m_logit_distill_alpha",
387
- "0.30",
388
- "--detachable_50m_logit_distill_temperature",
389
- "1.0",
390
- "--detachable_50m_logit_distill_tokens",
391
- "64",
392
- "--detachable_50m_hidden_distill_weight",
393
- "0.0",
394
- "--detachable_50m_anchor_every",
395
- "128",
396
- "--detachable_50m_anchor_tokens",
397
- "32",
398
- "--detachable_50m_seed",
399
- "20260830",
400
- "--detachable_50m_log_every",
401
- "10",
402
- "--detachable_50m_telemetry_keep",
403
- "128",
404
- "--weight_accel_mode",
405
- "shadow",
406
- "--weight_accel_dir",
407
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel",
408
- "--weight_accel_every",
409
- "1",
410
- "--weight_accel_train_every",
411
- "4",
412
- "--weight_accel_record_every",
413
- "14",
414
- "--weight_accel_state_every",
415
- "256",
416
- "--detachable_50m",
417
- "--detachable_50m_optimizer",
418
- "paged_adamw8bit",
419
- "--delta_every_steps",
420
- "0",
421
- "--delta_every_sec",
422
- "0",
423
- "--resume",
424
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt",
425
- "--save_dir",
426
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B",
427
- "--data_seed",
428
- "2845332",
429
- "--batch_size",
430
- "56",
431
- "--dblock_blocks",
432
- "14",
433
- "--dblock_direct56_enabled",
434
- "1",
435
- "--dblock_layers_per_update",
436
- "0",
437
- "--dblock_router",
438
- "shadow",
439
- "--dblock_router_blend",
440
- "0.0",
441
- "--dblock_router_full_anchor_every",
442
- "128",
443
- "--dblock_router_train_layer_policy",
444
- "cyclic",
445
- "--dblock_router_train_layers_per_update",
446
- "1",
447
- "--dblock_sublayer_mode",
448
- "off",
449
- "--dblock_target_route_activate_from_checkpoint",
450
- "1",
451
- "--dblock_target_route_activation_checkpoint_step",
452
- "2717509",
453
- "--dblock_target_route_activation_parent_pointer_sha256",
454
- "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
455
- "--dblock_target_route_offset",
456
- "1",
457
- "--dblock_target_route_policy",
458
- "xor_fold56_v1",
459
- "--dblock_train_layer_policy",
460
- "cyclic",
461
- "--dblock_train_layers_per_update",
462
- "1",
463
- "--dblock_train_sublayer_policy",
464
- "cyclic",
465
- "--dblock_train_sublayers_per_update",
466
- "1",
467
- "--granularity_label",
468
- "one-sublayer-local-grad-direct56-teacher-v27.16",
469
- "--oom_known_safe_batch",
470
- "56",
471
- "--val_every_sec",
472
- "600",
473
- "--val_file",
474
- "/workspace/agillm43_heldout_v3/token_ids.json",
475
- "--val_sha256",
476
- "513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21"
477
- ],
478
- "argv_string": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
479
- "cwd": "/workspace/agillm44_throughput_80k_20260910",
480
- "env": {
481
- "AGILLM43_DBLOCK_CC_NOISE_SCALE": "0.0",
482
- "AGILLM43_DBLOCK_CLEAN_CONTEXT": "e2e",
483
- "AGILLM43_DBLOCK_CLEAN_CONTEXT_NOISE_SCALE": "0.0",
484
- "AGILLM43_DBLOCK_DEPTH_ANCHOR_EVERY": "50",
485
- "AGILLM43_DBLOCK_FULLSTACK_ANCHOR_CAP": "256",
486
- "AGILLM43_DBLOCK_FULLSTACK_AR_ROWS": "1",
487
- "AGILLM43_DBLOCK_LOCAL_ROWS": "2",
488
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATE_FROM_CHECKPOINT": "1",
489
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_CHECKPOINT_STEP": "2717509",
490
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_PARENT_POINTER_SHA256": "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
491
- "AGILLM43_DBLOCK_TARGET_ROUTE_OFFSET": "1",
492
- "AGILLM43_DBLOCK_TARGET_ROUTE_POLICY": "xor_fold56_v1",
493
- "AGILLM43_DECOMPRESS_CACHE": "0",
494
- "AGILLM43_DETACHABLE50M_ANCHOR_EVERY": "128",
495
- "AGILLM43_DETACHABLE50M_ANCHOR_TOKENS": "<redacted>",
496
- "AGILLM43_DETACHABLE50M_BATCH": "1",
497
- "AGILLM43_DETACHABLE50M_CREATE_FROM_LEGACY": "1",
498
- "AGILLM43_DETACHABLE50M_DISTILL_WEIGHT": "1.0",
499
- "AGILLM43_DETACHABLE50M_ENABLED": "1",
500
- "AGILLM43_DETACHABLE50M_EVERY": "1",
501
- "AGILLM43_DETACHABLE50M_GATE_V2713_START_CHECKPOINT_SHA256": "39f442fc0e6f9fad6f875e067f63db67f33a41a53e8cbc49ce1591010bb4aec3",
502
- "AGILLM43_DETACHABLE50M_HIDDEN_DISTILL_WEIGHT": "0.0",
503
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_ALPHA": "0.30",
504
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TEMPERATURE": "1.0",
505
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TOKENS": "<redacted>",
506
- "AGILLM43_DETACHABLE50M_LOG_EVERY": "10",
507
- "AGILLM43_DETACHABLE50M_LOSS_TOKENS": "<redacted>",
508
- "AGILLM43_DETACHABLE50M_LR": "0.0001",
509
- "AGILLM43_DETACHABLE50M_OPTIMIZER": "paged_adamw8bit",
510
- "AGILLM43_DETACHABLE50M_SEED": "20260830",
511
- "AGILLM43_DETACHABLE50M_SEED_FROM_MATURE": "1",
512
- "AGILLM43_DETACHABLE50M_TEACHER_EVERY": "1",
513
- "AGILLM43_DETACHABLE50M_TELEMETRY_KEEP": "128",
514
- "AGILLM43_DETACHABLE50M_TOKENS": "<redacted>",
515
- "AGILLM43_DETACHABLE50M_UPDATE_MODE": "sublayer",
516
- "AGILLM43_DETACHABLE50M_UPGRADE_LOGIT_DISTILL": "1",
517
- "AGILLM43_DETACHABLE50M_UPGRADE_MULTIMODE": "1",
518
- "AGILLM43_DETACHABLE50M_UPGRADE_SAT_GATE_V2713": "1",
519
- "AGILLM43_DETACHABLE50M_WEIGHT_DECAY": "0.0",
520
- "AGILLM43_SEQUENCE_PREFETCH_DEPTH": "256",
521
- "AGILLM43_SINGLEFILE_PROFILE": "intelligence-v25-satvar-pretrain",
522
- "AGILLM43_SUPERVISE_SET_ARGS_JSON": "{\"--batch_size\":\"56\",\"--dblock_blocks\":\"14\",\"--dblock_direct56_enabled\":\"1\",\"--dblock_layers_per_update\":\"0\",\"--dblock_router\":\"shadow\",\"--dblock_router_blend\":\"0.0\",\"--dblock_router_full_anchor_every\":\"128\",\"--dblock_router_train_layer_policy\":\"cyclic\",\"--dblock_router_train_layers_per_update\":\"1\",\"--dblock_sublayer_mode\":\"off\",\"--dblock_target_route_activate_from_checkpoint\":\"1\",\"--dblock_target_route_activation_checkpoint_step\":\"2717509\",\"--dblock_target_route_activation_parent_pointer_sha256\":\"04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827\",\"--dblock_target_route_offset\":\"1\",\"--dblock_target_route_policy\":\"xor_fold56_v1\",\"--dblock_train_layer_policy\":\"cyclic\",\"--dblock_train_layers_per_update\":\"1\",\"--dblock_train_sublayer_policy\":\"cyclic\",\"--dblock_train_sublayers_per_update\":\"1\",\"--granularity_label\":\"one-sublayer-local-grad-direct56-teacher-v27.16\",\"--oom_known_safe_batch\":\"56\",\"--val_every_sec\":\"600\",\"--val_file\":\"/workspace/agillm43_heldout_v3/token_ids.json\",\"--val_sha256\":\"513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21\"}",
523
- "AGILLM43_WEIGHT_ACCEL_MODE": "shadow",
524
- "AGILLM_DATASETS_STREAMING_READ_MAX_RETRIES": "2",
525
- "AGILLM_DATASETS_STREAMING_READ_RETRY_INTERVAL": "1",
526
- "AGILLM_DATASET_AGENT_ROUTER": "0",
527
- "AGILLM_DATASET_HOTLOAD_CONFIG": "/workspace/agillm43_dataset_hotload.json",
528
- "AGILLM_DATASET_HOT_RELOAD_CHECK_TOKENS": "<redacted>",
529
- "AGILLM_DATASET_HOT_RELOAD_SEC": "5",
530
- "AGILLM_DATASET_NN_ROUTER": "1",
531
- "AGILLM_DATASET_ROUTER_BLEND": "0.25",
532
- "AGILLM_DATASET_ROUTER_EXPLORE": "0.03",
533
- "AGILLM_DATASET_ROUTER_LR": "0.005",
534
- "AGILLM_DATASET_ROUTER_MIN_SCORE": "0.15",
535
- "AGILLM_DATASET_ROUTER_SHARPNESS": "2.0",
536
- "AGILLM_ENABLE_CUPY": "0",
537
- "AGILLM_HF_HUB_DOWNLOAD_TIMEOUT": "30",
538
- "AGILLM_HF_TRANSPORT": "geth_hf_relay",
539
- "AGILLM_HOT_CONFIG": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
540
- "AGILLM_HOT_CONFIG_SHA256": "",
541
- "AGILLM_MAX_EXAMPLE_CHARS": "32768",
542
- "AGILLM_MAX_EXAMPLE_TOKENS": "<redacted>",
543
- "AGILLM_PARENT_POINTER_SHA256": "9619410d2c62c3ec7ad43f6ba4d2a0300719df18226fc584592765aa774618db",
544
- "AGILLM_REMOTE_SHARDS_PER_ITER": "4",
545
- "AGILLM_RUN_ID": "singlefile_intelligence_v25_20260911T070604Z_from_step000002845290",
546
- "AGILLM_STREAM_NEXT_TIMEOUT_SEC": "60",
547
- "AGILLM_STREAM_OPEN_TIMEOUT_SEC": "120",
548
- "AGILLM_STREAM_SOCKET_TIMEOUT_SEC": "30",
549
- "AGILLM_STREAM_SOURCE_FATAL_COOLDOWN_SEC": "600",
550
- "AGILLM_STREAM_SOURCE_MAX_COOLDOWN_SEC": "120",
551
- "AGILLM_SYNTHETIC_TOKENIZER": "<redacted>",
552
- "CUDA_MODULE_LOADING": "LAZY",
553
- "CUDA_VERSION": "12.1.1",
554
- "CUDA_VISIBLE_DEVICES": "0",
555
- "HF_HUB_DOWNLOAD_TIMEOUT": "30",
556
- "HF_HUB_ETAG_TIMEOUT": "10",
557
- "PYTORCH_CUDA_ALLOC_CONF": "backend:cudaMallocAsync",
558
- "PYTORCH_VERSION": "2.2.1"
559
- },
560
- "parent_proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py supervise --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --log /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/logs/singlefile_intelligence_v25_satvar_pretrain.log --sleep_sec 15",
561
- "pid": 1042995,
562
- "ppid": 1042934,
563
- "proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
564
- "schema": "agillm.launch.v1"
565
- },
566
- "local_step": 2847537,
567
- "loss": 5.969868183135986,
568
- "nat_mask_contract": "agillm.nat-mask.v1",
569
- "nat_mask_contract_source": "explicit-legacy-recovery",
570
- "nat_mask_eos_id": 1,
571
- "nat_mask_legacy_blank_id": 1,
572
- "nat_mask_migrated_from_id": 1,
573
- "nat_mask_schema_version": 1,
574
- "nat_mask_token_id": 2,
575
- "pid": 1042995,
576
- "schema_version": 1,
577
- "seen_tok": 283463718912,
578
- "source_is_lane_checkpoint": true,
579
- "source_is_master_checkpoint": false,
580
- "source_is_non_master_checkpoint": false,
581
- "step": 2847537,
582
- "train_argv": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
583
- "train_script": "runtime_retain_activations_v2.py",
584
- "train_script_sha256": "2dbd1fa60fd5e5abd53e23930b34a0f20a2289a13b45a9f15b6de4a44b102afb",
585
- "warmstart_base_seen_tok": 0,
586
- "warmstart_base_step": 2845290,
587
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
588
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt",
589
- "warmstart_source_step": 2845290
590
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
checkpoints/step2847537_20260911/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_scaler.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c
3
- size 149
 
 
 
 
checkpoints/step2847537_20260911/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/scaler.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c
3
- size 149
 
 
 
 
checkpoints/step2847537_20260911/serving_candidate.json DELETED
@@ -1,694 +0,0 @@
1
- {
2
- "agillm43_provenance": {
3
- "batch_size": 56,
4
- "block_size": 2048,
5
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
6
- "checkpoint_type": "full",
7
- "created_at_iso": "2026-09-11T08:25:29Z",
8
- "created_at_unix": 1789115129.1128435,
9
- "created_from_scratch": false,
10
- "dataset_provenance": {
11
- "chat": false,
12
- "dataset_field_text": "text",
13
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
14
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
15
- "dataset_hotload_config_stamp": [
16
- 50,
17
- 1124990893,
18
- 1789091829464680808,
19
- 7210
20
- ],
21
- "dataset_hotload_config_used": true,
22
- "dataset_hotload_enabled": true,
23
- "dataset_hotload_generation": 13,
24
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
25
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
26
- "dataset_revisions": {
27
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
28
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
29
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
30
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
31
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
32
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
33
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
34
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
35
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
36
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
37
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
38
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
39
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
40
- },
41
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
42
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
43
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
44
- "has_language_mix": true,
45
- "has_numeracy": false,
46
- "hot_config_expected_sha256": "",
47
- "hot_config_has_append": true,
48
- "hot_config_has_datasets": false,
49
- "hot_config_mtime": 1789092253.2119184,
50
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
51
- "hot_config_used": true,
52
- "hot_reload_capable": true,
53
- "phase": "pretrain",
54
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
55
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
56
- "schema": "agillm.dataset_provenance.v1",
57
- "sequence_prefetch_depth": 256,
58
- "source_count": 23,
59
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
60
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
61
- "source_specs": [
62
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
63
- "wikimedia/wikipedia:20231101.en|1.63",
64
- "allenai/c4:en|0.36",
65
- "Skylion007/openwebtext|0.18",
66
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
67
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
68
- "EleutherAI/proof-pile-2:all|0.67",
69
- "allenai/dolma:v1_6-sample|0.5",
70
- "codeparrot/codeparrot-clean|0.24",
71
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
72
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
73
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
74
- "allenai/dolmino-mix-1124|0.78",
75
- "mlfoundations/dclm-baseline-1.0|1.05",
76
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
77
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
78
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
79
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
80
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
81
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
82
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
83
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
84
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
85
- ],
86
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
87
- "val_source_requested": ""
88
- },
89
- "effective_global_step": 5692827,
90
- "effective_seen_tok": 283463718912,
91
- "global_origin_seen_tok": 0,
92
- "global_origin_step": 2845290,
93
- "gpu": {
94
- "allocated_gb": 7.13,
95
- "peak_allocated_gb": 12.0,
96
- "reserved_gb": 13.34
97
- },
98
- "hostname": "ead6505f92c4",
99
- "lane": "pretrain",
100
- "launch": {
101
- "argv": [
102
- "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py",
103
- "train",
104
- "--preset",
105
- "agillm4_floor",
106
- "--block",
107
- "2048",
108
- "--source",
109
- "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
110
- "--nat_mask_token_id",
111
- "2",
112
- "--tie_weights",
113
- "--tie_kv",
114
- "--moe_ffn",
115
- "--moe_experts",
116
- "2",
117
- "--moe_top_k",
118
- "1",
119
- "--moe_mlp_mult",
120
- "4",
121
- "--moe_shared_experts",
122
- "1",
123
- "--moe_shared_mlp_mult",
124
- "2",
125
- "--moe_aux_coef",
126
- "0.01",
127
- "--moe_z_coef",
128
- "0.001",
129
- "--amp",
130
- "--grad_checkpoint",
131
- "--attn_backend",
132
- "sdpa",
133
- "--sublinear_window",
134
- "128",
135
- "--sublinear_stride",
136
- "128",
137
- "--sublinear_max_anchors",
138
- "128",
139
- "--sublinear_chunk",
140
- "128",
141
- "--sublinear_sinks",
142
- "4",
143
- "--sublinear_recent_anchors",
144
- "64",
145
- "--no-sublinear_pooled_landmarks",
146
- "--anchor_stride",
147
- "256",
148
- "--anchor_max",
149
- "2048",
150
- "--anchor_position",
151
- "-1",
152
- "--alibi_mode",
153
- "corrected",
154
- "--alibi_scale",
155
- "0.0",
156
- "--optimizer",
157
- "adamw8bit",
158
- "--weight_decay",
159
- "0.0",
160
- "--lr_decay",
161
- "cosine",
162
- "--lr_decay_tokens",
163
- "49000954880",
164
- "--lr_min_mult",
165
- "0.2",
166
- "--lr_schedule_reset_on_resume",
167
- "--lr_warmup_tokens",
168
- "10000000",
169
- "--lr_warmup_min_mult",
170
- "0.1",
171
- "--dblock",
172
- "--dblock_schedule",
173
- "roundrobin",
174
- "--dblock_router_ramp_steps",
175
- "64",
176
- "--dblock_warmup_steps",
177
- "14",
178
- "--dblock_checkpoint_skip_tail",
179
- "0",
180
- "--dblock_loop_layers",
181
- "0",
