PowerMachine commited on
Commit
9f8c324
·
verified ·
1 Parent(s): 64465e3

V6.5-attn-v2 cleanup: remove legacy v6_5_7ds_attn_v2_report.json

Browse files
Files changed (1) hide show
  1. v6_5_7ds_attn_v2_report.json +0 -741
v6_5_7ds_attn_v2_report.json DELETED
@@ -1,741 +0,0 @@
1
- {
2
- "version": "V6.5-attn-v2-100batch-6000meta-save-states-user-questions",
3
- "timestamp": "2026-08-08T03:11:28.820924",
4
- "user_requirements_checklist": {
5
- "HF_TOKEN_deleted_after_use": "PENDING (will delete after upload)",
6
- "streaming_datasets_active_REAL": true,
7
- "xeon_runtime_active": true,
8
- "V65_ENABLE_STREAMING_forced": true,
9
- "attention_active_and_logic_functional": true,
10
- "streaming_100_per_batch": true,
11
- "meta_minima_6000_atingida": true,
12
- "no_synthetic_data": true,
13
- "streaming_with_pauses": true,
14
- "storage_critical_check": true,
15
- "model_states_saved_for_evaluation": true,
16
- "user_questions_launched_without_help": true,
17
- "aggressive_ram_cleanup": true,
18
- "aggressive_storage_cleanup": true,
19
- "metrics_reasoning_response_verified": true,
20
- "exhausted_7_datasets_in_sequence": true,
21
- "som_grid_864_neurons": true,
22
- "mtp_head_size_increased_K6": true,
23
- "vqvae2_active_in_pipeline": true,
24
- "reasoning_engine_integrated_to_kls": true,
25
- "smoothquant_w8a8_integrated_to_kls": true,
26
- "tool_coordinator_workers_reactivated": true,
27
- "ewc_w8a8_dequant_benchmark_active": true,
28
- "logic_and_bugfixes_verified": true
29
- },
30
- "config": {
31
- "BATCH_SIZE": 16,
32
- "STREAM_BATCH_SIZE": 100,
33
- "datasets_to_exhaust": [
34
- "dominguesm/restore-punctuation-ptbr-dataset",
35
- "carolina-c4ai/corpus-carolina",
36
- "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1",
37
- "dominguesm/Canarim-Instruct-PTBR-Dataset",
38
- "adalbertojunior/punctuation-ptbr",
39
- "iara-project/news-articles-ptbr-dataset",
40
- "manoela/noticias_ptbr"
41
- ],
42
- "MAX_SAMPLES_PER_DATASET": 860,
43
- "TOTAL_SAMPLES_TARGET": 6020,
44
- "META_MINIMA_SAMPLES": 6000,
45
- "SOM_GRID": [
46
- 6,
47
- 6,
48
- 6,
49
- 4
50
- ],
51
- "n_neurons": 864,
52
- "HIDDEN_DIM": 1024,
53
- "VOCAB_SIZE": 16384,
54
- "MAX_SEQ_LEN": 8,
55
- "MTP_K": 6,
56
- "INTER_BATCH_PAUSE_S": 0.3,
57
- "INTER_STREAM_BATCH_PAUSE_S": 0.5,
58
- "INTER_DATASET_PAUSE_S": 1.5,
59
- "STORAGE_CRITICAL_PCT": 90,
60
- "ATTENTION_N_HEADS": 8
61
- },
62
- "xeon_status": {
63
- "version": "V6",
64
- "physical_cores": 2,
65
- "env": {
66
- "MKL_ENABLE_INSTRUCTIONS": "AVX512",
67
- "MKL_NUM_THREADS": "2",
68
- "OMP_NUM_THREADS": "2",
69
- "MKL_DYNAMIC": "FALSE",
70
- "DNNL_PRIMITIVE_CACHE_CAPACITY": "1024",
71
- "ONEDNN_MAX_CPU_ISA": "AMX_INT8",
72
- "KMP_AFFINITY": "granularity=fine,compact,1,0",
73
- "KMP_BLOCKTIME": "1"
74
- },
75
- "avx512": {
76
- "supported": true,
77
- "desc": "AVX512_VNNI (full INT8 acceleration)"
78
- },
79
- "amx": {
80
- "supported": true,
81
- "desc": "AMX (tile + int8 + bf16) — full AMX acceleration"
82
- },
83
- "ipex_available": false,
84
- "init_done": true
85
- },
86