182
- "--dblock_loop_start",
183
- "0",
184
- "--dblock_loop_cond_scale",
185
- "1.0",
186
- "--dblock_sigma_curriculum_steps",
187
- "2000",
188
- "--dblock_sigma_sampling",
189
- "lognormal",
190
- "--dblock_sigma_stratified",
191
- "--dblock_sigma_min",
192
- "0.002",
193
- "--dblock_sigma_max",
194
- "2.0",
195
- "--dblock_sigma_pmean",
196
- "-1.2",
197
- "--dblock_sigma_pstd",
198
- "1.2",
199
- "--dblock_edm_wmax",
200
- "5.0",
201
- "--dblock_ar_prob",
202
- "0.8",
203
- "--dblock_sat_prob",
204
- "0.1",
205
- "--dblock_nat_prob",
206
- "0.1",
207
- "--dblock_ar_loss_tokens",
208
- "8192",
209
- "--dblock_sat_loss_tokens",
210
- "4096",
211
- "--dblock_nat_loss_tokens",
212
- "4096",
213
- "--dblock_nat_embed_noise_mode",
214
- "off",
215
- "--dblock_nat_embed_noise_scale",
216
- "1.0",
217
- "--nat_loss_weight",
218
- "1.0",
219
- "--nat_mask_ratio",
220
- "0.5",
221
- "--nat_max_tokens",
222
- "0",
223
- "--nat_span_mask_prob",
224
- "0.35",
225
- "--nat_suffix_mask_prob",
226
- "0.20",
227
- "--nat_span_max_tokens",
228
- "0",
229
- "--loss_spike_skip",
230
- "0.0",
231
- "--sat_every",
232
- "1",
233
- "--nat_every",
234
- "1",
235
- "--dblock_fullstack_anchor_spike_skip",
236
- "0.0",
237
- "--dblock_fullstack_anchor_max_ce",
238
- "0.0",
239
- "--dblock_anchor_softcap_mult",
240
- "4.0",
241
- "--dblock_anchor_softcap_min_ce",
242
- "25.0",
243
- "--dblock_anchor_softcap_max_ce",
244
- "50.0",
245
- "--dblock_fullstack_ar_every",
246
- "3",
247
- "--dblock_fullstack_ar_offset",
248
- "0",
249
- "--dblock_fullstack_ar_tokens",
250
- "256",
251
- "--dblock_fullstack_sat_every",
252
- "3",
253
- "--dblock_fullstack_sat_offset",
254
- "1",
255
- "--dblock_fullstack_sat_tokens",
256
- "128",
257
- "--dblock_fullstack_nat_every",
258
- "3",
259
- "--dblock_fullstack_nat_offset",
260
- "2",
261
- "--dblock_fullstack_nat_tokens",
262
- "128",
263
- "--dblock_fullstack_nat_mask_id",
264
- "-1",
265
- "--ckpt_codec",
266
- "block-sharded-zstd",
267
- "--async_update_every_steps",
268
- "0",
269
- "--dblock_router_hidden",
270
- "64",
271
- "--dblock_router_heads",
272
- "4",
273
- "--dblock_router_layers",
274
- "2",
275
- "--dblock_router_lr",
276
- "0.0005",
277
- "--dblock_explore",
278
- "0.08",
279
- "--dblock_max_stale_steps",
280
- "64",
281
- "--dblock_max_count_skew",
282
- "1.35",
283
- "--dblock_stale_bonus",
284
- "0.35",
285
- "--dblock_undertrain_bonus",
286
- "0.25",
287
- "--dblock_activation_offload_min_mb",
288
- "1.0",
289
- "--nat_document_boundary_aware",
290
- "--dblock_fullstack_anchor_deterministic_eval",
291
- "--dblock_fullstack_anchor_spike_retry_limit",
292
- "3",
293
- "--dblock_local_spike_retry_limit",
294
- "3",
295
- "--dblock_fullstack_nat_no_valid_crop_limit",
296
- "8",
297
- "--target_tokens",
298
- "400000000000",
299
- "--lr_schedule_reanchor_on_resume",
300
- "--save_every_sec",
301
- "1800",
302
- "--delta_max_keep",
303
- "0",
304
- "--max_ckpts",
305
- "2",
306
- "--disk_free_floor_gb",
307
- "12.0",
308
- "--heartbeat_every_sec",
309
- "30",
310
- "--dblock_fullstack_ar_weight",
311
- "1.0",
312
- "--dblock_fullstack_sat_weight",
313
- "0.1",
314
- "--dblock_fullstack_nat_weight",
315
- "0.1",
316
- "--dblock_local_aux_warn_ce",
317
- "0.0",
318
- "--dblock_local_aux_max_ce",
319
- "0.0",
320
- "--lr_schedule_origin_tokens",
321
- "85002387456",
322
- "--dblock_stop_after_commits",
323
- "0",
324
- "--no-oom_auto_backoff",
325
- "--require_exact_resume_state",
326
- "--dblock_objective_mode",
327
- "committed_9_1_1",
328
- "--dblock_ar_weight",
329
- "0.25",
330
- "--dblock_sat_weight",
331
- "0.25",
332
- "--dblock_nat_weight",
333
- "0.25",
334
- "--dblock_aux_grad_ratio",
335
- "0.5",
336
- "--dblock_min_ar_share",
337
- "0.8",
338
- "--lr_core",
339
- "2e-7",
340
- "--lr_head",
341
- "2e-7",
342
- "--dblock_checkpoint_stride",
343
- "2",
344
- "--empty_cache_every_steps",
345
- "0",
346
- "--dblock_log_every",
347
- "25",
348
- "--cuda_max_reserved_mib",
349
- "0",
350
- "--cuda_min_free_mib",
351
- "0",
352
- "--dblock_clean_context",
353
- "e2e",
354
- "--dblock_local_rows",
355
- "2",
356
- "--dblock_clean_context_noise_scale",
357
- "0.0",
358
- "--dblock_depth_anchor_every",
359
- "50",
360
- "--dblock_fullstack_anchor_cap",
361
- "256",
362
- "--dblock_fullstack_ar_rows",
363
- "1",
364
- "--detachable_50m_upgrade_multimode",
365
- "--detachable_50m_upgrade_logit_distill",
366
- "--detachable_50m_upgrade_sat_gate_v2713",
367
- "--detachable_50m_create_from_legacy",
368
- "--detachable_50m_seed_from_mature",
369
- "--detachable_50m_lr",
370
- "0.0001",
371
- "--detachable_50m_weight_decay",
372
- "0.0",
373
- "--detachable_50m_update_mode",
374
- "sublayer",
375
- "--detachable_50m_batch",
376
- "1",
377
- "--detachable_50m_tokens",
378
- "64",
379
- "--detachable_50m_loss_tokens",
380
- "64",
381
- "--detachable_50m_every",
382
- "1",
383
- "--detachable_50m_teacher_every",
384
- "1",
385
- "--detachable_50m_distill_weight",
386
- "1.0",
387
- "--detachable_50m_logit_distill_alpha",
388
- "0.30",
389
- "--detachable_50m_logit_distill_temperature",
390
- "1.0",
391
- "--detachable_50m_logit_distill_tokens",
392
- "64",
393
- "--detachable_50m_hidden_distill_weight",
394
- "0.0",
395
- "--detachable_50m_anchor_every",
396
- "128",
397
- "--detachable_50m_anchor_tokens",
398
- "32",
399
- "--detachable_50m_seed",
400
- "20260830",
401
- "--detachable_50m_log_every",
402
- "10",
403
- "--detachable_50m_telemetry_keep",
404
- "128",
405
- "--weight_accel_mode",
406
- "shadow",
407
- "--weight_accel_dir",
408
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel",
409
- "--weight_accel_every",
410
- "1",
411
- "--weight_accel_train_every",
412
- "4",
413
- "--weight_accel_record_every",
414
- "14",
415
- "--weight_accel_state_every",
416
- "256",
417
- "--detachable_50m",
418
- "--detachable_50m_optimizer",
419
- "paged_adamw8bit",
420
- "--delta_every_steps",
421
- "0",
422
- "--delta_every_sec",
423
- "0",
424
- "--resume",
425
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt",
426
- "--save_dir",
427
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B",
428
- "--data_seed",
429
- "2845332",
430
- "--batch_size",
431
- "56",
432
- "--dblock_blocks",
433
- "14",
434
- "--dblock_direct56_enabled",
435
- "1",
436
- "--dblock_layers_per_update",
437
- "0",
438
- "--dblock_router",
439
- "shadow",
440
- "--dblock_router_blend",
441
- "0.0",
442
- "--dblock_router_full_anchor_every",
443
- "128",
444
- "--dblock_router_train_layer_policy",
445
- "cyclic",
446
- "--dblock_router_train_layers_per_update",
447
- "1",
448
- "--dblock_sublayer_mode",
449
- "off",
450
- "--dblock_target_route_activate_from_checkpoint",
451
- "1",
452
- "--dblock_target_route_activation_checkpoint_step",
453
- "2717509",
454
- "--dblock_target_route_activation_parent_pointer_sha256",
455
- "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
456
- "--dblock_target_route_offset",
457
- "1",
458
- "--dblock_target_route_policy",
459
- "xor_fold56_v1",
460
- "--dblock_train_layer_policy",
461
- "cyclic",
462
- "--dblock_train_layers_per_update",
463
- "1",
464
- "--dblock_train_sublayer_policy",
465
- "cyclic",
466
- "--dblock_train_sublayers_per_update",
467
- "1",
468
- "--granularity_label",
469
- "one-sublayer-local-grad-direct56-teacher-v27.16",
470
- "--oom_known_safe_batch",
471
- "56",
472
- "--val_every_sec",
473
- "600",
474
- "--val_file",
475
- "/workspace/agillm43_heldout_v3/token_ids.json",
476
- "--val_sha256",
477
- "513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21"
478
- ],
479
- "argv_string": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
480
- "cwd": "/workspace/agillm44_throughput_80k_20260910",
481
- "env": {
482
- "AGILLM43_DBLOCK_CC_NOISE_SCALE": "0.0",
483
- "AGILLM43_DBLOCK_CLEAN_CONTEXT": "e2e",
484
- "AGILLM43_DBLOCK_CLEAN_CONTEXT_NOISE_SCALE": "0.0",
485
- "AGILLM43_DBLOCK_DEPTH_ANCHOR_EVERY": "50",
486
- "AGILLM43_DBLOCK_FULLSTACK_ANCHOR_CAP": "256",
487
- "AGILLM43_DBLOCK_FULLSTACK_AR_ROWS": "1",
488
- "AGILLM43_DBLOCK_LOCAL_ROWS": "2",
489
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATE_FROM_CHECKPOINT": "1",
490
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_CHECKPOINT_STEP": "2717509",
491
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_PARENT_POINTER_SHA256": "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
492
- "AGILLM43_DBLOCK_TARGET_ROUTE_OFFSET": "1",
493
- "AGILLM43_DBLOCK_TARGET_ROUTE_POLICY": "xor_fold56_v1",
494
- "AGILLM43_DECOMPRESS_CACHE": "0",
495
- "AGILLM43_DETACHABLE50M_ANCHOR_EVERY": "128",
496
- "AGILLM43_DETACHABLE50M_ANCHOR_TOKENS": "<redacted>",
497
- "AGILLM43_DETACHABLE50M_BATCH": "1",
498
- "AGILLM43_DETACHABLE50M_CREATE_FROM_LEGACY": "1",
499
- "AGILLM43_DETACHABLE50M_DISTILL_WEIGHT": "1.0",
500
- "AGILLM43_DETACHABLE50M_ENABLED": "1",
501
- "AGILLM43_DETACHABLE50M_EVERY": "1",
502
- "AGILLM43_DETACHABLE50M_GATE_V2713_START_CHECKPOINT_SHA256": "39f442fc0e6f9fad6f875e067f63db67f33a41a53e8cbc49ce1591010bb4aec3",
503
- "AGILLM43_DETACHABLE50M_HIDDEN_DISTILL_WEIGHT": "0.0",
504
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_ALPHA": "0.30",
505
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TEMPERATURE": "1.0",
506
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TOKENS": "<redacted>",
507
- "AGILLM43_DETACHABLE50M_LOG_EVERY": "10",
508
- "AGILLM43_DETACHABLE50M_LOSS_TOKENS": "<redacted>",
509
- "AGILLM43_DETACHABLE50M_LR": "0.0001",
510
- "AGILLM43_DETACHABLE50M_OPTIMIZER": "paged_adamw8bit",
511
- "AGILLM43_DETACHABLE50M_SEED": "20260830",
512
- "AGILLM43_DETACHABLE50M_SEED_FROM_MATURE": "1",
513
- "AGILLM43_DETACHABLE50M_TEACHER_EVERY": "1",
514
- "AGILLM43_DETACHABLE50M_TELEMETRY_KEEP": "128",
515
- "AGILLM43_DETACHABLE50M_TOKENS": "<redacted>",
516
- "AGILLM43_DETACHABLE50M_UPDATE_MODE": "sublayer",
517
- "AGILLM43_DETACHABLE50M_UPGRADE_LOGIT_DISTILL": "1",
518
- "AGILLM43_DETACHABLE50M_UPGRADE_MULTIMODE": "1",
519
- "AGILLM43_DETACHABLE50M_UPGRADE_SAT_GATE_V2713": "1",
520
- "AGILLM43_DETACHABLE50M_WEIGHT_DECAY": "0.0",
521
- "AGILLM43_SEQUENCE_PREFETCH_DEPTH": "256",
522
- "AGILLM43_SINGLEFILE_PROFILE": "intelligence-v25-satvar-pretrain",
523
- "AGILLM43_SUPERVISE_SET_ARGS_JSON": "{\"--batch_size\":\"56\",\"--dblock_blocks\":\"14\",\"--dblock_direct56_enabled\":\"1\",\"--dblock_layers_per_update\":\"0\",\"--dblock_router\":\"shadow\",\"--dblock_router_blend\":\"0.0\",\"--dblock_router_full_anchor_every\":\"128\",\"--dblock_router_train_layer_policy\":\"cyclic\",\"--dblock_router_train_layers_per_update\":\"1\",\"--dblock_sublayer_mode\":\"off\",\"--dblock_target_route_activate_from_checkpoint\":\"1\",\"--dblock_target_route_activation_checkpoint_step\":\"2717509\",\"--dblock_target_route_activation_parent_pointer_sha256\":\"04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827\",\"--dblock_target_route_offset\":\"1\",\"--dblock_target_route_policy\":\"xor_fold56_v1\",\"--dblock_train_layer_policy\":\"cyclic\",\"--dblock_train_layers_per_update\":\"1\",\"--dblock_train_sublayer_policy\":\"cyclic\",\"--dblock_train_sublayers_per_update\":\"1\",\"--granularity_label\":\"one-sublayer-local-grad-direct56-teacher-v27.16\",\"--oom_known_safe_batch\":\"56\",\"--val_every_sec\":\"600\",\"--val_file\":\"/workspace/agillm43_heldout_v3/token_ids.json\",\"--val_sha256\":\"513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21\"}",
524
- "AGILLM43_WEIGHT_ACCEL_MODE": "shadow",
525
- "AGILLM_DATASETS_STREAMING_READ_MAX_RETRIES": "2",
526
- "AGILLM_DATASETS_STREAMING_READ_RETRY_INTERVAL": "1",
527
- "AGILLM_DATASET_AGENT_ROUTER": "0",
528
- "AGILLM_DATASET_HOTLOAD_CONFIG": "/workspace/agillm43_dataset_hotload.json",
529
- "AGILLM_DATASET_HOT_RELOAD_CHECK_TOKENS": "<redacted>",
530
- "AGILLM_DATASET_HOT_RELOAD_SEC": "5",
531
- "AGILLM_DATASET_NN_ROUTER": "1",
532
- "AGILLM_DATASET_ROUTER_BLEND": "0.25",
533
- "AGILLM_DATASET_ROUTER_EXPLORE": "0.03",
534
- "AGILLM_DATASET_ROUTER_LR": "0.005",
535
- "AGILLM_DATASET_ROUTER_MIN_SCORE": "0.15",
536
- "AGILLM_DATASET_ROUTER_SHARPNESS": "2.0",
537
- "AGILLM_ENABLE_CUPY": "0",
538
- "AGILLM_HF_HUB_DOWNLOAD_TIMEOUT": "30",
539
- "AGILLM_HF_TRANSPORT": "geth_hf_relay",
540
- "AGILLM_HOT_CONFIG": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
541
- "AGILLM_HOT_CONFIG_SHA256": "",
542
- "AGILLM_MAX_EXAMPLE_CHARS": "32768",
543
- "AGILLM_MAX_EXAMPLE_TOKENS": "<redacted>",
544
- "AGILLM_PARENT_POINTER_SHA256": "9619410d2c62c3ec7ad43f6ba4d2a0300719df18226fc584592765aa774618db",
545
- "AGILLM_REMOTE_SHARDS_PER_ITER": "4",
546
- "AGILLM_RUN_ID": "singlefile_intelligence_v25_20260911T070604Z_from_step000002845290",
547
- "AGILLM_STREAM_NEXT_TIMEOUT_SEC": "60",
548
- "AGILLM_STREAM_OPEN_TIMEOUT_SEC": "120",
549
- "AGILLM_STREAM_SOCKET_TIMEOUT_SEC": "30",
550
- "AGILLM_STREAM_SOURCE_FATAL_COOLDOWN_SEC": "600",
551
- "AGILLM_STREAM_SOURCE_MAX_COOLDOWN_SEC": "120",
552
- "AGILLM_SYNTHETIC_TOKENIZER": "<redacted>",
553
- "CUDA_MODULE_LOADING": "LAZY",
554
- "CUDA_VERSION": "12.1.1",
555
- "CUDA_VISIBLE_DEVICES": "0",
556
- "HF_HUB_DOWNLOAD_TIMEOUT": "30",
557
- "HF_HUB_ETAG_TIMEOUT": "10",
558
- "PYTORCH_CUDA_ALLOC_CONF": "backend:cudaMallocAsync",
559
- "PYTORCH_VERSION": "2.2.1"
560
- },
561
- "parent_proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py supervise --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --log /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/logs/singlefile_intelligence_v25_satvar_pretrain.log --sleep_sec 15",
562
- "pid": 1042995,
563
- "ppid": 1042934,
564
- "proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
565
- "schema": "agillm.launch.v1"
566
- },
567
- "local_step": 2847537,
568
- "loss": 5.969868183135986,
569
- "nat_mask_contract": "agillm.nat-mask.v1",
570
- "nat_mask_contract_source": "explicit-legacy-recovery",
571
- "nat_mask_eos_id": 1,
572
- "nat_mask_legacy_blank_id": 1,
573
- "nat_mask_migrated_from_id": 1,
574
- "nat_mask_schema_version": 1,
575
- "nat_mask_token_id": 2,
576
- "pid": 1042995,
577
- "schema_version": 1,
578
- "seen_tok": 283463718912,
579
- "source_is_lane_checkpoint": true,
580
- "source_is_master_checkpoint": false,
581
- "source_is_non_master_checkpoint": false,
582
- "step": 2847537,
583
- "train_argv": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2845332 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
584
- "train_script": "runtime_retain_activations_v2.py",
585
- "train_script_sha256": "2dbd1fa60fd5e5abd53e23930b34a0f20a2289a13b45a9f15b6de4a44b102afb",
586
- "warmstart_base_seen_tok": 0,
587
- "warmstart_base_step": 2845290,
588
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
589
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt",
590
- "warmstart_source_step": 2845290
591
- },
592
- "alibi_mode": "corrected",
593
- "alibi_scale": 0.0,
594
- "checkpoint_summary": "warmstarted_from_lane_checkpoint; source=/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt; path=/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
595
- "dataset_provenance": {
596
- "chat": false,
597
- "dataset_field_text": "text",
598
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
599
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
600
- "dataset_hotload_config_stamp": [
601
- 50,
602
- 1124990893,
603
- 1789091829464680808,
604
- 7210
605
- ],
606
- "dataset_hotload_config_used": true,
607
- "dataset_hotload_enabled": true,
608
- "dataset_hotload_generation": 13,
609
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
610
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
611
- "dataset_revisions": {
612
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
613
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
614
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
615
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
616
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
617
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
618
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
619
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
620
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
621
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
622
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
623
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
624
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
625
- },
626
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
627
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
628
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
629
- "has_language_mix": true,
630
- "has_numeracy": false,
631
- "hot_config_expected_sha256": "",
632
- "hot_config_has_append": true,
633
- "hot_config_has_datasets": false,
634
- "hot_config_mtime": 1789092253.2119184,
635
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
636
- "hot_config_used": true,
637
- "hot_reload_capable": true,
638
- "phase": "pretrain",
639
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
640
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
641
- "schema": "agillm.dataset_provenance.v1",
642
- "sequence_prefetch_depth": 256,
643
- "source_count": 23,
644
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
645
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
646
- "source_specs": [
647
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
648
- "wikimedia/wikipedia:20231101.en|1.63",
649
- "allenai/c4:en|0.36",
650
- "Skylion007/openwebtext|0.18",
651
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
652
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
653
- "EleutherAI/proof-pile-2:all|0.67",
654
- "allenai/dolma:v1_6-sample|0.5",
655
- "codeparrot/codeparrot-clean|0.24",
656
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
657
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
658
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
659
- "allenai/dolmino-mix-1124|0.78",
660
- "mlfoundations/dclm-baseline-1.0|1.05",
661
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
662
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
663
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
664
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
665
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
666
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
667
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
668
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
669
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
670
- ],
671
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
672
- "val_source_requested": ""
673
- },
674
- "lr_schedule_origin_tok": 85002387456,
675
- "nat_mask_contract": "agillm.nat-mask.v1",
676
- "nat_mask_contract_source": "explicit-legacy-recovery",
677
- "nat_mask_eos_id": 1,
678
- "nat_mask_legacy_blank_id": 1,
679
- "nat_mask_migrated_from_id": 1,
680
- "nat_mask_schema_version": 1,
681
- "nat_mask_token_id": 2,
682
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
683
- "promotion_required": true,
684
- "quality_gate_used_for_training": false,
685
- "repair_schema": "",
686
- "saved_at": "2026-09-11T08:26:27Z",
687
- "schema": "agillm43.serving.candidate.v1",
688
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
689
- "step": 2847537,
690
- "training_pointer_schema": "agillm43.training.resume.v2",
691
- "training_pointer_sha256": "3af634bfe6a879b6f06701dd01c8c51cf9c4dbaf4bdc1619fa662f57a4c0e41e",
692
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
693
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02845290_from02842785_20260911T063539_1789108539467944706_706357_000001.pt"
694
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
checkpoints/step2847537_20260911/training_latest.json DELETED
@@ -1,455 +0,0 @@
1
- {
2
- "checkpoint_main_nbytes": 18496394,
3
- "checkpoint_main_sha256": "0f8047e91a79072538e657e726ced888a7c2fc80fcec5bf35550bc8d4102bd18",
4
- "checkpoint_name": "pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
5
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
6
- "continuation_state_schema": "agillm43.continuation.state.v2",
7
- "continuation_state_sha256": "c3c8e0309210d7ef7c24d2d49c8c43daaedc3a796283246df885559a88dc2fe1",
8
- "dblock_committed_step": 1117186,
9
- "dblock_target_route_clock": 130028,
10
- "dblock_target_route_controller_source_sha256": "714ee71c440ba9e9699034d7ea5f1de6f73d65cc80caf9f9350cf30f11a33e28",
11
- "dblock_target_route_last_committed_target": {
12
- "attempts": 1,
13
- "block": 7,
14
- "canonical_index": 52,
15
- "committed_step_after": 1117186,
16
- "first_attempt_step": 130027,
17
- "k": 10,
18
- "last_attempt_step": 130027,
19
- "layer_offset": 0,
20
- "phase": 2,
21
- "physical_layer": 14,
22
- "q": 3,
23
- "route_commit_clock": 130027,
24
- "route_offset": 1,
25
- "selection": {
26
- "authority": "learned",
27
- "authority_granted": true,
28
- "baseline_prediction": 6.006411273581128,
29
- "candidate_hash": "7366174b2e1fbeb670854a846f1be186bf4933755de5e3bcaf7aa6b0bd537baa",
30
- "candidate_persisted": true,
31
- "clock": 130027,
32
- "fault_count": 0,
33
- "guard_override": false,
34
- "guard_reason": null,
35
- "nominated": true,
36
- "override": false,
37
- "phase": 52,
38
- "raw_prediction": 30,
39
- "reason": "candidate_scheduled",
40
- "schema": "agillm44.direct56.controller.v1",
41
- "selected": 30,
42
- "selected_target": 30,
43
- "shadow_override": false,
44
- "stage": "learned_full",
45
- "teacher_ce": 0.0001746263587847352,
46
- "teacher_target": 30,
47
- "teacher_training_receipts": 49331,
48
- "teacher_value_prediction": 6.0404791831970215,
49
- "validation_samples": 1227,
50
- "value_prediction": 6.0404791831970215
51
- },
52
- "sublayer": "ffn",
53
- "target_id": 30,
54
- "target_sha256": "715ed54507c97086031c68d9c54c0b0f6cfc1b8b1c1668bb53c7a2637dda11c5",
55
- "teacher_target_id": 30
56
- },
57
- "dblock_target_route_offset": 1,
58
- "dblock_target_route_payload_sha256": "1dc889e4667f009a560603a107be45a6b90324a86f999dadb09de6a0ffd73b3c",
59
- "dblock_target_route_pending_target": null,
60
- "dblock_target_route_policy_id": "guarded-direct56-route-equiv56-v2",
61
- "dblock_target_route_policy_sha256": "d4241a9c3a0c319cab6b268ea3792277c2cf7088b8d11492e459d2e58e2bde47",
62
- "dblock_target_route_schema": "agillm43.dblock.target-route.direct56.v1",
63
- "dblock_target_route_target_map_sha256": "31e64396588ebf72dd6aa20de6cbea14c20b6b0823269fade6f7471d37cc0836",
64
- "dblock_target_route_topology_sha256": "c3b1679046220655820501bb30dc2733efc0628f661273ac3f2217355bdc2102",
65
- "direct56_checkpointed": true,
66
- "direct56_payload_digest": "213d97b9956404a356c20e28ea5bb85714c04874cb5e592200aa45d6b7c62bfb",
67
- "direct56_schema": "agillm44.direct56.controller.v1",
68
- "package_format": "block-sharded",
69
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt",
70
- "quality_gate_used_for_training": false,
71
- "record_sha256": "3af634bfe6a879b6f06701dd01c8c51cf9c4dbaf4bdc1619fa662f57a4c0e41e",
72
- "role": "training",
73
- "saved_at": "2026-09-11T08:26:27Z",
74
- "schema": "agillm43.training.resume.v2",
75
- "seen_tok": 283463718912,
76
- "shard_dir": "pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards",
77
- "shards": [
78
- {
79
- "name": "core_base.pt",
80
- "nbytes": 608660242,
81
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_base.pt",
82
- "sha256": "f9aa1a90ccf14d5b9ae92ef5711d0083f4e6a2bd7cd09954bef9b186192dd37b"
83
- },
84
- {
85
- "name": "core_block_0000.pt",
86
- "nbytes": 139938848,
87
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0000.pt",
88
- "sha256": "99fab444fd4b2a2c9e96c89f8422e3f72fd8c0022d0e1fcce93a524256224848"
89
- },
90
- {
91
- "name": "core_block_0001.pt",
92
- "nbytes": 140209523,
93
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0001.pt",
94
- "sha256": "62595fa411827eb506cab64743ec7b19213e160560fdba232d1fd84a8800f6fb"
95
- },
96
- {
97
- "name": "core_block_0002.pt",
98
- "nbytes": 140184224,
99
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0002.pt",
100
- "sha256": "fe5f3b12358feb095de03a8039f4566b4fa6fa0e26d3dd905214a0e593174716"
101
- },
102
- {
103
- "name": "core_block_0003.pt",
104
- "nbytes": 140171398,
105
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0003.pt",
106
- "sha256": "5a29384a1e30dc49a513b1a7797e131d51bc46810b05e1d8ebc881d644fac1d9"
107
- },
108
- {
109
- "name": "core_block_0004.pt",
110
- "nbytes": 140327349,
111
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0004.pt",
112
- "sha256": "9b72403a74f469917d3436523baff2b6c00e28d07a731e552a16a2bffb6d36c6"
113
- },
114
- {
115
- "name": "core_block_0005.pt",
116
- "nbytes": 140270184,
117
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0005.pt",
118
- "sha256": "f679b7dd6581b83cf1c340ad28c498bde82b691eafcb4a81358684ada9cbda17"
119
- },
120
- {
121
- "name": "core_block_0006.pt",
122
- "nbytes": 140627412,
123
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0006.pt",
124
- "sha256": "a190fa793ae38f8f5d7d839774ea5f6317604bfeeb07c5882c853f082ebf423a"
125
- },
126
- {
127
- "name": "core_block_0007.pt",
128
- "nbytes": 140777683,
129
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0007.pt",
130
- "sha256": "b12d553cd77177070b90193140741ebcafcd5a960551f58f8b0b3a4bfe07a5b3"
131
- },
132
- {
133
- "name": "core_block_0008.pt",
134
- "nbytes": 140659532,
135
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0008.pt",
136
- "sha256": "d1be3f2980c7499250753929baa178e42bc8ce7f349c653101845644c3458101"
137
- },
138
- {
139
- "name": "core_block_0009.pt",
140
- "nbytes": 140485548,
141
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0009.pt",
142
- "sha256": "807135fc8e97216df5af208438acdf69781f4eabf123aec78762c2537d0e32a3"
143
- },
144
- {
145
- "name": "core_block_0010.pt",
146
- "nbytes": 140551029,
147
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0010.pt",
148
- "sha256": "c95740ccadc6de5249f4ac6cebf3bd21acb262af70d36ea2ebcc3373f67f23d6"
149
- },
150
- {
151
- "name": "core_block_0011.pt",
152
- "nbytes": 140591018,
153
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0011.pt",
154
- "sha256": "cd7ff43d4fecec485865da27700b8369993894e9cdc2980c220796e23efb142c"
155
- },
156
- {
157
- "name": "core_block_0012.pt",
158
- "nbytes": 140610366,
159
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0012.pt",
160
- "sha256": "ff6d1d7dea641a405fcff8b89b400e8e9742b868858a6cd291df2ae0b754eb92"
161
- },
162
- {
163
- "name": "core_block_0013.pt",
164
- "nbytes": 140407299,
165
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0013.pt",
166
- "sha256": "03c520745fe32f6d5c1cd4bb09a8fb2af5b77d03cbad1efc70077e9c6c7eea92"
167
- },
168
- {
169
- "name": "core_block_0014.pt",
170
- "nbytes": 140501663,
171
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0014.pt",
172
- "sha256": "3749c1ba95b083ea9eaf9a1c156f6e11a794fa4cca61b4e945b0c0a4ba814750"
173
- },
174
- {
175
- "name": "core_block_0015.pt",
176
- "nbytes": 140220223,
177
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0015.pt",
178
- "sha256": "b8dc7be738a3ae3745092b9c362edd359e8354e37ea82c10aeaaa13f681901df"
179
- },
180
- {
181
- "name": "core_block_0016.pt",
182
- "nbytes": 140489407,
183
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0016.pt",
184
- "sha256": "45b5ef558cc309f0c9a9e9e3e9ce5d6a5236e60052833d5938642815a1fe3bdf"
185
- },
186
- {
187
- "name": "core_block_0017.pt",
188
- "nbytes": 140095249,
189
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0017.pt",
190
- "sha256": "59c46e5d931553b62bf4cf7188f7e38629defc0c3ecad264eed29ad2d292dca1"
191
- },
192
- {
193
- "name": "core_block_0018.pt",
194
- "nbytes": 140429656,
195
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0018.pt",
196
- "sha256": "0366c5cbd1a9469e0f381d3531dbf6480a38e6c0916ef2e9ecd7a3e64fd4cf95"
197
- },
198
- {
199
- "name": "core_block_0019.pt",
200
- "nbytes": 140272712,
201
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0019.pt",
202
- "sha256": "444b2f9a6d52ed3a6ca7fc004003013ed9d5af43fbd90606632de3c6425aa286"
203
- },
204
- {
205
- "name": "core_block_0020.pt",
206
- "nbytes": 140353129,
207
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0020.pt",
208
- "sha256": "e6946904dec1be96c5587179a1adad78925fc9af1ae351b32067a01bbfd2fc20"
209
- },
210
- {
211
- "name": "core_block_0021.pt",
212
- "nbytes": 140165186,
213
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0021.pt",
214
- "sha256": "47704ee4da8535a7756ab3f94d28d45516d495047004cb43f676072c96d59876"
215
- },
216
- {
217
- "name": "core_block_0022.pt",
218
- "nbytes": 140519225,
219
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0022.pt",
220
- "sha256": "25caa715e4b837036509311feaad3e2214faa893571f0f206208a3c05f10f399"
221
- },
222
- {
223
- "name": "core_block_0023.pt",
224
- "nbytes": 140770939,
225
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0023.pt",
226
- "sha256": "df9cd21e4b705a026cad85b337978ad4c0371b85e9cbb5e040c40923dfe5b1e0"
227
- },
228
- {
229
- "name": "core_block_0024.pt",
230
- "nbytes": 139687856,
231
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0024.pt",
232
- "sha256": "21e723f119143f48bc565d23bcffd035b6310b3f0c9c359447455d83e6d57e71"
233
- },
234
- {
235
- "name": "core_block_0025.pt",
236
- "nbytes": 139711189,
237
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0025.pt",
238
- "sha256": "d96e6dadd43697cf15ec93e90a3b6d54d569cb1db16d24d9260671977135a1a5"
239
- },
240
- {
241
- "name": "core_block_0026.pt",
242
- "nbytes": 139650780,
243
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0026.pt",
244
- "sha256": "99a4ec0b18db5c6ea1a322f0492bfb8c64abf5d6161d8d1e77459021bd053b0b"
245
- },
246
- {
247
- "name": "core_block_0027.pt",
248
- "nbytes": 139828500,
249
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/core_block_0027.pt",
250
- "sha256": "7eeb7f46d31428152fa95eacf441f4ac3a40faf46db2279b30af133c30213a67"
251
- },
252
- {
253
- "name": "detachable_50m_base.pt",
254
- "nbytes": 122165841,
255
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_base.pt",
256
- "sha256": "5526a4481073566daa330a4890770c0f237de28d46b62ea15a55ce0de56625fa"
257
- },
258
- {
259
- "name": "detachable_50m_block_0000.pt",
260
- "nbytes": 2721112,
261
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0000.pt",
262
- "sha256": "387374193aec3a957a9ce601a1dbc73671356a9b41877d8f5d42884a3aa222ee"
263
- },
264
- {
265
- "name": "detachable_50m_block_0001.pt",
266
- "nbytes": 2699699,
267
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0001.pt",
268
- "sha256": "be7d9d7cdb2347ddcdbcd477b89081785dd9d23658a9272e12dff47ad28db487"
269
- },
270
- {
271
- "name": "detachable_50m_block_0002.pt",
272
- "nbytes": 2703503,
273
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0002.pt",
274
- "sha256": "fdf8c32bfd76a59f13debcfa44d67e5f9c87b7febd172f7395820da6cdba4b25"
275
- },
276
- {
277
- "name": "detachable_50m_block_0003.pt",
278
- "nbytes": 2703434,
279
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0003.pt",
280
- "sha256": "e5d46203d34baf7f01463dae01c777932b5d160d9b38676f52efbc6a85b5590e"
281
- },
282
- {
283
- "name": "detachable_50m_block_0004.pt",
284
- "nbytes": 2704928,
285
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0004.pt",
286
- "sha256": "3dbee936ae06937f9cdb2ad0b553bb64716a1817a5527406cea1843c103cdaf9"
287
- },
288
- {
289
- "name": "detachable_50m_block_0005.pt",
290
- "nbytes": 2711191,
291
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0005.pt",
292
- "sha256": "c9a6c46a4fb080f9e356bcd3bf2f16e6521c45e95ae1d54c6d36b22a912dcba7"
293
- },
294
- {
295
- "name": "detachable_50m_block_0006.pt",
296
- "nbytes": 2720386,
297
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0006.pt",
298
- "sha256": "c71e92f8475f91bfcfb4392f3ae2d68178285d055d9df05088b488ac343aa073"
299
- },
300
- {
301
- "name": "detachable_50m_block_0007.pt",
302
- "nbytes": 2708902,
303
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0007.pt",
304
- "sha256": "4276ad0a28ffa61bb72c1b2cc1a6f966788e43fa7d0fdc8cb307d3ea23a3209c"
305
- },
306
- {
307
- "name": "detachable_50m_block_0008.pt",