- "fp16_benchmark": {
87
- "best_time_ms": 94.03223100161995,
88
- "avg_time_ms": 94.2973510009324,
89
- "best_tflops": 1.3612353831931827,
90
- "avg_tflops": 1.3574082266503367,
91
- "matrix_size": 4000.0
92
- },
93
- "training": {
94
- "duration_s": 561.1727123260498,
95
- "n_steps": 420,
96
- "samples_per_dataset_actual": {
97
- "dominguesm/restore-punctuation-ptbr-dataset": 860,
98
- "carolina-c4ai/corpus-carolina": 860,
99
- "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1": 860,
100
- "dominguesm/Canarim-Instruct-PTBR-Dataset": 860,
101
- "adalbertojunior/punctuation-ptbr": 860,
102
- "iara-project/news-articles-ptbr-dataset": 860,
103
- "manoela/noticias_ptbr": 860
104
- },
105
- "total_samples_processed": 6020,
106
- "meta_minima_atingida": true,
107
- "streaming_failures": {
108
- "dominguesm/restore-punctuation-ptbr-dataset": 0,
109
- "carolina-c4ai/corpus-carolina": 0,
110
- "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1": 0,
111
- "dominguesm/Canarim-Instruct-PTBR-Dataset": 0,
112
- "adalbertojunior/punctuation-ptbr": 0,
113
- "iara-project/news-articles-ptbr-dataset": 0,
114
- "manoela/noticias_ptbr": 0
115
- },
116
- "storage_critical_stopped": false,
117
- "inter_batch_pause_s": 0.3,
118
- "inter_stream_batch_pause_s": 0.5,
119
- "inter_dataset_pause_s": 1.5
120
- },
121
- "summary": {
122
- "n_steps": 420,
123
- "final_loss": -1.0,
124
- "mean_loss": -0.9793500178662157,
125
- "min_loss": -1.0,
126
- "max_loss": -0.7071067811865476,
127
- "final_acc": 1.0,
128
- "mean_acc": 0.9608372896302778,
129
- "sigma_start": 1.3846745195799537,
130
- "sigma_end": 1.2194385807030765e-05,
131
- "vqvae2_mean_total_loss": 0.19771544410708988,
132
- "vqvae2_final_total_loss": 0.3119245767593384,
133
- "w8a8_mean_relative_error": 0.029152057293855862,
134
- "w8a8_final_relative_error": 0.005616033231113835,
135
- "attention_final_n_calls": 6160,
136
- "attention_max_n_calls": 6160,
137
- "rss_max_mb": 1770.1796875,
138
- "rss_final_mb": 1770.1796875,
139
- "disk_max_pct": 15.96249353869993,
140
- "disk_final_pct": 15.96249353869993,
141
- "n_alerts": 419,
142
- "alerts": [
143
- {
144
- "type": "3.3_loss_vanishing",
145
- "step": 1,
146
- "value": -1.0
147
- },
148
- {
149
- "type": "3.3_loss_vanishing",
150
- "step": 2,
151
- "value": -0.9682458365518543
152
- },
153
- {
154
- "type": "3.3_loss_vanishing",
155
- "step": 3,
156
- "value": -1.0
157
- },
158
- {
159
- "type": "3.3_loss_vanishing",
160
- "step": 4,
161
- "value": -1.0
162
- },
163
- {
164
- "type": "3.3_loss_vanishing",
165
- "step": 5,
166
- "value": -0.9682458365518543
167
- },
168
- {
169
- "type": "3.3_loss_vanishing",
170
- "step": 6,
171
- "value": -1.0
172
- },
173
- {
174
- "type": "3.3_loss_vanishing",
175
- "step": 7,
176
- "value": -0.8660254037844386
177
- },
178
- {
179
- "type": "3.3_loss_vanishing",
180
- "step": 8,
181
- "value": -1.0
182
- },
183
- {
184
- "type": "3.3_loss_vanishing",
185
- "step": 9,
186
- "value": -0.9354143466934853
187
- },
188
- {
189
- "type": "3.3_loss_vanishing",
190
- "step": 10,