308
- "nbytes": 2717825,
309
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0008.pt",
310
- "sha256": "0070cd13087e114a0cd6a2508f17d87e495ead4287179e00e2c6e728c1250b2f"
311
- },
312
- {
313
- "name": "detachable_50m_block_0009.pt",
314
- "nbytes": 2714571,
315
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0009.pt",
316
- "sha256": "832ff495c20ae486edd110d0573ee594687090baeef30d9561173774dbb00742"
317
- },
318
- {
319
- "name": "detachable_50m_block_0010.pt",
320
- "nbytes": 2708715,
321
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0010.pt",
322
- "sha256": "059912bae2b7f9972bdd2ff2969d4858a0502b080ecaf148c6cf038a823044f7"
323
- },
324
- {
325
- "name": "detachable_50m_block_0011.pt",
326
- "nbytes": 2719947,
327
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0011.pt",
328
- "sha256": "196085d4d51d90cc6e13591f6bc3e1e351361ba4bc285b35bb0f3d6c7b1c6ccd"
329
- },
330
- {
331
- "name": "detachable_50m_block_0012.pt",
332
- "nbytes": 2702585,
333
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0012.pt",
334
- "sha256": "3b9d8bf592fe978b5186184f58be196cd2eb297c16a6d73ab54125bdde2e8863"
335
- },
336
- {
337
- "name": "detachable_50m_block_0013.pt",
338
- "nbytes": 2698081,
339
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0013.pt",
340
- "sha256": "54a7951bb111ab774b15178107fdd550c9746506969dfa9539d76ded7c54ca1f"
341
- },
342
- {
343
- "name": "detachable_50m_block_0014.pt",
344
- "nbytes": 2704320,
345
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0014.pt",
346
- "sha256": "0e74ab2716fbdc7cf14e3f810b37f43bfc6d059207eb239e223f827929ae66e4"
347
- },
348
- {
349
- "name": "detachable_50m_block_0015.pt",
350
- "nbytes": 2701065,
351
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0015.pt",
352
- "sha256": "fae0810b9416cbace5f1b5cb15ed1abec7a33ebe6455a3f523bb5377c9b93315"
353
- },
354
- {
355
- "name": "detachable_50m_block_0016.pt",
356
- "nbytes": 2699015,
357
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0016.pt",
358
- "sha256": "4581469635ecb2e7593af3af000a0af3f5bb16ab443f0be16a922cf6f30c5456"
359
- },
360
- {
361
- "name": "detachable_50m_block_0017.pt",
362
- "nbytes": 2702253,
363
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0017.pt",
364
- "sha256": "7e5eaac6958ba25ee8cded92896a1ecf474f1d92acc12ac3589febbdb2369e82"
365
- },
366
- {
367
- "name": "detachable_50m_block_0018.pt",
368
- "nbytes": 2709069,
369
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0018.pt",
370
- "sha256": "8a1c064319d4d3c039ab0a7c28ba8fcb322a67014181a6fbb657334ecf49bb28"
371
- },
372
- {
373
- "name": "detachable_50m_block_0019.pt",
374
- "nbytes": 1970314,
375
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0019.pt",
376
- "sha256": "c5cae4e0287c0645f769269369cce8d68ca7581821279c5879f969c7124c06af"
377
- },
378
- {
379
- "name": "detachable_50m_block_0020.pt",
380
- "nbytes": 1968762,
381
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0020.pt",
382
- "sha256": "902ec3fbe92eca397678e7e0e56406047d49fd1c6d56c1036c4faf6e7eb0c118"
383
- },
384
- {
385
- "name": "detachable_50m_block_0021.pt",
386
- "nbytes": 1969967,
387
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/detachable_50m_block_0021.pt",
388
- "sha256": "601c67cdeb60a7bd46d6c8a2164985099365431d29b1015d3b2ff79f562104da"
389
- },
390
- {
391
- "name": "ar.pt",
392
- "nbytes": 608664135,
393
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/ar.pt",
394
- "sha256": "e37df7485ffcf2d5ea8e2135d66201f0cc8523b6286393c3960052f4169a6216"
395
- },
396
- {
397
- "name": "sat.pt",
398
- "nbytes": 608659405,
399
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/sat.pt",
400
- "sha256": "532fb07b269d364991ad7ca3acebf69275843ea4382fcc568fd3a5d3dcebc912"
401
- },
402
- {
403
- "name": "nat.pt",
404
- "nbytes": 608664135,
405
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/nat.pt",
406
- "sha256": "81411382f5ce333da34b773c100a8776d8fa0a7cd253aa5dce61ac64245ef830"
407
- },
408
- {
409
- "name": "opt.pt",
410
- "nbytes": 1901401510,
411
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/opt.pt",
412
- "sha256": "8de2b11252f9396eedc0e03c5de391b2d924bc52d97d8934e6bae86abc31344d"
413
- },
414
- {
415
- "name": "scaler.pt",
416
- "nbytes": 149,
417
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/scaler.pt",
418
- "sha256": "c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c"
419
- },
420
- {
421
- "name": "agillm5_detachable_50m_bridge.pt",
422
- "nbytes": 1224856,
423
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_bridge.pt",
424
- "sha256": "3091c5282a6be7c7d326a4d2f6dd70791fa7bc00b433f142bbf44edd95a05f8e"
425
- },
426
- {
427
- "name": "agillm5_detachable_50m_opt.pt",
428
- "nbytes": 64498216,
429
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_opt.pt",
430
- "sha256": "d090a07334dc827ef2d89172cf9272135c96b86ddfe88b8425557816700ad228"
431
- },
432
- {
433
- "name": "agillm5_detachable_50m_scaler.pt",
434
- "nbytes": 149,
435
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_scaler.pt",
436
- "sha256": "c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c"
437
- },
438
- {
439
- "name": "agillm5_detachable_50m_rng.pt",
440
- "nbytes": 3542,
441
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_rng.pt",
442
- "sha256": "e68186f3ba08e3c9a4e27d1258805b7d6045bfd89fc3369c0a0c889350d15f7f"
443
- },
444
- {
445
- "name": "agillm5_detachable_50m_counters.pt",
446
- "nbytes": 30517,
447
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.shards/agillm5_detachable_50m_counters.pt",
448
- "sha256": "981c6cb6c1aaee63ea7941544a935899a5a87747edb5ba4eb5892e294be6d70f"
449
- }
450
- ],
451
- "step": 2847537,
452
- "tokenizer_sidecar_nbytes": 17465554,
453
- "tokenizer_sidecar_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02847537_from02845290_20260911T082529_1789115129082045530_1042995_000002.pt.tokenizer.json",
454
- "tokenizer_sidecar_sha256": "02c3d6fa78c602ced6565e5f46b7a6f10ba73e3f53aafbcff21e598a45acf031"
455
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
checkpoints/step2849585_20260911/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.provenance.json DELETED
@@ -1,606 +0,0 @@
1
- {
2
- "batch_size": 56,
3
- "block_size": 2048,
4
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
5
- "checkpoint_type": "full",
6
- "created_at_iso": "2026-09-11T10:32:22Z",
7
- "created_at_unix": 1789122742.7115822,
8
- "created_from_scratch": false,
9
- "dataset_provenance": {
10
- "chat": false,
11
- "dataset_field_text": "text",
12
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
13
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
14
- "dataset_hotload_config_stamp": [
15
- 50,
16
- 1124990893,
17
- 1789091829464680808,
18
- 7210
19
- ],
20
- "dataset_hotload_config_used": true,
21
- "dataset_hotload_enabled": true,
22
- "dataset_hotload_generation": 13,
23
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
24
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
25
- "dataset_revisions": {
26
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
27
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
28
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
29
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
30
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
31
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
32
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
33
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
34
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
35
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
36
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
37
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
38
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
39
- },
40
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
41
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
42
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
43
- "has_language_mix": true,
44
- "has_numeracy": false,
45
- "hot_config_expected_sha256": "",
46
- "hot_config_has_append": true,
47
- "hot_config_has_datasets": false,
48
- "hot_config_mtime": 1789092253.2119184,
49
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
50
- "hot_config_used": true,
51
- "hot_reload_capable": true,
52
- "phase": "pretrain",
53
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
54
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
55
- "schema": "agillm.dataset_provenance.v1",
56
- "sequence_prefetch_depth": 256,
57
- "source_count": 23,
58
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
59
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
60
- "source_specs": [
61
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
62
- "wikimedia/wikipedia:20231101.en|1.63",
63
- "allenai/c4:en|0.36",
64
- "Skylion007/openwebtext|0.18",
65
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
66
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
67
- "EleutherAI/proof-pile-2:all|0.67",
68
- "allenai/dolma:v1_6-sample|0.5",
69
- "codeparrot/codeparrot-clean|0.24",
70
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
71
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
72
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
73
- "allenai/dolmino-mix-1124|0.78",
74
- "mlfoundations/dclm-baseline-1.0|1.05",
75
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
76
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
77
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
78
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
79
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
80
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
81
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
82
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
83
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
84
- ],
85
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
86
- "val_source_requested": ""
87
- },
88
- "effective_global_step": 5698273,
89
- "effective_seen_tok": 283698599936,
90
- "global_origin_seen_tok": 0,
91
- "global_origin_step": 2848688,
92
- "gpu": {
93
- "allocated_gb": 7.17,
94
- "peak_allocated_gb": 12.3,
95
- "reserved_gb": 13.5
96
- },
97
- "hostname": "ead6505f92c4",
98
- "lane": "pretrain",
99
- "launch": {
100
- "argv": [
101
- "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py",
102
- "train",
103
- "--preset",
104
- "agillm4_floor",
105
- "--block",
106
- "2048",
107
- "--source",
108
- "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
109
- "--nat_mask_token_id",
110
- "2",
111
- "--tie_weights",
112
- "--tie_kv",
113
- "--moe_ffn",
114
- "--moe_experts",
115
- "2",
116
- "--moe_top_k",
117
- "1",
118
- "--moe_mlp_mult",
119
- "4",
120
- "--moe_shared_experts",
121
- "1",
122
- "--moe_shared_mlp_mult",
123
- "2",
124
- "--moe_aux_coef",
125
- "0.01",
126
- "--moe_z_coef",
127
- "0.001",
128
- "--amp",
129
- "--grad_checkpoint",
130
- "--attn_backend",
131
- "sdpa",
132
- "--sublinear_window",
133
- "128",
134
- "--sublinear_stride",
135
- "128",
136
- "--sublinear_max_anchors",
137
- "128",
138
- "--sublinear_chunk",
139
- "128",
140
- "--sublinear_sinks",
141
- "4",
142
- "--sublinear_recent_anchors",
143
- "64",
144
- "--no-sublinear_pooled_landmarks",
145
- "--anchor_stride",
146
- "256",
147
- "--anchor_max",
148
- "2048",
149
- "--anchor_position",
150
- "-1",
151
- "--alibi_mode",
152
- "corrected",
153
- "--alibi_scale",
154
- "0.0",
155
- "--optimizer",
156
- "adamw8bit",
157
- "--weight_decay",
158
- "0.0",
159
- "--lr_decay",
160
- "cosine",
161
- "--lr_decay_tokens",
162
- "49000954880",
163
- "--lr_min_mult",
164
- "0.2",
165
- "--lr_schedule_reset_on_resume",
166
- "--lr_warmup_tokens",
167
- "10000000",
168
- "--lr_warmup_min_mult",
169
- "0.1",
170
- "--dblock",
171
- "--dblock_schedule",
172
- "roundrobin",
173
- "--dblock_router_ramp_steps",
174
- "64",
175
- "--dblock_warmup_steps",
176
- "14",
177
- "--dblock_checkpoint_skip_tail",
178
- "0",
179
- "--dblock_loop_layers",
180
- "0",
181
- "--dblock_loop_start",
182
- "0",
183
- "--dblock_loop_cond_scale",
184
- "1.0",
185
- "--dblock_sigma_curriculum_steps",
186
- "2000",
187
- "--dblock_sigma_sampling",
188
- "lognormal",
189
- "--dblock_sigma_stratified",
190
- "--dblock_sigma_min",
191
- "0.002",
192
- "--dblock_sigma_max",
193
- "2.0",
194
- "--dblock_sigma_pmean",
195
- "-1.2",
196
- "--dblock_sigma_pstd",
197
- "1.2",
198
- "--dblock_edm_wmax",
199
- "5.0",
200
- "--dblock_ar_prob",
201
- "0.8",
202
- "--dblock_sat_prob",
203
- "0.1",
204
- "--dblock_nat_prob",
205
- "0.1",
206
- "--dblock_ar_loss_tokens",
207
- "8192",
208
- "--dblock_sat_loss_tokens",
209
- "4096",
210
- "--dblock_nat_loss_tokens",
211
- "4096",
212
- "--dblock_nat_embed_noise_mode",
213
- "off",
214
- "--dblock_nat_embed_noise_scale",
215
- "1.0",
216
- "--nat_loss_weight",
217
- "1.0",
218
- "--nat_mask_ratio",
219
- "0.5",
220
- "--nat_max_tokens",
221
- "0",
222
- "--nat_span_mask_prob",
223
- "0.35",
224
- "--nat_suffix_mask_prob",
225
- "0.20",
226
- "--nat_span_max_tokens",
227
- "0",
228
- "--loss_spike_skip",
229
- "0.0",
230
- "--sat_every",
231
- "1",
232
- "--nat_every",
233
- "1",
234
- "--dblock_fullstack_anchor_spike_skip",
235
- "0.0",
236
- "--dblock_fullstack_anchor_max_ce",
237
- "0.0",
238
- "--dblock_anchor_softcap_mult",
239
- "4.0",
240
- "--dblock_anchor_softcap_min_ce",
241
- "25.0",
242
- "--dblock_anchor_softcap_max_ce",
243
- "50.0",
244
- "--dblock_fullstack_ar_every",
245
- "3",
246
- "--dblock_fullstack_ar_offset",
247
- "0",
248
- "--dblock_fullstack_ar_tokens",
249
- "256",
250
- "--dblock_fullstack_sat_every",
251
- "3",
252
- "--dblock_fullstack_sat_offset",
253
- "1",
254
- "--dblock_fullstack_sat_tokens",
255
- "128",
256
- "--dblock_fullstack_nat_every",
257
- "3",
258
- "--dblock_fullstack_nat_offset",
259
- "2",
260
- "--dblock_fullstack_nat_tokens",
261
- "128",
262
- "--dblock_fullstack_nat_mask_id",
263
- "-1",
264
- "--ckpt_codec",
265
- "block-sharded-zstd",
266
- "--async_update_every_steps",
267
- "0",
268
- "--dblock_router_hidden",
269
- "64",
270
- "--dblock_router_heads",
271
- "4",
272
- "--dblock_router_layers",
273
- "2",
274
- "--dblock_router_lr",
275
- "0.0005",
276
- "--dblock_explore",
277
- "0.08",
278
- "--dblock_max_stale_steps",
279
- "64",
280
- "--dblock_max_count_skew",
281
- "1.35",
282
- "--dblock_stale_bonus",
283
- "0.35",
284
- "--dblock_undertrain_bonus",
285
- "0.25",
286
- "--dblock_activation_offload_min_mb",
287
- "1.0",
288
- "--nat_document_boundary_aware",
289
- "--dblock_fullstack_anchor_deterministic_eval",
290
- "--dblock_fullstack_anchor_spike_retry_limit",
291
- "3",
292
- "--dblock_local_spike_retry_limit",
293
- "3",
294
- "--dblock_fullstack_nat_no_valid_crop_limit",
295
- "8",
296
- "--target_tokens",
297
- "400000000000",
298
- "--lr_schedule_reanchor_on_resume",
299
- "--save_every_sec",
300
- "1800",
301
- "--delta_max_keep",
302
- "0",
303
- "--max_ckpts",
304
- "2",
305
- "--disk_free_floor_gb",
306
- "12.0",
307
- "--heartbeat_every_sec",
308
- "30",
309
- "--dblock_fullstack_ar_weight",
310
- "1.0",
311
- "--dblock_fullstack_sat_weight",
312
- "0.1",
313
- "--dblock_fullstack_nat_weight",
314
- "0.1",
315
- "--dblock_local_aux_warn_ce",
316
- "0.0",
317
- "--dblock_local_aux_max_ce",
318
- "0.0",
319
- "--lr_schedule_origin_tokens",
320
- "85002387456",
321
- "--dblock_stop_after_commits",
322
- "0",
323
- "--no-oom_auto_backoff",
324
- "--require_exact_resume_state",
325
- "--dblock_objective_mode",
326
- "committed_9_1_1",
327
- "--dblock_ar_weight",
328
- "0.25",
329
- "--dblock_sat_weight",
330
- "0.25",
331
- "--dblock_nat_weight",
332
- "0.25",
333
- "--dblock_aux_grad_ratio",
334
- "0.5",
335
- "--dblock_min_ar_share",
336
- "0.8",
337
- "--lr_core",
338
- "2e-7",
339
- "--lr_head",
340
- "2e-7",
341
- "--dblock_checkpoint_stride",
342
- "2",
343
- "--empty_cache_every_steps",
344
- "0",
345
- "--dblock_log_every",
346
- "25",
347
- "--cuda_max_reserved_mib",
348
- "0",
349
- "--cuda_min_free_mib",
350
- "0",
351
- "--repo",
352
- "--repo_retrofit",
353
- "--repo_start_layer",
354
- "9",
355
- "--repo_dim",
356
- "160",
357
- "--repo_theta",
358
- "500000.0",
359
- "--repo_lr",
360
- "1e-5",
361
- "--dblock_clean_context",
362
- "e2e",
363
- "--dblock_local_rows",
364
- "2",
365
- "--dblock_clean_context_noise_scale",
366
- "0.0",
367
- "--dblock_depth_anchor_every",
368
- "50",
369
- "--dblock_fullstack_anchor_cap",
370
- "256",
371
- "--dblock_fullstack_ar_rows",
372
- "1",
373
- "--detachable_50m_upgrade_multimode",
374
- "--detachable_50m_upgrade_logit_distill",
375
- "--detachable_50m_upgrade_sat_gate_v2713",
376
- "--detachable_50m_create_from_legacy",
377
- "--detachable_50m_seed_from_mature",
378
- "--detachable_50m_lr",
379
- "0.0001",
380
- "--detachable_50m_weight_decay",
381
- "0.0",
382
- "--detachable_50m_update_mode",
383
- "sublayer",
384
- "--detachable_50m_batch",
385
- "1",
386
- "--detachable_50m_tokens",
387
- "64",
388
- "--detachable_50m_loss_tokens",
389
- "64",
390
- "--detachable_50m_every",
391
- "1",
392
- "--detachable_50m_teacher_every",
393
- "1",
394
- "--detachable_50m_distill_weight",
395
- "1.0",
396
- "--detachable_50m_logit_distill_alpha",
397
- "0.30",
398
- "--detachable_50m_logit_distill_temperature",
399
- "1.0",
400
- "--detachable_50m_logit_distill_tokens",
401
- "64",
402
- "--detachable_50m_hidden_distill_weight",
403
- "0.0",
404
- "--detachable_50m_anchor_every",
405
- "128",
406
- "--detachable_50m_anchor_tokens",
407
- "32",
408
- "--detachable_50m_seed",
409
- "20260830",
410
- "--detachable_50m_log_every",
411
- "10",
412
- "--detachable_50m_telemetry_keep",
413
- "128",
414
- "--weight_accel_mode",
415
- "shadow",
416
- "--weight_accel_dir",
417
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel",
418
- "--weight_accel_every",
419
- "1",
420
- "--weight_accel_train_every",
421
- "4",
422
- "--weight_accel_record_every",
423
- "14",
424
- "--weight_accel_state_every",
425
- "256",
426
- "--detachable_50m",
427
- "--detachable_50m_optimizer",
428
- "paged_adamw8bit",
429
- "--delta_every_steps",
430
- "0",
431
- "--delta_every_sec",
432
- "0",
433
- "--resume",
434
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt",
435
- "--save_dir",
436
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B",
437
- "--data_seed",
438
- "2848730",
439
- "--batch_size",
440
- "56",
441
- "--dblock_blocks",
442
- "14",
443
- "--dblock_direct56_enabled",
444
- "1",
445
- "--dblock_layers_per_update",
446
- "0",
447
- "--dblock_router",
448
- "shadow",
449
- "--dblock_router_blend",
450
- "0.0",
451
- "--dblock_router_full_anchor_every",
452
- "128",
453
- "--dblock_router_train_layer_policy",
454
- "cyclic",
455
- "--dblock_router_train_layers_per_update",
456
- "1",
457
- "--dblock_sublayer_mode",
458
- "off",
459
- "--dblock_target_route_activate_from_checkpoint",
460
- "1",
461
- "--dblock_target_route_activation_checkpoint_step",
462
- "2717509",
463
- "--dblock_target_route_activation_parent_pointer_sha256",
464
- "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
465
- "--dblock_target_route_offset",
466
- "1",
467
- "--dblock_target_route_policy",
468
- "xor_fold56_v1",
469
- "--dblock_train_layer_policy",
470
- "cyclic",
471
- "--dblock_train_layers_per_update",
472
- "1",
473
- "--dblock_train_sublayer_policy",
474
- "cyclic",
475
- "--dblock_train_sublayers_per_update",
476
- "1",
477
- "--granularity_label",
478
- "one-sublayer-local-grad-direct56-teacher-v27.16",
479
- "--oom_known_safe_batch",
480
- "56",
481
- "--val_every_sec",
482
- "600",
483
- "--val_file",
484
- "/workspace/agillm43_heldout_v3/token_ids.json",
485
- "--val_sha256",
486
- "513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21"
487
- ],
488
- "argv_string": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
489
- "cwd": "/workspace/agillm44_throughput_80k_20260910",
490
- "env": {
491
- "AGILLM43_DBLOCK_CC_NOISE_SCALE": "0.0",
492
- "AGILLM43_DBLOCK_CLEAN_CONTEXT": "e2e",
493
- "AGILLM43_DBLOCK_CLEAN_CONTEXT_NOISE_SCALE": "0.0",
494
- "AGILLM43_DBLOCK_DEPTH_ANCHOR_EVERY": "50",
495
- "AGILLM43_DBLOCK_FULLSTACK_ANCHOR_CAP": "256",
496
- "AGILLM43_DBLOCK_FULLSTACK_AR_ROWS": "1",
497
- "AGILLM43_DBLOCK_LOCAL_ROWS": "2",
498
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATE_FROM_CHECKPOINT": "1",
499
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_CHECKPOINT_STEP": "2717509",
500
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_PARENT_POINTER_SHA256": "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
501
- "AGILLM43_DBLOCK_TARGET_ROUTE_OFFSET": "1",
502
- "AGILLM43_DBLOCK_TARGET_ROUTE_POLICY": "xor_fold56_v1",
503
- "AGILLM43_DECOMPRESS_CACHE": "0",
504
- "AGILLM43_DETACHABLE50M_ANCHOR_EVERY": "128",
505
- "AGILLM43_DETACHABLE50M_ANCHOR_TOKENS": "<redacted>",
506
- "AGILLM43_DETACHABLE50M_BATCH": "1",
507
- "AGILLM43_DETACHABLE50M_CREATE_FROM_LEGACY": "1",
508
- "AGILLM43_DETACHABLE50M_DISTILL_WEIGHT": "1.0",
509
- "AGILLM43_DETACHABLE50M_ENABLED": "1",
510
- "AGILLM43_DETACHABLE50M_EVERY": "1",
511
- "AGILLM43_DETACHABLE50M_GATE_V2713_START_CHECKPOINT_SHA256": "39f442fc0e6f9fad6f875e067f63db67f33a41a53e8cbc49ce1591010bb4aec3",
512
- "AGILLM43_DETACHABLE50M_HIDDEN_DISTILL_WEIGHT": "0.0",
513
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_ALPHA": "0.30",
514
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TEMPERATURE": "1.0",
515
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TOKENS": "<redacted>",
516
- "AGILLM43_DETACHABLE50M_LOG_EVERY": "10",
517
- "AGILLM43_DETACHABLE50M_LOSS_TOKENS": "<redacted>",
518
- "AGILLM43_DETACHABLE50M_LR": "0.0001",
519
- "AGILLM43_DETACHABLE50M_OPTIMIZER": "paged_adamw8bit",
520
- "AGILLM43_DETACHABLE50M_SEED": "20260830",
521
- "AGILLM43_DETACHABLE50M_SEED_FROM_MATURE": "1",
522
- "AGILLM43_DETACHABLE50M_TEACHER_EVERY": "1",
523
- "AGILLM43_DETACHABLE50M_TELEMETRY_KEEP": "128",
524
- "AGILLM43_DETACHABLE50M_TOKENS": "<redacted>",
525
- "AGILLM43_DETACHABLE50M_UPDATE_MODE": "sublayer",
526
- "AGILLM43_DETACHABLE50M_UPGRADE_LOGIT_DISTILL": "1",
527
- "AGILLM43_DETACHABLE50M_UPGRADE_MULTIMODE": "1",
528
- "AGILLM43_DETACHABLE50M_UPGRADE_SAT_GATE_V2713": "1",
529
- "AGILLM43_DETACHABLE50M_WEIGHT_DECAY": "0.0",
530
- "AGILLM43_REPO_DIM": "160",
531
- "AGILLM43_REPO_ENABLED": "1",
532
- "AGILLM43_REPO_LR": "1e-5",
533
- "AGILLM43_REPO_RETROFIT": "1",
534
- "AGILLM43_REPO_START_LAYER": "9",
535
- "AGILLM43_REPO_THETA": "500000.0",
536
- "AGILLM43_SEQUENCE_PREFETCH_DEPTH": "256",
537
- "AGILLM43_SINGLEFILE_PROFILE": "intelligence-v25-satvar-pretrain",
538
- "AGILLM43_SUPERVISE_SET_ARGS_JSON": "{\"--batch_size\":\"56\",\"--dblock_blocks\":\"14\",\"--dblock_direct56_enabled\":\"1\",\"--dblock_layers_per_update\":\"0\",\"--dblock_router\":\"shadow\",\"--dblock_router_blend\":\"0.0\",\"--dblock_router_full_anchor_every\":\"128\",\"--dblock_router_train_layer_policy\":\"cyclic\",\"--dblock_router_train_layers_per_update\":\"1\",\"--dblock_sublayer_mode\":\"off\",\"--dblock_target_route_activate_from_checkpoint\":\"1\",\"--dblock_target_route_activation_checkpoint_step\":\"2717509\",\"--dblock_target_route_activation_parent_pointer_sha256\":\"04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827\",\"--dblock_target_route_offset\":\"1\",\"--dblock_target_route_policy\":\"xor_fold56_v1\",\"--dblock_train_layer_policy\":\"cyclic\",\"--dblock_train_layers_per_update\":\"1\",\"--dblock_train_sublayer_policy\":\"cyclic\",\"--dblock_train_sublayers_per_update\":\"1\",\"--granularity_label\":\"one-sublayer-local-grad-direct56-teacher-v27.16\",\"--oom_known_safe_batch\":\"56\",\"--val_every_sec\":\"600\",\"--val_file\":\"/workspace/agillm43_heldout_v3/token_ids.json\",\"--val_sha256\":\"513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21\"}",
539
- "AGILLM43_WEIGHT_ACCEL_MODE": "shadow",
540
- "AGILLM_DATASETS_STREAMING_READ_MAX_RETRIES": "2",
541
- "AGILLM_DATASETS_STREAMING_READ_RETRY_INTERVAL": "1",
542
- "AGILLM_DATASET_AGENT_ROUTER": "0",
543
- "AGILLM_DATASET_HOTLOAD_CONFIG": "/workspace/agillm43_dataset_hotload.json",
544
- "AGILLM_DATASET_HOT_RELOAD_CHECK_TOKENS": "<redacted>",
545
- "AGILLM_DATASET_HOT_RELOAD_SEC": "5",
546
- "AGILLM_DATASET_NN_ROUTER": "1",
547
- "AGILLM_DATASET_ROUTER_BLEND": "0.25",
548
- "AGILLM_DATASET_ROUTER_EXPLORE": "0.03",
549
- "AGILLM_DATASET_ROUTER_LR": "0.005",
550
- "AGILLM_DATASET_ROUTER_MIN_SCORE": "0.15",
551
- "AGILLM_DATASET_ROUTER_SHARPNESS": "2.0",
552
- "AGILLM_ENABLE_CUPY": "0",
553
- "AGILLM_HF_HUB_DOWNLOAD_TIMEOUT": "30",
554
- "AGILLM_HF_TRANSPORT": "geth_hf_relay",
555
- "AGILLM_HOT_CONFIG": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
556
- "AGILLM_HOT_CONFIG_SHA256": "",
557
- "AGILLM_MAX_EXAMPLE_CHARS": "32768",
558
- "AGILLM_MAX_EXAMPLE_TOKENS": "<redacted>",
559
- "AGILLM_PARENT_POINTER_SHA256": "443aed3f7eb17df42d0840fe208c735ab58671bebb26c99be4c80d96c28d756c",
560
- "AGILLM_REMOTE_SHARDS_PER_ITER": "4",
561
- "AGILLM_RUN_ID": "singlefile_intelligence_v25_20260911T095413Z_from_step000002848688",
562
- "AGILLM_STREAM_NEXT_TIMEOUT_SEC": "60",
563
- "AGILLM_STREAM_OPEN_TIMEOUT_SEC": "120",
564
- "AGILLM_STREAM_SOCKET_TIMEOUT_SEC": "30",
565
- "AGILLM_STREAM_SOURCE_FATAL_COOLDOWN_SEC": "600",
566
- "AGILLM_STREAM_SOURCE_MAX_COOLDOWN_SEC": "120",
567
- "AGILLM_SYNTHETIC_TOKENIZER": "<redacted>",
568
- "CUDA_MODULE_LOADING": "LAZY",
569
- "CUDA_VERSION": "12.1.1",
570
- "CUDA_VISIBLE_DEVICES": "0",
571
- "HF_HUB_DOWNLOAD_TIMEOUT": "30",
572
- "HF_HUB_ETAG_TIMEOUT": "10",
573
- "PYTORCH_CUDA_ALLOC_CONF": "backend:cudaMallocAsync",
574
- "PYTORCH_VERSION": "2.2.1"
575
- },
576
- "parent_proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py supervise --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --log /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/logs/singlefile_intelligence_v25_satvar_pretrain.log --sleep_sec 15",
577
- "pid": 1518633,
578
- "ppid": 1518567,
579
- "proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
580
- "schema": "agillm.launch.v1"
581
- },
582
- "local_step": 2849585,
583
- "loss": 5.471035957336426,
584
- "nat_mask_contract": "agillm.nat-mask.v1",
585
- "nat_mask_contract_source": "explicit-legacy-recovery",
586
- "nat_mask_eos_id": 1,
587
- "nat_mask_legacy_blank_id": 1,
588
- "nat_mask_migrated_from_id": 1,
589
- "nat_mask_schema_version": 1,
590
- "nat_mask_token_id": 2,
591
- "pid": 1518633,
592
- "schema_version": 1,
593
- "seen_tok": 283698599936,
594
- "source_is_lane_checkpoint": true,
595
- "source_is_master_checkpoint": false,
596
- "source_is_non_master_checkpoint": false,
597
- "step": 2849585,
598
- "train_argv": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
599
- "train_script": "runtime_retain_activations_v2.py",
600
- "train_script_sha256": "989a7ee5e4344b7b53b67b70d0df27dc42461d864084283d1d7acb262e0a7081",
601
- "warmstart_base_seen_tok": 0,
602
- "warmstart_base_step": 2848688,
603
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
604
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt",
605
- "warmstart_source_step": 2848688
606
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
checkpoints/step2849585_20260911/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_scaler.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c
3
- size 149
 
 
 
 
checkpoints/step2849585_20260911/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/scaler.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c
3
- size 149
 
 
 
 
checkpoints/step2849585_20260911/serving_candidate.json DELETED
@@ -1,710 +0,0 @@
1
- {
2
- "agillm43_provenance": {
3
- "batch_size": 56,
4
- "block_size": 2048,
5
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
6
- "checkpoint_type": "full",
7
- "created_at_iso": "2026-09-11T10:32:22Z",
8
- "created_at_unix": 1789122742.7115822,
9
- "created_from_scratch": false,
10
- "dataset_provenance": {
11
- "chat": false,
12
- "dataset_field_text": "text",
13
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
14
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
15
- "dataset_hotload_config_stamp": [
16
- 50,
17
- 1124990893,
18
- 1789091829464680808,
19
- 7210
20
- ],
21
- "dataset_hotload_config_used": true,
22
- "dataset_hotload_enabled": true,
23
- "dataset_hotload_generation": 13,
24
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
25
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
26
- "dataset_revisions": {
27
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
28
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
29
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
30
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
31
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
32
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
33
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
34
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
35
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
36
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
37
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
38
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
39
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
40
- },
41
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
42
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
43
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
44
- "has_language_mix": true,
45
- "has_numeracy": false,
46
- "hot_config_expected_sha256": "",
47
- "hot_config_has_append": true,
48
- "hot_config_has_datasets": false,
49
- "hot_config_mtime": 1789092253.2119184,
50
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
51
- "hot_config_used": true,
52
- "hot_reload_capable": true,
53
- "phase": "pretrain",
54
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
55
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
56
- "schema": "agillm.dataset_provenance.v1",
57
- "sequence_prefetch_depth": 256,
58
- "source_count": 23,
59
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
60
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
61
- "source_specs": [
62
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
63
- "wikimedia/wikipedia:20231101.en|1.63",
64
- "allenai/c4:en|0.36",
65
- "Skylion007/openwebtext|0.18",
66
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
67
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
68
- "EleutherAI/proof-pile-2:all|0.67",
69
- "allenai/dolma:v1_6-sample|0.5",
70
- "codeparrot/codeparrot-clean|0.24",
71
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
72
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
73
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
74
- "allenai/dolmino-mix-1124|0.78",
75
- "mlfoundations/dclm-baseline-1.0|1.05",
76
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
77
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
78
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
79
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
80
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
81
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
82
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
83
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
84
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
85
- ],
86
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
87
- "val_source_requested": ""
88
- },
89
- "effective_global_step": 5698273,
90
- "effective_seen_tok": 283698599936,
91
- "global_origin_seen_tok": 0,
92
- "global_origin_step": 2848688,
93
- "gpu": {
94
- "allocated_gb": 7.17,
95
- "peak_allocated_gb": 12.3,
96
- "reserved_gb": 13.5
97
- },
98
- "hostname": "ead6505f92c4",
99
- "lane": "pretrain",
100
- "launch": {
101
- "argv": [
102
- "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py",
103
- "train",
104
- "--preset",
105
- "agillm4_floor",
106
- "--block",
107
- "2048",
108
- "--source",
109
- "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
110
- "--nat_mask_token_id",
111
- "2",
112
- "--tie_weights",
113
- "--tie_kv",
114
- "--moe_ffn",
115
- "--moe_experts",