191
- "value": -1.0
192
- },
193
- {
194
- "type": "3.3_loss_vanishing",
195
- "step": 11,
196
- "value": -0.9682458365518543
197
- },
198
- {
199
- "type": "3.3_loss_vanishing",
200
- "step": 12,
201
- "value": -1.0
202
- },
203
- {
204
- "type": "3.3_loss_vanishing",
205
- "step": 13,
206
- "value": -1.0
207
- },
208
- {
209
- "type": "3.3_loss_vanishing",
210
- "step": 14,
211
- "value": -0.9486832980505138
212
- },
213
- {
214
- "type": "3.3_loss_vanishing",
215
- "step": 15,
216
- "value": -1.0
217
- },
218
- {
219
- "type": "3.3_loss_vanishing",
220
- "step": 16,
221
- "value": -0.9354143466934853
222
- },
223
- {
224
- "type": "3.3_loss_vanishing",
225
- "step": 17,
226
- "value": -1.0
227
- },
228
- {
229
- "type": "3.3_loss_vanishing",
230
- "step": 18,
231
- "value": -0.9682458365518543
232
- },
233
- {
234
- "type": "3.3_loss_vanishing",
235
- "step": 19,
236
- "value": -1.0
237
- },
238
- {
239
- "type": "3.3_loss_vanishing",
240
- "step": 20,
241
- "value": -0.8660254037844386
242
- },
243
- {
244
- "type": "3.3_loss_vanishing",
245
- "step": 21,
246
- "value": -0.9486832980505138
247
- },
248
- {
249
- "type": "3.3_loss_vanishing",
250
- "step": 22,
251
- "value": -1.0
252
- },
253
- {
254
- "type": "3.3_loss_vanishing",
255
- "step": 23,
256
- "value": -0.9682458365518543
257
- },
258
- {
259
- "type": "3.3_loss_vanishing",
260
- "step": 24,
261
- "value": -1.0
262
- },
263
- {
264
- "type": "3.3_loss_vanishing",
265
- "step": 25,
266
- "value": -0.9354143466934853
267
- },
268
- {
269
- "type": "3.3_loss_vanishing",
270
- "step": 26,
271
- "value": -1.0
272
- },
273
- {
274
- "type": "3.3_loss_vanishing",
275
- "step": 27,
276
- "value": -0.8660254037844386
277
- },
278
- {
279
- "type": "3.3_loss_vanishing",
280
- "step": 28,
281
- "value": -0.9219544457292888
282
- },
283
- {
284
- "type": "3.3_loss_vanishing",
285
- "step": 29,
286
- "value": -1.0
287
- },
288
- {
289
- "type": "3.3_loss_vanishing",
290
- "step": 30,
291
- "value": -0.8660254037844386
292
- }
293
- ],
294
- "kohonen_final": {
295
- "sigma_t": 1.2194385807030765e-05,
296
- "alpha_t": 0.00028512436826796326,
297
- "t": 11720,
298
- "n_neurons": 864,
299
- "fisher_w_mean": 0.0,
300
- "fisher_w_max": 0.0,
301
- "fisher_accum_count": 0,
302
- "has_ewc_reference": true,
303
- "weights_norm": 1973.36376953125,
304
- "weights_w_mean": -0.01385963149368763
305
- },
306
- "hypothesis_final": {
307
- "classifier_trained": true,
308
- "punishment_count": 0,
309
- "success_count": 2,
310
- "training_ready": true,
311
- "buffer_size": 28,
312
- "required_new_samples": 10,
313
- "histogram_max": 28,
314
- "time_counter": 6160
315
- },
316
- "mtp_final": {
317
- "active": false,
318
- "mtp_total_loss": 0.0,
319
- "mtp_weighted_loss": 0.0,
320
- "mtp_entropy_reg": 0.0,
321
- "mtp_alphas": [],
322
- "mtp_per_k_loss": [],
323
- "active_in_val": true,
324
- "entropy_beta": 0.01,
325
- "K": 6
326
- },
327
- "vqvae2_final": {
328
- "active": true,
329
- "n_calls": 100,
330
- "latest_total_loss": 0.3119245767593384,
331
- "latest_recon_loss": 0.31192177534103394,
332
- "latest_vq_loss": 2.7990381568088196e-06,
333
- "mean_total_loss": 0.2312491385638714
334
- },
335
- "w8a8_final": {
336
- "active": true,
337
- "n_calls": 100,
338
- "latest_relative_error": 0.005616033231113835,
339
- "latest_compression_ratio": 4.0,
340
- "latest_preserves_topology": true,
341
- "mean_relative_error": 0.015359933862165893,
342
- "all_preserve_topology": false
343
- },
344
- "attention_final": {
345
- "active": true,
346
- "n_calls": 6160,
347
- "n_errors": 0,
348
- "last_norm_in": 112.62097930908203,
349
- "last_norm_out": 117.49085235595703,
350
- "last_attn_activated": true,
351
- "last_attn_diff_norm": 119.20104217529297,
352
- "n_heads": 8,
353
- "logic_functional": true
354
- },
355
- "reasoning_final": {
356
- "active": true,
357
- "n_history": 0,
358
- "n_steps_last": 0,
359
- "phases_used": [],
360
- "tool_coordinator_workers_active": true
361
- },
362
- "ewc_final": {
363
- "active": true,
364
- "ewc_eval_mode_penalty": true,
365
- "w8a8_dequant_active": true,
366
- "fisher_w_mean": 0.0,
367
- "fisher_w_max": 0.0,
368
- "fisher_accum_count": 0
369
- }
370
- },
371
- "kohonen_final": {
372
- "t": 11720,
373
- "sigma_t": 1.2194385807030765e-05,
374
- "alpha_t": 0.00028512436826796326,
375
- "sigma0": 1.5,
376
- "alpha0": 0.1,
377
- "lambda_ewc": 0.02,
378
- "grid_shape": [
379
- 6,
380
- 6,
381
- 6,
382
- 4
383
- ],
384
- "n_neurons": 864,
385
- "has_ewc_reference": true,
386
- "fisher_w_mean": 0.0,
387
- "fisher_w_max": 0.0,
388
- "fisher_accum_count": 0,
389
- "weights_norm": 1973.36376953125,
390
- "weights_w_mean": -0.01385963149368763
391
- },
392
- "kls_state": {
393
- "som": {
394
- "t": 11720,
395
- "sigma_t": 1.2194385807030765e-05,
396
- "alpha_t": 0.00028512436826796326,
397
- "sigma0": 1.5,
398
- "alpha0": 0.1,
399
- "lambda_ewc": 0.02,
400
- "grid_shape": [
401
- 6,
402
- 6,
403
- 6,
404
- 4
405
- ],
406
- "n_neurons": 864,
407
- "has_ewc_reference": true,
408
- "fisher_w_mean": 0.0,
409
- "fisher_w_max": 0.0,
410
- "fisher_accum_count": 0,
411
- "weights_norm": 1973.36376953125,
412
- "weights_w_mean": -0.01385963149368763
413
- },
414
- "kls": {
415
- "time_counter": 6180,
416
- "T_max": 10000,
417
- "buffer_size": 28,
418
- "training_ready": true,
419
- "punishment_count": 0,
420
- "success_count": 2,
421
- "classifier_trained": true,
422
- "histogram_size": 1,
423
- "histogram_max": 28,
424
- "N_start": 10,
425
- "dim_choice": "y",
426
- "som_neuron_count": 864,
427
- "required_new_samples": 10,
428
- "has_classifier": true,
429
- "vocab_size": 16384,
430
- "hidden_dim": 1024,
431
- "seq_len": 8,
432
- "enable_vqvae2": true,
433
- "enable_reasoning": true,
434
- "enable_w8a8": true,
435
- "vqvae2_n_calls": 100,
436
- "w8a8_n_calls": 100,
437
- "tool_coordinator_workers_active": true,
438
- "enable_attention": true,
439
- "attention_active": true,
440
- "attention_n_heads": 8
441
- },
442
- "vqvae2": {
443
- "active": true,
444
- "n_calls": 100,
445
- "latest": {
446
- "vq_loss": 2.7990381568088196e-06,
447