116
- "2",
117
- "--moe_top_k",
118
- "1",
119
- "--moe_mlp_mult",
120
- "4",
121
- "--moe_shared_experts",
122
- "1",
123
- "--moe_shared_mlp_mult",
124
- "2",
125
- "--moe_aux_coef",
126
- "0.01",
127
- "--moe_z_coef",
128
- "0.001",
129
- "--amp",
130
- "--grad_checkpoint",
131
- "--attn_backend",
132
- "sdpa",
133
- "--sublinear_window",
134
- "128",
135
- "--sublinear_stride",
136
- "128",
137
- "--sublinear_max_anchors",
138
- "128",
139
- "--sublinear_chunk",
140
- "128",
141
- "--sublinear_sinks",
142
- "4",
143
- "--sublinear_recent_anchors",
144
- "64",
145
- "--no-sublinear_pooled_landmarks",
146
- "--anchor_stride",
147
- "256",
148
- "--anchor_max",
149
- "2048",
150
- "--anchor_position",
151
- "-1",
152
- "--alibi_mode",
153
- "corrected",
154
- "--alibi_scale",
155
- "0.0",
156
- "--optimizer",
157
- "adamw8bit",
158
- "--weight_decay",
159
- "0.0",
160
- "--lr_decay",
161
- "cosine",
162
- "--lr_decay_tokens",
163
- "49000954880",
164
- "--lr_min_mult",
165
- "0.2",
166
- "--lr_schedule_reset_on_resume",
167
- "--lr_warmup_tokens",
168
- "10000000",
169
- "--lr_warmup_min_mult",
170
- "0.1",
171
- "--dblock",
172
- "--dblock_schedule",
173
- "roundrobin",
174
- "--dblock_router_ramp_steps",
175
- "64",
176
- "--dblock_warmup_steps",
177
- "14",
178
- "--dblock_checkpoint_skip_tail",
179
- "0",
180
- "--dblock_loop_layers",
181
- "0",
182
- "--dblock_loop_start",
183
- "0",
184
- "--dblock_loop_cond_scale",
185
- "1.0",
186
- "--dblock_sigma_curriculum_steps",
187
- "2000",
188
- "--dblock_sigma_sampling",
189
- "lognormal",
190
- "--dblock_sigma_stratified",
191
- "--dblock_sigma_min",
192
- "0.002",
193
- "--dblock_sigma_max",
194
- "2.0",
195
- "--dblock_sigma_pmean",
196
- "-1.2",
197
- "--dblock_sigma_pstd",
198
- "1.2",
199
- "--dblock_edm_wmax",
200
- "5.0",
201
- "--dblock_ar_prob",
202
- "0.8",
203
- "--dblock_sat_prob",
204
- "0.1",
205
- "--dblock_nat_prob",
206
- "0.1",
207
- "--dblock_ar_loss_tokens",
208
- "8192",
209
- "--dblock_sat_loss_tokens",
210
- "4096",
211
- "--dblock_nat_loss_tokens",
212
- "4096",
213
- "--dblock_nat_embed_noise_mode",
214
- "off",
215
- "--dblock_nat_embed_noise_scale",
216
- "1.0",
217
- "--nat_loss_weight",
218
- "1.0",
219
- "--nat_mask_ratio",
220
- "0.5",
221
- "--nat_max_tokens",
222
- "0",
223
- "--nat_span_mask_prob",
224
- "0.35",
225
- "--nat_suffix_mask_prob",
226
- "0.20",
227
- "--nat_span_max_tokens",
228
- "0",
229
- "--loss_spike_skip",
230
- "0.0",
231
- "--sat_every",
232
- "1",
233
- "--nat_every",
234
- "1",
235
- "--dblock_fullstack_anchor_spike_skip",
236
- "0.0",
237
- "--dblock_fullstack_anchor_max_ce",
238
- "0.0",
239
- "--dblock_anchor_softcap_mult",
240
- "4.0",
241
- "--dblock_anchor_softcap_min_ce",
242
- "25.0",
243
- "--dblock_anchor_softcap_max_ce",
244
- "50.0",
245
- "--dblock_fullstack_ar_every",
246
- "3",
247
- "--dblock_fullstack_ar_offset",
248
- "0",
249
- "--dblock_fullstack_ar_tokens",
250
- "256",
251
- "--dblock_fullstack_sat_every",
252
- "3",
253
- "--dblock_fullstack_sat_offset",
254
- "1",
255
- "--dblock_fullstack_sat_tokens",
256
- "128",
257
- "--dblock_fullstack_nat_every",
258
- "3",
259
- "--dblock_fullstack_nat_offset",
260
- "2",
261
- "--dblock_fullstack_nat_tokens",
262
- "128",
263
- "--dblock_fullstack_nat_mask_id",
264
- "-1",
265
- "--ckpt_codec",
266
- "block-sharded-zstd",
267
- "--async_update_every_steps",
268
- "0",
269
- "--dblock_router_hidden",
270
- "64",
271
- "--dblock_router_heads",
272
- "4",
273
- "--dblock_router_layers",
274
- "2",
275
- "--dblock_router_lr",
276
- "0.0005",
277
- "--dblock_explore",
278
- "0.08",
279
- "--dblock_max_stale_steps",
280
- "64",
281
- "--dblock_max_count_skew",
282
- "1.35",
283
- "--dblock_stale_bonus",
284
- "0.35",
285
- "--dblock_undertrain_bonus",
286
- "0.25",
287
- "--dblock_activation_offload_min_mb",
288
- "1.0",
289
- "--nat_document_boundary_aware",
290
- "--dblock_fullstack_anchor_deterministic_eval",
291
- "--dblock_fullstack_anchor_spike_retry_limit",
292
- "3",
293
- "--dblock_local_spike_retry_limit",
294
- "3",
295
- "--dblock_fullstack_nat_no_valid_crop_limit",
296
- "8",
297
- "--target_tokens",
298
- "400000000000",
299
- "--lr_schedule_reanchor_on_resume",
300
- "--save_every_sec",
301
- "1800",
302
- "--delta_max_keep",
303
- "0",
304
- "--max_ckpts",
305
- "2",
306
- "--disk_free_floor_gb",
307
- "12.0",
308
- "--heartbeat_every_sec",
309
- "30",
310
- "--dblock_fullstack_ar_weight",
311
- "1.0",
312
- "--dblock_fullstack_sat_weight",
313
- "0.1",
314
- "--dblock_fullstack_nat_weight",
315
- "0.1",
316
- "--dblock_local_aux_warn_ce",
317
- "0.0",
318
- "--dblock_local_aux_max_ce",
319
- "0.0",
320
- "--lr_schedule_origin_tokens",
321
- "85002387456",
322
- "--dblock_stop_after_commits",
323
- "0",
324
- "--no-oom_auto_backoff",
325
- "--require_exact_resume_state",
326
- "--dblock_objective_mode",
327
- "committed_9_1_1",
328
- "--dblock_ar_weight",
329
- "0.25",
330
- "--dblock_sat_weight",
331
- "0.25",
332
- "--dblock_nat_weight",
333
- "0.25",
334
- "--dblock_aux_grad_ratio",
335
- "0.5",
336
- "--dblock_min_ar_share",
337
- "0.8",
338
- "--lr_core",
339
- "2e-7",
340
- "--lr_head",
341
- "2e-7",
342
- "--dblock_checkpoint_stride",
343
- "2",
344
- "--empty_cache_every_steps",
345
- "0",
346
- "--dblock_log_every",
347
- "25",
348
- "--cuda_max_reserved_mib",
349
- "0",
350
- "--cuda_min_free_mib",
351
- "0",
352
- "--repo",
353
- "--repo_retrofit",
354
- "--repo_start_layer",
355
- "9",
356
- "--repo_dim",
357
- "160",
358
- "--repo_theta",
359
- "500000.0",
360
- "--repo_lr",
361
- "1e-5",
362
- "--dblock_clean_context",
363
- "e2e",
364
- "--dblock_local_rows",
365
- "2",
366
- "--dblock_clean_context_noise_scale",
367
- "0.0",
368
- "--dblock_depth_anchor_every",
369
- "50",
370
- "--dblock_fullstack_anchor_cap",
371
- "256",
372
- "--dblock_fullstack_ar_rows",
373
- "1",
374
- "--detachable_50m_upgrade_multimode",
375
- "--detachable_50m_upgrade_logit_distill",
376
- "--detachable_50m_upgrade_sat_gate_v2713",
377
- "--detachable_50m_create_from_legacy",
378
- "--detachable_50m_seed_from_mature",
379
- "--detachable_50m_lr",
380
- "0.0001",
381
- "--detachable_50m_weight_decay",
382
- "0.0",
383
- "--detachable_50m_update_mode",
384
- "sublayer",
385
- "--detachable_50m_batch",
386
- "1",
387
- "--detachable_50m_tokens",
388
- "64",
389
- "--detachable_50m_loss_tokens",
390
- "64",
391
- "--detachable_50m_every",
392
- "1",
393
- "--detachable_50m_teacher_every",
394
- "1",
395
- "--detachable_50m_distill_weight",
396
- "1.0",
397
- "--detachable_50m_logit_distill_alpha",
398
- "0.30",
399
- "--detachable_50m_logit_distill_temperature",
400
- "1.0",
401
- "--detachable_50m_logit_distill_tokens",
402
- "64",
403
- "--detachable_50m_hidden_distill_weight",
404
- "0.0",
405
- "--detachable_50m_anchor_every",
406
- "128",
407
- "--detachable_50m_anchor_tokens",
408
- "32",
409
- "--detachable_50m_seed",
410
- "20260830",
411
- "--detachable_50m_log_every",
412
- "10",
413
- "--detachable_50m_telemetry_keep",
414
- "128",
415
- "--weight_accel_mode",
416
- "shadow",
417
- "--weight_accel_dir",
418
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel",
419
- "--weight_accel_every",
420
- "1",
421
- "--weight_accel_train_every",
422
- "4",
423
- "--weight_accel_record_every",
424
- "14",
425
- "--weight_accel_state_every",
426
- "256",
427
- "--detachable_50m",
428
- "--detachable_50m_optimizer",
429
- "paged_adamw8bit",
430
- "--delta_every_steps",
431
- "0",
432
- "--delta_every_sec",
433
- "0",
434
- "--resume",
435
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt",
436
- "--save_dir",
437
- "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B",
438
- "--data_seed",
439
- "2848730",
440
- "--batch_size",
441
- "56",
442
- "--dblock_blocks",
443
- "14",
444
- "--dblock_direct56_enabled",
445
- "1",
446
- "--dblock_layers_per_update",
447
- "0",
448
- "--dblock_router",
449
- "shadow",
450
- "--dblock_router_blend",
451
- "0.0",
452
- "--dblock_router_full_anchor_every",
453
- "128",
454
- "--dblock_router_train_layer_policy",
455
- "cyclic",
456
- "--dblock_router_train_layers_per_update",
457
- "1",
458
- "--dblock_sublayer_mode",
459
- "off",
460
- "--dblock_target_route_activate_from_checkpoint",
461
- "1",
462
- "--dblock_target_route_activation_checkpoint_step",
463
- "2717509",
464
- "--dblock_target_route_activation_parent_pointer_sha256",
465
- "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
466
- "--dblock_target_route_offset",
467
- "1",
468
- "--dblock_target_route_policy",
469
- "xor_fold56_v1",
470
- "--dblock_train_layer_policy",
471
- "cyclic",
472
- "--dblock_train_layers_per_update",
473
- "1",
474
- "--dblock_train_sublayer_policy",
475
- "cyclic",
476
- "--dblock_train_sublayers_per_update",
477
- "1",
478
- "--granularity_label",
479
- "one-sublayer-local-grad-direct56-teacher-v27.16",
480
- "--oom_known_safe_batch",
481
- "56",
482
- "--val_every_sec",
483
- "600",
484
- "--val_file",
485
- "/workspace/agillm43_heldout_v3/token_ids.json",
486
- "--val_sha256",
487
- "513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21"
488
- ],
489
- "argv_string": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
490
- "cwd": "/workspace/agillm44_throughput_80k_20260910",
491
- "env": {
492
- "AGILLM43_DBLOCK_CC_NOISE_SCALE": "0.0",
493
- "AGILLM43_DBLOCK_CLEAN_CONTEXT": "e2e",
494
- "AGILLM43_DBLOCK_CLEAN_CONTEXT_NOISE_SCALE": "0.0",
495
- "AGILLM43_DBLOCK_DEPTH_ANCHOR_EVERY": "50",
496
- "AGILLM43_DBLOCK_FULLSTACK_ANCHOR_CAP": "256",
497
- "AGILLM43_DBLOCK_FULLSTACK_AR_ROWS": "1",
498
- "AGILLM43_DBLOCK_LOCAL_ROWS": "2",
499
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATE_FROM_CHECKPOINT": "1",
500
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_CHECKPOINT_STEP": "2717509",
501
- "AGILLM43_DBLOCK_TARGET_ROUTE_ACTIVATION_PARENT_POINTER_SHA256": "04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827",
502
- "AGILLM43_DBLOCK_TARGET_ROUTE_OFFSET": "1",
503
- "AGILLM43_DBLOCK_TARGET_ROUTE_POLICY": "xor_fold56_v1",
504
- "AGILLM43_DECOMPRESS_CACHE": "0",
505
- "AGILLM43_DETACHABLE50M_ANCHOR_EVERY": "128",
506
- "AGILLM43_DETACHABLE50M_ANCHOR_TOKENS": "<redacted>",
507
- "AGILLM43_DETACHABLE50M_BATCH": "1",
508
- "AGILLM43_DETACHABLE50M_CREATE_FROM_LEGACY": "1",
509
- "AGILLM43_DETACHABLE50M_DISTILL_WEIGHT": "1.0",
510
- "AGILLM43_DETACHABLE50M_ENABLED": "1",
511
- "AGILLM43_DETACHABLE50M_EVERY": "1",
512
- "AGILLM43_DETACHABLE50M_GATE_V2713_START_CHECKPOINT_SHA256": "39f442fc0e6f9fad6f875e067f63db67f33a41a53e8cbc49ce1591010bb4aec3",
513
- "AGILLM43_DETACHABLE50M_HIDDEN_DISTILL_WEIGHT": "0.0",
514
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_ALPHA": "0.30",
515
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TEMPERATURE": "1.0",
516
- "AGILLM43_DETACHABLE50M_LOGIT_DISTILL_TOKENS": "<redacted>",
517
- "AGILLM43_DETACHABLE50M_LOG_EVERY": "10",
518
- "AGILLM43_DETACHABLE50M_LOSS_TOKENS": "<redacted>",
519
- "AGILLM43_DETACHABLE50M_LR": "0.0001",
520
- "AGILLM43_DETACHABLE50M_OPTIMIZER": "paged_adamw8bit",
521
- "AGILLM43_DETACHABLE50M_SEED": "20260830",
522
- "AGILLM43_DETACHABLE50M_SEED_FROM_MATURE": "1",
523
- "AGILLM43_DETACHABLE50M_TEACHER_EVERY": "1",
524
- "AGILLM43_DETACHABLE50M_TELEMETRY_KEEP": "128",
525
- "AGILLM43_DETACHABLE50M_TOKENS": "<redacted>",
526
- "AGILLM43_DETACHABLE50M_UPDATE_MODE": "sublayer",
527
- "AGILLM43_DETACHABLE50M_UPGRADE_LOGIT_DISTILL": "1",
528
- "AGILLM43_DETACHABLE50M_UPGRADE_MULTIMODE": "1",
529
- "AGILLM43_DETACHABLE50M_UPGRADE_SAT_GATE_V2713": "1",
530
- "AGILLM43_DETACHABLE50M_WEIGHT_DECAY": "0.0",
531
- "AGILLM43_REPO_DIM": "160",
532
- "AGILLM43_REPO_ENABLED": "1",
533
- "AGILLM43_REPO_LR": "1e-5",
534
- "AGILLM43_REPO_RETROFIT": "1",
535
- "AGILLM43_REPO_START_LAYER": "9",
536
- "AGILLM43_REPO_THETA": "500000.0",
537
- "AGILLM43_SEQUENCE_PREFETCH_DEPTH": "256",
538
- "AGILLM43_SINGLEFILE_PROFILE": "intelligence-v25-satvar-pretrain",
539
- "AGILLM43_SUPERVISE_SET_ARGS_JSON": "{\"--batch_size\":\"56\",\"--dblock_blocks\":\"14\",\"--dblock_direct56_enabled\":\"1\",\"--dblock_layers_per_update\":\"0\",\"--dblock_router\":\"shadow\",\"--dblock_router_blend\":\"0.0\",\"--dblock_router_full_anchor_every\":\"128\",\"--dblock_router_train_layer_policy\":\"cyclic\",\"--dblock_router_train_layers_per_update\":\"1\",\"--dblock_sublayer_mode\":\"off\",\"--dblock_target_route_activate_from_checkpoint\":\"1\",\"--dblock_target_route_activation_checkpoint_step\":\"2717509\",\"--dblock_target_route_activation_parent_pointer_sha256\":\"04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827\",\"--dblock_target_route_offset\":\"1\",\"--dblock_target_route_policy\":\"xor_fold56_v1\",\"--dblock_train_layer_policy\":\"cyclic\",\"--dblock_train_layers_per_update\":\"1\",\"--dblock_train_sublayer_policy\":\"cyclic\",\"--dblock_train_sublayers_per_update\":\"1\",\"--granularity_label\":\"one-sublayer-local-grad-direct56-teacher-v27.16\",\"--oom_known_safe_batch\":\"56\",\"--val_every_sec\":\"600\",\"--val_file\":\"/workspace/agillm43_heldout_v3/token_ids.json\",\"--val_sha256\":\"513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21\"}",
540
- "AGILLM43_WEIGHT_ACCEL_MODE": "shadow",
541
- "AGILLM_DATASETS_STREAMING_READ_MAX_RETRIES": "2",
542
- "AGILLM_DATASETS_STREAMING_READ_RETRY_INTERVAL": "1",
543
- "AGILLM_DATASET_AGENT_ROUTER": "0",
544
- "AGILLM_DATASET_HOTLOAD_CONFIG": "/workspace/agillm43_dataset_hotload.json",
545
- "AGILLM_DATASET_HOT_RELOAD_CHECK_TOKENS": "<redacted>",
546
- "AGILLM_DATASET_HOT_RELOAD_SEC": "5",
547
- "AGILLM_DATASET_NN_ROUTER": "1",
548
- "AGILLM_DATASET_ROUTER_BLEND": "0.25",
549
- "AGILLM_DATASET_ROUTER_EXPLORE": "0.03",
550
- "AGILLM_DATASET_ROUTER_LR": "0.005",
551
- "AGILLM_DATASET_ROUTER_MIN_SCORE": "0.15",
552
- "AGILLM_DATASET_ROUTER_SHARPNESS": "2.0",
553
- "AGILLM_ENABLE_CUPY": "0",
554
- "AGILLM_HF_HUB_DOWNLOAD_TIMEOUT": "30",
555
- "AGILLM_HF_TRANSPORT": "geth_hf_relay",
556
- "AGILLM_HOT_CONFIG": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
557
- "AGILLM_HOT_CONFIG_SHA256": "",
558
- "AGILLM_MAX_EXAMPLE_CHARS": "32768",
559
- "AGILLM_MAX_EXAMPLE_TOKENS": "<redacted>",
560
- "AGILLM_PARENT_POINTER_SHA256": "443aed3f7eb17df42d0840fe208c735ab58671bebb26c99be4c80d96c28d756c",
561
- "AGILLM_REMOTE_SHARDS_PER_ITER": "4",
562
- "AGILLM_RUN_ID": "singlefile_intelligence_v25_20260911T095413Z_from_step000002848688",
563
- "AGILLM_STREAM_NEXT_TIMEOUT_SEC": "60",
564
- "AGILLM_STREAM_OPEN_TIMEOUT_SEC": "120",
565
- "AGILLM_STREAM_SOCKET_TIMEOUT_SEC": "30",
566
- "AGILLM_STREAM_SOURCE_FATAL_COOLDOWN_SEC": "600",
567
- "AGILLM_STREAM_SOURCE_MAX_COOLDOWN_SEC": "120",
568
- "AGILLM_SYNTHETIC_TOKENIZER": "<redacted>",
569
- "CUDA_MODULE_LOADING": "LAZY",
570
- "CUDA_VERSION": "12.1.1",
571
- "CUDA_VISIBLE_DEVICES": "0",
572
- "HF_HUB_DOWNLOAD_TIMEOUT": "30",
573
- "HF_HUB_ETAG_TIMEOUT": "10",
574
- "PYTORCH_CUDA_ALLOC_CONF": "backend:cudaMallocAsync",
575
- "PYTORCH_VERSION": "2.2.1"
576
- },
577
- "parent_proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py supervise --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --log /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/logs/singlefile_intelligence_v25_satvar_pretrain.log --sleep_sec 15",
578
- "pid": 1518633,
579
- "ppid": 1518567,
580
- "proc_cmdline": "/opt/conda/bin/python -u /workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
581
- "schema": "agillm.launch.v1"
582
- },
583
- "local_step": 2849585,
584
- "loss": 5.471035957336426,
585
- "nat_mask_contract": "agillm.nat-mask.v1",
586
- "nat_mask_contract_source": "explicit-legacy-recovery",
587
- "nat_mask_eos_id": 1,
588
- "nat_mask_legacy_blank_id": 1,
589
- "nat_mask_migrated_from_id": 1,
590
- "nat_mask_schema_version": 1,
591
- "nat_mask_token_id": 2,
592
- "pid": 1518633,
593
- "schema_version": 1,
594
- "seen_tok": 283698599936,
595
- "source_is_lane_checkpoint": true,
596
- "source_is_master_checkpoint": false,
597
- "source_is_non_master_checkpoint": false,