- "recon_loss": 0.31192177534103394,
448
- "total_loss": 0.3119245767593384,
449
- "n_used_top": 64,
450
- "n_used_bot": 128,
451
- "usage_ratio_top": 1.0,
452
- "usage_ratio_bot": 1.0,
453
- "codebook_ppl_top": 13.45434264387107,
454
- "codebook_ppl_bot": 15.999999999743997,
455
- "n_restarted_top": 0,
456
- "n_restarted_bot": 0,
457
- "goose_temp": 0.5,
458
- "active": true
459
- },
460
- "mean_total_loss": 0.2312491385638714,
461
- "mean_recon_loss": 0.2312279748916626,
462
- "n_nan_skipped": 0
463
- },
464
- "reasoning": {
465
- "active": true,
466
- "stats": {
467
- "n_steps": 8,
468
- "n_history": 10,
469
- "tools": [
470
- "som_query",
471
- "buffer_stats"
472
- ],
473
- "tool_stats": {
474
- "som_query": {
475
- "n_calls": 150,
476
- "n_success": 150,
477
- "n_failures": 0,
478
- "total_time_s": 1.0832843780517578,
479
- "avg_time_s": 0.007221895853678386,
480
- "cache_size": 0,
481
- "instructions_count": 0,
482
- "error_count": 0,
483
- "success_rate": 1.0,
484
- "circuit_breaker": {
485
- "state": "closed",
486
- "failure_count": 0,
487
- "success_count": 0,
488
- "failure_threshold": 5,
489
- "recovery_timeout_s": 30.0
490
- }
491
- },
492
- "buffer_stats": {
493
- "n_calls": 140,
494
- "n_success": 140,
495
- "n_failures": 0,
496
- "total_time_s": 0.2509152889251709,
497
- "avg_time_s": 0.0017922520637512207,
498
- "cache_size": 0,
499
- "instructions_count": 0,
500
- "error_count": 0,
501
- "success_rate": 1.0,
502
- "circuit_breaker": {
503
- "state": "closed",
504
- "failure_count": 0,
505
- "success_count": 0,
506
- "failure_threshold": 5,
507
- "recovery_timeout_s": 30.0
508
- }
509
- }
510
- },
511
- "phases_used": [
512
- "executing",
513
- "answering",
514
- "predicting",
515
- "planning",
516
- "monitoring",
517
- "thinking",
518
- "decomposing",
519
- "adjusting"
520
- ]
521
- },
522
- "n_history": 10
523
- },
524
- "attention": {
525
- "active": true,
526
- "n_calls": 6181,
527
- "n_errors": 0,
528
- "last_norm_in": 114.05921936035156,
529
- "last_norm_out": 117.0058822631836,
530
- "last_attn_activated": true,
531
- "last_attn_diff_norm": 117.8182144165039,
532
- "n_heads": 8,
533
- "logic_functional": true
534
- },
535
- "w8a8": {
536
- "active": true,
537
- "n_calls": 100,
538
- "latest": {
539
- "active": true,
540
- "n_neurons": 864,
541
- "n_features": 4,
542
- "n_calibration_samples": 16,
543
- "smooth_scale_mean": 0.11529327929019928,
544
- "smooth_scale_std": 0.22994910180568695,
545
- "smooth_scale_min": 0.00031872352701611817,
546
- "smooth_scale_max": 0.4602169394493103,
547
- "alpha": 0.5,
548
- "n_bits": 8,
549
- "compression_ratio": 4.0,
550
- "weights_norm_original": 1962.3458251953125,
551
- "weights_norm_dequant": 1973.36376953125,
552
- "absolute_error": 11.020599365234375,
553
- "relative_error": 0.005616033231113835,
554
- "preserves_topology": true,
555
- "w8a8_int8_range": [
556
- -122,
557
- 127
558
- ],
559
- "elapsed_ms": 0.5145072937011719
560
- },
561
- "mean_relative_error": 0.015359933862165893,
562
- "max_relative_error": 0.16713649906961597,
563