598
- "step": 2849585,
599
- "train_argv": "/workspace/agillm44_throughput_80k_20260910/runtime_retain_activations_v2.py train --preset agillm4_floor --block 2048 --source HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25 --nat_mask_token_id 2 --tie_weights --tie_kv --moe_ffn --moe_experts 2 --moe_top_k 1 --moe_mlp_mult 4 --moe_shared_experts 1 --moe_shared_mlp_mult 2 --moe_aux_coef 0.01 --moe_z_coef 0.001 --amp --grad_checkpoint --attn_backend sdpa --sublinear_window 128 --sublinear_stride 128 --sublinear_max_anchors 128 --sublinear_chunk 128 --sublinear_sinks 4 --sublinear_recent_anchors 64 --no-sublinear_pooled_landmarks --anchor_stride 256 --anchor_max 2048 --anchor_position -1 --alibi_mode corrected --alibi_scale 0.0 --optimizer adamw8bit --weight_decay 0.0 --lr_decay cosine --lr_decay_tokens 49000954880 --lr_min_mult 0.2 --lr_schedule_reset_on_resume --lr_warmup_tokens 10000000 --lr_warmup_min_mult 0.1 --dblock --dblock_schedule roundrobin --dblock_router_ramp_steps 64 --dblock_warmup_steps 14 --dblock_checkpoint_skip_tail 0 --dblock_loop_layers 0 --dblock_loop_start 0 --dblock_loop_cond_scale 1.0 --dblock_sigma_curriculum_steps 2000 --dblock_sigma_sampling lognormal --dblock_sigma_stratified --dblock_sigma_min 0.002 --dblock_sigma_max 2.0 --dblock_sigma_pmean -1.2 --dblock_sigma_pstd 1.2 --dblock_edm_wmax 5.0 --dblock_ar_prob 0.8 --dblock_sat_prob 0.1 --dblock_nat_prob 0.1 --dblock_ar_loss_tokens 8192 --dblock_sat_loss_tokens 4096 --dblock_nat_loss_tokens 4096 --dblock_nat_embed_noise_mode off --dblock_nat_embed_noise_scale 1.0 --nat_loss_weight 1.0 --nat_mask_ratio 0.5 --nat_max_tokens 0 --nat_span_mask_prob 0.35 --nat_suffix_mask_prob 0.20 --nat_span_max_tokens 0 --loss_spike_skip 0.0 --sat_every 1 --nat_every 1 --dblock_fullstack_anchor_spike_skip 0.0 --dblock_fullstack_anchor_max_ce 0.0 --dblock_anchor_softcap_mult 4.0 --dblock_anchor_softcap_min_ce 25.0 --dblock_anchor_softcap_max_ce 50.0 --dblock_fullstack_ar_every 3 --dblock_fullstack_ar_offset 0 --dblock_fullstack_ar_tokens 256 --dblock_fullstack_sat_every 3 --dblock_fullstack_sat_offset 1 --dblock_fullstack_sat_tokens 128 --dblock_fullstack_nat_every 3 --dblock_fullstack_nat_offset 2 --dblock_fullstack_nat_tokens 128 --dblock_fullstack_nat_mask_id -1 --ckpt_codec block-sharded-zstd --async_update_every_steps 0 --dblock_router_hidden 64 --dblock_router_heads 4 --dblock_router_layers 2 --dblock_router_lr 0.0005 --dblock_explore 0.08 --dblock_max_stale_steps 64 --dblock_max_count_skew 1.35 --dblock_stale_bonus 0.35 --dblock_undertrain_bonus 0.25 --dblock_activation_offload_min_mb 1.0 --nat_document_boundary_aware --dblock_fullstack_anchor_deterministic_eval --dblock_fullstack_anchor_spike_retry_limit 3 --dblock_local_spike_retry_limit 3 --dblock_fullstack_nat_no_valid_crop_limit 8 --target_tokens 400000000000 --lr_schedule_reanchor_on_resume --save_every_sec 1800 --delta_max_keep 0 --max_ckpts 2 --disk_free_floor_gb 12.0 --heartbeat_every_sec 30 --dblock_fullstack_ar_weight 1.0 --dblock_fullstack_sat_weight 0.1 --dblock_fullstack_nat_weight 0.1 --dblock_local_aux_warn_ce 0.0 --dblock_local_aux_max_ce 0.0 --lr_schedule_origin_tokens 85002387456 --dblock_stop_after_commits 0 --no-oom_auto_backoff --require_exact_resume_state --dblock_objective_mode committed_9_1_1 --dblock_ar_weight 0.25 --dblock_sat_weight 0.25 --dblock_nat_weight 0.25 --dblock_aux_grad_ratio 0.5 --dblock_min_ar_share 0.8 --lr_core 2e-7 --lr_head 2e-7 --dblock_checkpoint_stride 2 --empty_cache_every_steps 0 --dblock_log_every 25 --cuda_max_reserved_mib 0 --cuda_min_free_mib 0 --repo --repo_retrofit --repo_start_layer 9 --repo_dim 160 --repo_theta 500000.0 --repo_lr 1e-5 --dblock_clean_context e2e --dblock_local_rows 2 --dblock_clean_context_noise_scale 0.0 --dblock_depth_anchor_every 50 --dblock_fullstack_anchor_cap 256 --dblock_fullstack_ar_rows 1 --detachable_50m_upgrade_multimode --detachable_50m_upgrade_logit_distill --detachable_50m_upgrade_sat_gate_v2713 --detachable_50m_create_from_legacy --detachable_50m_seed_from_mature --detachable_50m_lr 0.0001 --detachable_50m_weight_decay 0.0 --detachable_50m_update_mode sublayer --detachable_50m_batch 1 --detachable_50m_tokens 64 --detachable_50m_loss_tokens 64 --detachable_50m_every 1 --detachable_50m_teacher_every 1 --detachable_50m_distill_weight 1.0 --detachable_50m_logit_distill_alpha 0.30 --detachable_50m_logit_distill_temperature 1.0 --detachable_50m_logit_distill_tokens 64 --detachable_50m_hidden_distill_weight 0.0 --detachable_50m_anchor_every 128 --detachable_50m_anchor_tokens 32 --detachable_50m_seed 20260830 --detachable_50m_log_every 10 --detachable_50m_telemetry_keep 128 --weight_accel_mode shadow --weight_accel_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/weight_trajectory_accel --weight_accel_every 1 --weight_accel_train_every 4 --weight_accel_record_every 14 --weight_accel_state_every 256 --detachable_50m --detachable_50m_optimizer paged_adamw8bit --delta_every_steps 0 --delta_every_sec 0 --resume /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt --save_dir /workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B --data_seed 2848730 --batch_size 56 --dblock_blocks 14 --dblock_direct56_enabled 1 --dblock_layers_per_update 0 --dblock_router shadow --dblock_router_blend 0.0 --dblock_router_full_anchor_every 128 --dblock_router_train_layer_policy cyclic --dblock_router_train_layers_per_update 1 --dblock_sublayer_mode off --dblock_target_route_activate_from_checkpoint 1 --dblock_target_route_activation_checkpoint_step 2717509 --dblock_target_route_activation_parent_pointer_sha256 04638cfbc19c0f9db6e6dfa03b2509d50a0bf3c56048a2cbd6e67154bc85c827 --dblock_target_route_offset 1 --dblock_target_route_policy xor_fold56_v1 --dblock_train_layer_policy cyclic --dblock_train_layers_per_update 1 --dblock_train_sublayer_policy cyclic --dblock_train_sublayers_per_update 1 --granularity_label one-sublayer-local-grad-direct56-teacher-v27.16 --oom_known_safe_batch 56 --val_every_sec 600 --val_file /workspace/agillm43_heldout_v3/token_ids.json --val_sha256 513b30203ffd18167655deb2d9cd610541df87d1ea0b8a0f9241c58e68703d21",
600
- "train_script": "runtime_retain_activations_v2.py",
601
- "train_script_sha256": "989a7ee5e4344b7b53b67b70d0df27dc42461d864084283d1d7acb262e0a7081",
602
- "warmstart_base_seen_tok": 0,
603
- "warmstart_base_step": 2848688,
604
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
605
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt",
606
- "warmstart_source_step": 2848688
607
- },
608
- "alibi_mode": "corrected",
609
- "alibi_scale": 0.0,
610
- "checkpoint_summary": "warmstarted_from_lane_checkpoint; source=/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt; path=/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
611
- "dataset_provenance": {
612
- "chat": false,
613
- "dataset_field_text": "text",
614
- "dataset_hotload_config_path": "/workspace/agillm43_dataset_hotload.json",
615
- "dataset_hotload_config_sha256": "3f48fe33b475c719d43e14cc4d0867c19e17dfa1d2cb304cd291eea6a879a7c8",
616
- "dataset_hotload_config_stamp": [
617
- 50,
618
- 1124990893,
619
- 1789091829464680808,
620
- 7210
621
- ],
622
- "dataset_hotload_config_used": true,
623
- "dataset_hotload_enabled": true,
624
- "dataset_hotload_generation": 13,
625
- "dataset_hotload_schema": "agillm.dataset_hotload.config.v2",
626
- "dataset_revision_map_sha256": "8361995a504a2365136b20e1a879ba78944a06309f13f3f013bbd0cdfc72225b",
627
- "dataset_revisions": {
628
- "EleutherAI/proof-pile-2": "901a9273a770e9d4138c5ddd91802f9c5c6cdc4b",
629
- "HuggingFaceFW/fineweb": "9bb295ddab0e05d785b879661af7260fed5140fc",
630
- "HuggingFaceFW/fineweb-edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
631
- "HuggingFaceTB/cosmopedia": "0ae6ec63f91742bd2d1eaef4f02232c55d719385",
632
- "HuggingFaceTB/finemath": "e92b25a616738fe95dc186b64dfb19f9c8525594",
633
- "HuggingFaceTB/smollm-corpus": "3ba9d605774198c5868892d7a8deda78031a781f",
634
- "Skylion007/openwebtext": "79d93d786212f7344586290adb811d4ae6a1762c",
635
- "allenai/c4": "1588ec454efa1a09f29cd18ddd04fe05fc8653a2",
636
- "allenai/dolma": "7f48140530a023e9ea4c5cfb141160922727d4d3",
637
- "allenai/dolmino-mix-1124": "a319f19eef1e257417b11ea8c30da266ae175557",
638
- "codeparrot/codeparrot-clean": "35a59fb025bc0a102f7d96eac09d145b896d487b",
639
- "mlfoundations/dclm-baseline-1.0": "a3b142c183aebe5af344955ae20836eb34dcf69b",
640
- "wikimedia/wikipedia": "b04c8d1ceb2f5cd4588862100d08de323dccfbaa"
641
- },
642
- "dolma_payload_integrity": "external_url_manifest_pinned_payload_hashes_unavailable",
643
- "dolma_url_list_sha256": "bc6ef505be920c94bb9e7147445638f606fdc3f5af622cd1d887dc48e21a2d5b",
644
- "dolma_url_manifest_raw_sha256": "6fe8c40d51e5bebc9303453d7f533b232d69f7ae122123621a82621b1cd779b7",
645
- "has_language_mix": true,
646
- "has_numeracy": false,
647
- "hot_config_expected_sha256": "",
648
- "hot_config_has_append": true,
649
- "hot_config_has_datasets": false,
650
- "hot_config_mtime": 1789092253.2119184,
651
- "hot_config_path": "/workspace/agillm43_production_v14/contracts/hot_config_owner_v25.json",
652
- "hot_config_used": true,
653
- "hot_reload_capable": true,
654
- "phase": "pretrain",
655
- "proof_pile_lfs_manifest_sha256": "d020dfb796b02a75564e4e11d9e03185a55b3f9c6639f87830cfbe10737ed329",
656
- "proof_pile_paths_sha256": "8a30018017e3b3d9a9bcfa06a3c404ccd113297b0b1d0c48e2e2aadf113e52b0",
657
- "schema": "agillm.dataset_provenance.v1",
658
- "sequence_prefetch_depth": 256,
659
- "source_count": 23,
660
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
661
- "source_requested": "HuggingFaceFW/fineweb-edu:sample-10BT|1.60,HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.50,wikimedia/wikipedia:20231101.en|1.10,allenai/c4:en|0.40,Skylion007/openwebtext|0.35,HuggingFaceTB/cosmopedia:web_samples_v2|1.30,HuggingFaceTB/smollm-corpus:cosmopedia-v2|1.30,EleutherAI/proof-pile-2:all|1.50,allenai/dolma:v1_6-sample|0.50,codeparrot/codeparrot-clean|1.25",
662
- "source_specs": [
663
- "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96",
664
- "wikimedia/wikipedia:20231101.en|1.63",
665
- "allenai/c4:en|0.36",
666
- "Skylion007/openwebtext|0.18",
667
- "HuggingFaceTB/cosmopedia:web_samples_v2|0.64",
668
- "HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65",
669
- "EleutherAI/proof-pile-2:all|0.67",
670
- "allenai/dolma:v1_6-sample|0.5",
671
- "codeparrot/codeparrot-clean|0.24",
672
- "HuggingFaceTB/finemath:finemath-4plus|0.32",
673
- "HuggingFaceTB/finemath:infiwebmath-4plus|0.23",
674
- "HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05",
675
- "allenai/dolmino-mix-1124|0.78",
676
- "mlfoundations/dclm-baseline-1.0|1.05",
677
- "HuggingFaceFW/fineweb-edu:sample-100BT|2.4",
678
- "json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03",
679
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45",
680
- "json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35",
681
- "json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35",
682
- "json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35",
683
- "json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2",
684
- "json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2",
685
- "json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35"
686
- ],
687
- "val_source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
688
- "val_source_requested": ""
689
- },
690
- "lr_schedule_origin_tok": 85002387456,
691
- "nat_mask_contract": "agillm.nat-mask.v1",
692
- "nat_mask_contract_source": "explicit-legacy-recovery",
693
- "nat_mask_eos_id": 1,
694
- "nat_mask_legacy_blank_id": 1,
695
- "nat_mask_migrated_from_id": 1,
696
- "nat_mask_schema_version": 1,
697
- "nat_mask_token_id": 2,
698
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
699
- "promotion_required": true,
700
- "quality_gate_used_for_training": false,
701
- "repair_schema": "",
702
- "saved_at": "2026-09-11T10:33:17Z",
703
- "schema": "agillm43.serving.candidate.v1",
704
- "source_effective": "HuggingFaceFW/fineweb:CC-MAIN-2024-10|0.96,wikimedia/wikipedia:20231101.en|1.63,allenai/c4:en|0.36,Skylion007/openwebtext|0.18,HuggingFaceTB/cosmopedia:web_samples_v2|0.64,HuggingFaceTB/smollm-corpus:cosmopedia-v2|0.65,EleutherAI/proof-pile-2:all|0.67,allenai/dolma:v1_6-sample|0.5,codeparrot/codeparrot-clean|0.24,HuggingFaceTB/finemath:finemath-4plus|0.32,HuggingFaceTB/finemath:infiwebmath-4plus|0.23,HuggingFaceTB/smollm-corpus:fineweb-edu-dedup|1.05,allenai/dolmino-mix-1124|0.78,mlfoundations/dclm-baseline-1.0|1.05,HuggingFaceFW/fineweb-edu:sample-100BT|2.4,json:/workspace/agillm_wikitext103_raw_docs/train.jsonl|0.03,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb.jsonl|0.45,json:/workspace/agillm44_ultradata_20260911/ultra_fineweb_l3.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_code.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_math.jsonl|0.35,json:/workspace/agillm44_ultradata_20260911/ultradata_rl.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultradata_sft_agent.jsonl|0.2,json:/workspace/agillm44_ultradata_20260911/ultrax.jsonl|0.35",
705
- "step": 2849585,
706
- "training_pointer_schema": "agillm43.training.resume.v2",
707
- "training_pointer_sha256": "66f9b811a8f2523946838a842cb05ab978e5b961bcc362960a372368720b760f",
708
- "warmstart_kind": "warmstarted_from_lane_checkpoint",
709
- "warmstart_source_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02848688_from02848427_20260911T093525_1789119325602978099_1470897_000001.pt"
710
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
checkpoints/step2849585_20260911/training_latest.json DELETED
@@ -1,455 +0,0 @@
1
- {
2
- "checkpoint_main_nbytes": 18517715,
3
- "checkpoint_main_sha256": "970938cd7df01e92858376340c3857f3df880752625c45f323cae3c3ac7c757c",
4
- "checkpoint_name": "pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
5
- "checkpoint_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
6
- "continuation_state_schema": "agillm43.continuation.state.v2",
7
- "continuation_state_sha256": "c2b1f7393a1d5ab5f5cb40c2ce8bad34bdbd81b14bc32d1c8d569e977731c2a3",
8
- "dblock_committed_step": 1119234,
9
- "dblock_target_route_clock": 132076,
10
- "dblock_target_route_controller_source_sha256": "714ee71c440ba9e9699034d7ea5f1de6f73d65cc80caf9f9350cf30f11a33e28",
11
- "dblock_target_route_last_committed_target": {
12
- "attempts": 1,
13
- "block": 0,
14
- "canonical_index": 28,
15
- "committed_step_after": 1119234,
16
- "first_attempt_step": 132075,
17
- "k": 0,
18
- "last_attempt_step": 132075,
19
- "layer_offset": 0,
20
- "phase": 2,
21
- "physical_layer": 0,
22
- "q": 2,
23
- "route_commit_clock": 132075,
24
- "route_offset": 1,
25
- "selection": {
26
- "authority": "learned",
27
- "authority_granted": true,
28
- "baseline_prediction": 5.971238519960543,
29
- "candidate_hash": "7366174b2e1fbeb670854a846f1be186bf4933755de5e3bcaf7aa6b0bd537baa",
30
- "candidate_persisted": true,
31
- "clock": 132075,
32
- "fault_count": 0,
33
- "guard_override": false,
34
- "guard_reason": null,
35
- "nominated": true,
36
- "override": false,
37
- "phase": 28,
38
- "raw_prediction": 2,
39
- "reason": "candidate_scheduled",
40
- "schema": "agillm44.direct56.controller.v1",
41
- "selected": 2,
42
- "selected_target": 2,
43
- "shadow_override": false,
44
- "stage": "learned_full",
45
- "teacher_ce": 0.00019333878299221396,
46
- "teacher_target": 2,
47
- "teacher_training_receipts": 51379,
48
- "teacher_value_prediction": 5.406079292297363,
49
- "validation_samples": 1427,
50
- "value_prediction": 5.406079292297363
51
- },
52
- "sublayer": "ffn",
53
- "target_id": 2,
54
- "target_sha256": "5b41e93a7940002979e182700d1035d8f373dccece004a0c846a263b09f4c3ca",
55
- "teacher_target_id": 2
56
- },
57
- "dblock_target_route_offset": 1,
58
- "dblock_target_route_payload_sha256": "cd86ec21182bb8c2692ecc0f67696b8acc7e16cdb1cf749f5d835b8702ee4a46",
59
- "dblock_target_route_pending_target": null,
60
- "dblock_target_route_policy_id": "guarded-direct56-route-equiv56-v2",
61
- "dblock_target_route_policy_sha256": "d4241a9c3a0c319cab6b268ea3792277c2cf7088b8d11492e459d2e58e2bde47",
62
- "dblock_target_route_schema": "agillm43.dblock.target-route.direct56.v1",
63
- "dblock_target_route_target_map_sha256": "31e64396588ebf72dd6aa20de6cbea14c20b6b0823269fade6f7471d37cc0836",
64
- "dblock_target_route_topology_sha256": "c3b1679046220655820501bb30dc2733efc0628f661273ac3f2217355bdc2102",
65
- "direct56_checkpointed": true,
66
- "direct56_payload_digest": "4db89ead4362f7cbd41d1eabc6c4872c4076afaaae837c83827f6c805f93bef3",
67
- "direct56_schema": "agillm44.direct56.controller.v1",
68
- "package_format": "block-sharded",
69
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt",
70
- "quality_gate_used_for_training": false,
71
- "record_sha256": "66f9b811a8f2523946838a842cb05ab978e5b961bcc362960a372368720b760f",
72
- "role": "training",
73
- "saved_at": "2026-09-11T10:33:17Z",
74
- "schema": "agillm43.training.resume.v2",
75
- "seen_tok": 283698599936,
76
- "shard_dir": "pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards",
77
- "shards": [
78
- {
79
- "name": "core_base.pt",
80
- "nbytes": 608663037,
81
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_base.pt",
82
- "sha256": "60daef3bc6dde54646694739bc9bf6bc3411cf34c14de7e5470ae47bfc7e5d48"
83
- },
84
- {
85
- "name": "core_block_0000.pt",
86
- "nbytes": 139944140,
87
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0000.pt",
88
- "sha256": "792bfdd583cb09f2be93cbd5988d50cafd5e7b1d257f11ca266e64918c065b87"
89
- },
90
- {
91
- "name": "core_block_0001.pt",
92
- "nbytes": 140211298,
93
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0001.pt",
94
- "sha256": "19cf90ce1266861e143bcaf779cbd418cc7e9378cbd3de59e2e83612b03acbdf"
95
- },
96
- {
97
- "name": "core_block_0002.pt",
98
- "nbytes": 140183714,
99
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0002.pt",
100
- "sha256": "1d18ec841348da4597ed3f2996978990cfd4f7da25c19a15e3cd2eb0641b6f43"
101
- },
102
- {
103
- "name": "core_block_0003.pt",
104
- "nbytes": 140171674,
105
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0003.pt",
106
- "sha256": "42ddc24cae803f66294f5dd30de8df57bc3064fbcc16901384d95c86cffdc622"
107
- },
108
- {
109
- "name": "core_block_0004.pt",
110
- "nbytes": 140325746,
111
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0004.pt",
112
- "sha256": "d63a8a531d99ebc365921781de6331f564a685113c89acfa8265e48f7fccfc82"
113
- },
114
- {
115
- "name": "core_block_0005.pt",
116
- "nbytes": 140269868,
117
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0005.pt",
118
- "sha256": "7d9e53192d9c8da591e6c03ac89d5f5bc694a0d4485c12438d4613c04f19f092"
119
- },
120
- {
121
- "name": "core_block_0006.pt",
122
- "nbytes": 140624553,
123
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0006.pt",
124
- "sha256": "63d39b314374e9bac5656ed67ecab431d1998c7944bdce15aba070e26e64def8"
125
- },
126
- {
127
- "name": "core_block_0007.pt",
128
- "nbytes": 140776857,
129
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0007.pt",
130
- "sha256": "52a417e3304237caba36f07abdf0f7659130218f2414cad8ea1cf01464db1f12"
131
- },
132
- {
133
- "name": "core_block_0008.pt",
134
- "nbytes": 140654988,
135
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0008.pt",
136
- "sha256": "88e91dc652d1666b40773ccd652a479d7a1028a4649733f7bd4abec6533d3df9"
137
- },
138
- {
139
- "name": "core_block_0009.pt",
140
- "nbytes": 141992832,
141
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0009.pt",
142
- "sha256": "26d62ff82615fa2e080f15ced53e48231fa6380dced48f9970192350f4fec2d3"
143
- },
144
- {
145
- "name": "core_block_0010.pt",
146
- "nbytes": 142017741,
147
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0010.pt",
148
- "sha256": "d665c2e120311b98df122ad5d7cad000000f78bf9f9392f3ab1902d08e989859"
149
- },
150
- {
151
- "name": "core_block_0011.pt",
152
- "nbytes": 142101449,
153
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0011.pt",
154
- "sha256": "a0a55a243048a3b43f0aceebc05ff7acb61fdbaa9edc63166d25a59066ce4e1d"
155
- },
156
- {
157
- "name": "core_block_0012.pt",
158
- "nbytes": 142112917,
159
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0012.pt",
160
- "sha256": "7b7e07a78a22b41c318c00ec82942422a6b03ff50af0ab5a3e6027be038bb8f0"
161
- },
162
- {
163
- "name": "core_block_0013.pt",
164
- "nbytes": 141915734,
165
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0013.pt",
166
- "sha256": "c4970d3014ec192c166fd7a9941840f133dabbc5b916626703cc28dea4c5c912"
167
- },
168
- {
169
- "name": "core_block_0014.pt",
170
- "nbytes": 141989992,
171
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0014.pt",
172
- "sha256": "c982636c1b5c8bccd91094514bb9c93d139c220cd7ae2573c099851852ff2f1d"
173
- },
174
- {
175
- "name": "core_block_0015.pt",
176
- "nbytes": 141728794,
177
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0015.pt",
178
- "sha256": "a385957f2ec2fe59a8995a8678101b34db821790611b2430f21919254f51bf87"
179
- },
180
- {
181
- "name": "core_block_0016.pt",
182
- "nbytes": 141997786,
183
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0016.pt",
184
- "sha256": "b02f98e6e1418108bc98601e7cf1b308854d883d632615685c3e6e6891c7bd7e"
185
- },
186
- {
187
- "name": "core_block_0017.pt",
188
- "nbytes": 141602311,
189
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0017.pt",
190
- "sha256": "30465d73895e8df51fe588d152436ba2ff8c737c266db9010c9f07e8510762a5"
191
- },
192
- {
193
- "name": "core_block_0018.pt",
194
- "nbytes": 141938667,
195
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0018.pt",
196
- "sha256": "6cb548dc4d6967541d8fe9cbee0b7028d8a2bac19348731afd871953f5fbbad7"
197
- },
198
- {
199
- "name": "core_block_0019.pt",
200
- "nbytes": 141783169,
201
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0019.pt",
202
- "sha256": "1682991a54e3e3e1dcaf27ea35e87a77ed836355290a72379b8658f7d19b2b4b"
203
- },
204
- {
205
- "name": "core_block_0020.pt",
206
- "nbytes": 141860617,
207
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0020.pt",
208
- "sha256": "cfe534aade222913b147f7b1477a7a1b62424fc55c7703ac9f41bd7e9102f4e1"
209
- },
210
- {
211
- "name": "core_block_0021.pt",
212
- "nbytes": 141671954,
213
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0021.pt",
214
- "sha256": "2315adbcc66246a89b221f163e6a52d788e0df4b9bce827a65333d6e01870cde"
215
- },
216
- {
217
- "name": "core_block_0022.pt",
218
- "nbytes": 142032377,
219
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0022.pt",
220
- "sha256": "78abe836a272c47dd4b1444f8b33dc7d2c3be508ccb2883bb1dfcd7b8f6b9759"
221
- },
222
- {
223
- "name": "core_block_0023.pt",
224
- "nbytes": 142280637,
225
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0023.pt",
226
- "sha256": "77b3fb20cad562bce87e0ea77061b1c475f7f92d3dc7f311f1d988d93880697b"
227
- },
228
- {
229
- "name": "core_block_0024.pt",
230
- "nbytes": 141197067,
231
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0024.pt",
232
- "sha256": "69eb6d648ee7d8c21753a8490f83cbfbe98c9ac631be93edcbc199b98ec0f6a3"
233
- },
234
- {
235
- "name": "core_block_0025.pt",
236
- "nbytes": 141223220,
237
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0025.pt",
238
- "sha256": "d2a59861e03181b11ff8bd6c195ec300ad0826361bcabe5ce4c37cacce359092"
239
- },
240
- {
241
- "name": "core_block_0026.pt",
242
- "nbytes": 141158785,
243
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0026.pt",
244
- "sha256": "f2d39872fce4ee9e418e60f40a60c5c2a2131a685ff9367f7bbc857d3cd0aa57"
245
- },
246
- {
247
- "name": "core_block_0027.pt",
248
- "nbytes": 141339982,
249
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/core_block_0027.pt",
250
- "sha256": "15f6fc676189f49ab442cfa78532db4265a955a3d9dcbb56de29e9813edb049d"
251
- },
252
- {
253
- "name": "detachable_50m_base.pt",
254
- "nbytes": 122164427,
255
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_base.pt",
256
- "sha256": "b897f22c4fd6dbf8677b74289d61ee8032a224f2ef445dffe5379f0e8883fcb8"
257
- },
258
- {
259
- "name": "detachable_50m_block_0000.pt",
260
- "nbytes": 2721180,
261
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0000.pt",
262
- "sha256": "28ed8fac4f86e0c91230a05f3154cfd032a724abc296322f45cd41cb7dda6ce4"
263
- },
264
- {
265
- "name": "detachable_50m_block_0001.pt",
266
- "nbytes": 2699959,
267
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0001.pt",
268
- "sha256": "408a23e011453032e6d402a518d422e0554e0d89b73090428f8d1ba2a418d885"
269
- },
270
- {
271
- "name": "detachable_50m_block_0002.pt",
272
- "nbytes": 2703344,
273
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0002.pt",
274
- "sha256": "2a3c9681362e9d2f661649c6e2e6d8a6801bc95fd30f4170f5df785df2246f07"
275
- },
276
- {
277
- "name": "detachable_50m_block_0003.pt",
278
- "nbytes": 2703361,
279
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0003.pt",
280
- "sha256": "c601132f3532499d68ab8999adb52adf5981f53a3fbf8bda1844a86c07498605"
281
- },
282
- {
283
- "name": "detachable_50m_block_0004.pt",
284
- "nbytes": 2704834,
285
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0004.pt",
286
- "sha256": "c07246d94f7722382521dc21c138c54c14a91957b1d9a6edf844045f2533be6e"
287
- },
288
- {
289
- "name": "detachable_50m_block_0005.pt",
290
- "nbytes": 2711087,
291
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0005.pt",
292
- "sha256": "b0c3bd48808a91a56cf5451e9a5dbf73cea54e7a43e482dc495f7e5d3bb2b399"
293
- },
294
- {
295
- "name": "detachable_50m_block_0006.pt",
296
- "nbytes": 2720292,
297
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0006.pt",
298
- "sha256": "c6bae8ae912ad3b03f445a01e82babbafa96347feda1a3ed520ab25d0d46871e"
299
- },
300
- {
301
- "name": "detachable_50m_block_0007.pt",
302
- "nbytes": 2708816,
303
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0007.pt",
304
- "sha256": "da9f0f4b9a718a9195cb4669ac0bfb6e6a008930e77fdfea070592d56a4dbeb7"
305
- },
306
- {
307
- "name": "detachable_50m_block_0008.pt",
308
- "nbytes": 2718007,
309
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0008.pt",
310
- "sha256": "2eb9b2024f9a6bcc544f0bec19bb372897c0c72900751d2b3e6c865f949cc151"
311
- },
312
- {
313
- "name": "detachable_50m_block_0009.pt",
314
- "nbytes": 2714332,
315
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0009.pt",
316
- "sha256": "646591c0d2edc388423b8ca180a3a4bf205780d063f77d2fc7a7770dbd461a71"
317
- },
318
- {
319
- "name": "detachable_50m_block_0010.pt",
320
- "nbytes": 2708922,
321
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0010.pt",
322
- "sha256": "31c047af84c2f76bf6164ee5e2da89616ec28af28a5c5f753c986fafdb6c316b"
323
- },
324
- {
325
- "name": "detachable_50m_block_0011.pt",
326
- "nbytes": 2719472,
327
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0011.pt",
328
- "sha256": "4153a0a9403074af50ddd84fa47ed67a60a9ca3d0ce032803c1c8f5046f06ba4"
329
- },
330
- {
331
- "name": "detachable_50m_block_0012.pt",
332
- "nbytes": 2702611,
333
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0012.pt",
334
- "sha256": "cbaf3410a20bd78563e94c3efb30134a9f2038ff878a5cc996249d7dce747227"
335
- },
336
- {
337
- "name": "detachable_50m_block_0013.pt",
338
- "nbytes": 2698209,
339
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0013.pt",
340
- "sha256": "d76fc35be8ccb10e2e0c4ccded80220931389ce6794b562263fa9518578f87c3"
341
- },
342
- {
343
- "name": "detachable_50m_block_0014.pt",
344
- "nbytes": 2704341,
345
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0014.pt",
346
- "sha256": "a795041d5efa9a208240ea12e4e5825b4b74adf5b1bc67d512b0b254150167c5"
347
- },
348
- {
349
- "name": "detachable_50m_block_0015.pt",
350
- "nbytes": 2701194,
351
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0015.pt",
352
- "sha256": "96bdedf3823b35ad31648b1e001971f20c203f34d25ccbfc6cfb2f1b94962289"
353
- },
354
- {
355
- "name": "detachable_50m_block_0016.pt",
356
- "nbytes": 2699095,
357
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0016.pt",
358
- "sha256": "48dedf5da7ec52a8ff572e20633ef289eff0a41e6776e8a7920b7faa1dc3734b"
359
- },
360
- {
361
- "name": "detachable_50m_block_0017.pt",
362
- "nbytes": 2702653,
363
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0017.pt",
364
- "sha256": "667f489f82d732ec35f2744d5106010047faba432f3ceb73bf68f7bee0fe2033"
365
- },
366
- {
367
- "name": "detachable_50m_block_0018.pt",
368
- "nbytes": 2709027,
369
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0018.pt",
370
- "sha256": "963fe3f241708a796ef18ab1977cce1c3817b958fd506bafd5775878d6f9502b"
371
- },
372
- {
373
- "name": "detachable_50m_block_0019.pt",
374
- "nbytes": 1970441,
375
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0019.pt",
376
- "sha256": "cf509b705593fe6dbd965edb25261a4f7c23b21a9215159423f1a94098e196e4"
377
- },
378
- {
379
- "name": "detachable_50m_block_0020.pt",
380
- "nbytes": 1968839,
381
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0020.pt",
382
- "sha256": "a65e9159c5ae0f9c106cbfc1a789e88920cde98d87d2be5ab4ddfefcb1061bd6"
383
- },
384
- {
385
- "name": "detachable_50m_block_0021.pt",
386
- "nbytes": 1969684,
387
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/detachable_50m_block_0021.pt",
388
- "sha256": "2387cd6d259ec26ad5be9cfb78cbb948ddce51a08a31e6925f073bc0c10051e9"
389
- },
390
- {
391
- "name": "ar.pt",
392
- "nbytes": 608662259,
393
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/ar.pt",
394
- "sha256": "04bb6a7cf55b201f11e4543db4fca1e85bab957b324796aa229f603fa78366f5"
395
- },
396
- {
397
- "name": "sat.pt",
398
- "nbytes": 608671867,
399
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/sat.pt",
400
- "sha256": "5cf70776ec8ac6802785051efdeda7e0240c2d87da3c87e73a37c83e56bb12ab"
401
- },
402
- {
403
- "name": "nat.pt",
404
- "nbytes": 608662259,
405
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/nat.pt",
406
- "sha256": "ce881c5cba39c237b6deb5bc6095f4da0bd9e42a39bf5215896ca07e54b514a9"
407
- },
408
- {
409
- "name": "opt.pt",
410
- "nbytes": 1912700319,
411
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/opt.pt",
412
- "sha256": "d5796f4d51212832fb6da5eb2ebf699772249b55a10c1cfc7e8815ee26b713c4"
413
- },
414
- {
415
- "name": "scaler.pt",
416
- "nbytes": 149,
417
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/scaler.pt",
418
- "sha256": "c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c"
419
- },
420
- {
421
- "name": "agillm5_detachable_50m_bridge.pt",
422
- "nbytes": 1224682,
423
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_bridge.pt",
424
- "sha256": "5a1cdb19964639a9d72838c609bc0e3d86b49034132a85d8299592ccadaf215c"
425
- },
426
- {
427
- "name": "agillm5_detachable_50m_opt.pt",
428
- "nbytes": 64297085,
429
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_opt.pt",
430
- "sha256": "4b06385ee4641ce87e09e2be5269397e2d9eb76919bb88c15ba9b1d4868486ba"
431
- },
432
- {
433
- "name": "agillm5_detachable_50m_scaler.pt",
434
- "nbytes": 149,
435
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_scaler.pt",
436
- "sha256": "c2110dfd7b7a6c034fb62244a6f8f76a22d9936d55b6d0fe133e43e65b38631c"
437
- },
438
- {
439
- "name": "agillm5_detachable_50m_rng.pt",
440
- "nbytes": 3574,
441
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_rng.pt",
442
- "sha256": "c8e111428c21e2a677964e74dcb5c8875394a3792b51cf2084f67ad2ff6e19c7"
443
- },
444
- {
445
- "name": "agillm5_detachable_50m_counters.pt",
446
- "nbytes": 29937,
447
- "path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.shards/agillm5_detachable_50m_counters.pt",
448
- "sha256": "763695ec0328235974c67f83f9a2a4d2b45ce753b52481d7e13c14bd29c97515"
449
- }
450
- ],
451
- "step": 2849585,
452
- "tokenizer_sidecar_nbytes": 17465554,
453
- "tokenizer_sidecar_path": "/workspace/agillm43_production_v14/production_quality_v18_hard_spike_gates/checkpoints_continue400B_from200B/pretrain_step02849585_from02848688_20260911T103222_1789122742704700305_1518633_000002.pt.tokenizer.json",
454
- "tokenizer_sidecar_sha256": "bd178d14a063c648112c4b51efbc389da06b6b66c23387418f4d6be0920a7b47"
455
- }