- "all_preserve_topology": false
564
- }
565
- },
566
- "verification": {
567
- "checks": {
568
- "kls_instantiation": {
569
- "status": "PASS",
570
- "details": "vqvae2=True, reasoning=True, w8a8=True, attention=True, workers_active=True"
571
- },
572
- "find_bmu_no_premature_return": {
573
- "status": "PASS",
574
- "details": "bmu=(0, 1, 1, 0) valid"
575
- },
576
- "activate_hypothesis_detach": {
577
- "status": "PASS",
578
- "details": "classifier_trained=True"
579
- },
580
- "pgvector_lookup_removed": {
581
- "status": "PASS"
582
- },
583
- "kohonen_refactored_removed": {
584
- "status": "PASS",
585
- "details": "old_folder=False, new_file=True"
586
- },
587
- "trainer_removed": {
588
- "status": "PASS"
589
- },
590
- "vqvae2_valid_output": {
591
- "status": "PASS",
592
- "details": "n_calls=1, total_loss=0.1517"
593
- },
594
- "reasoning_engine_tags": {
595
- "status": "PASS",
596
- "details": "reasoning length=934 chars"
597
- },
598
- "smoothquant_w8a8_som_weights": {
599
- "status": "PASS",
600
- "details": "relative_error=0.002880, compression_ratio=4.0x, n_neurons=54"
601
- },
602
- "tool_coordinator_workers_reactivated": {
603
- "status": "PASS",
604
- "details": "n_workers=4, n_tools=2, n_results=2, elapsed_ms=2.70"
605
- },
606
- "attention_integrated_and_functional": {
607
- "status": "PASS",
608
- "details": "n_calls=15, n_errors=0, last_attn_activated=True, last_norm_in=16.6656, last_norm_out=17.7281, logic_functional=True, n_heads=8"
609
- }
610
- },
611
- "n_pass": 11,
612
- "n_fail": 0,
613
- "all_pass": true
614
- },
615
- "attention_eval_summary": {
616
- "active": true,
617
- "n_calls": 6161,
618
- "n_errors": 0,
619
- "last_attn_activated": true,
620
- "logic_functional": true,
621
- "attention_modifies_output": true,
622
- "verdict": "ACTIVE_AND_FUNCTIONAL"
623
- },
624
- "ewc_w8a8_benchmark_summary": {
625
- "dequant_relative_error": 0.0,
626
- "int8_relative_error": 0.0,
627
- "dequant_preserves_accuracy": true,
628
- "conclusion": "V6.5-attn EWC+W8A8 eval com dequant: erro dequant=0.000000 (< 0.01 = OK), erro int8 direto=0.000000."
629
- },
630
- "reasoning_eval_summary": {
631
- "n_with_answer": 7,
632
- "n_with_think": 7,
633
- "answer_rate": 1.0,
634
- "think_rate": 1.0,
635
- "avg_reasoning_length": 1034.5714285714287,
636
- "reasoning_engine_active": true,
637
- "reasoning_engine_n_history": 7
638
- },
639
- "user_questions_eval_summary": {
640
- "n_with_answer": 3,
641
- "n_with_think": 3,
642
- "n_with_plan": 3,
643
- "n_with_decompose": 3,
644
- "answer_rate": 1.0,
645
- "think_rate": 1.0,
646
- "plan_rate": 1.0,
647
- "decompose_rate": 1.0,
648
- "avg_latency_ms": 12.185970942179361,
649
- "avg_reasoning_length": 1022.0
650
- },
651
- "user_questions_quality_assessment": {
652
- "response_quality": "GOOD",
653
- "reasoning_quality": "GOOD",
654
- "model_not_helped": true,
655
- "note": "As 3 perguntas foram enviadas verbatim, sem system prompt, sem few-shot, sem contexto adicional. Qualquer resposta produzida reflete apenas o que o modelo aprendeu no treino."
656
- },
657
- "model_states_save_info": {
658
- "saved": true,
659
- "path": "/home/z/my-project/BiGRU_T_version/v6_5_7ds_attn_v2_model_states.pt",
660
- "size_mb": 69.621303,
661
- "reason": "end_of_training",
662
- "step": 420,
663
- "total_samples": 6020,
664
- "n_tensors": 6,
665
- "n_buffer_tail": 28
666
- },
667
- "w8a8_compression_summary": {
668
- "active": true,
669
- "n_calls": 100,
670
- "latest": {
671
- "active": true,
672
- "n_neurons": 864,
673
- "n_features": 4,
674
- "n_calibration_samples": 16,
675
- "smooth_scale_mean": 0.11529327929019928,
676
- "smooth_scale_std": 0.22994910180568695,
677
- "smooth_scale_min": 0.00031872352701611817,
678
- "smooth_scale_max": 0.4602169394493103,
679
- "alpha": 0.5,
680
- "n_bits": 8,
681
- "compression_ratio": 4.0,
682
- "weights_norm_original": 1962.3458251953125,
683
- "weights_norm_dequant": 1973.36376953125,
684
- "absolute_error": 11.020599365234375,
685
- "relative_error": 0.005616033231113835,
686
- "preserves_topology": true,
687
- "w8a8_int8_range": [
688
- -122,
689
- 127
690
- ],
691
- "elapsed_ms": 0.5145072937011719
692
- },
693
- "mean_relative_error": 0.015359933862165893,
694
- "max_relative_error": 0.16713649906961597,
695
- "all_preserve_topology": false
696
- },
697
- "attention_metrics_final": {
698
- "active": true,
699
- "n_calls": 6181,
700
- "n_errors": 0,
701
- "last_norm_in": 114.05921936035156,
702
- "last_norm_out": 117.0058822631836,
703
- "last_attn_activated": true,
704
- "last_attn_diff_norm": 117.8182144165039,
705
- "n_heads": 8,
706
- "logic_functional": true
707
- },
708
- "module_analysis_summary": {
709
- "n_removed_modules": 9,
710
- "n_removed_folders": 1,
711
- "n_active_modules": 22
712
- },
713
- "math_analysis": {
714
- "text_to_4d": "SVD: M @ V[:3].T -> centroid 3D + w = time_step/T_max (LINEAR)",
715
- "attention_pipeline": "fused (L,D) -> MultiHeadAttention(self-attn, n_heads=8) -> fused + attn_out -> SVD",
716
- "bmu_distance": "||W - x||^2 (L2 squared in R^4)",
717
- "neighborhood": "Lambda(d, sigma) = exp(-d^2 / (2*sigma^2))",
718
- "weight_update": "dW = alpha * Lambda * (x - W)",
719
- "sigma_decay": "sigma_t = sigma0 * exp(-t/1000)",
720
- "alpha_decay": "alpha_t = alpha0 * exp(-t/2000)",
721
- "ewc_only_dim4": "penalty = lambda * F * (W_w - W*_w)",
722
- "vqvae2_loss": "L = recon_loss + vq_loss (commitment top + bottom + diversity)",
723
- "smoothquant_w8a8": "s_j = max|X_j|^alpha / max|W_j|^(1-alpha); W_smooth = W * diag(s); INT8 quant; dequant = W_int8 * scale / diag(s)",
724
- "mtp_loss": "L = sum_k(alpha_k * L_k) - beta * H(alpha), K=6 (V6.5-attn)",
725
- "ewc_w8a8_dequant": "W_float = (W_int8 * scale) / smooth_scale; EWC uses W_float for (p-w*)^2"
726
- },
727
- "datasets_used": [
728
- "dominguesm/restore-punctuation-ptbr-dataset",
729
- "carolina-c4ai/corpus-carolina",
730
- "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1",
731
- "dominguesm/Canarim-Instruct-PTBR-Dataset",
732
- "adalbertojunior/punctuation-ptbr",
733
- "iara-project/news-articles-ptbr-dataset",
734
- "manoela/noticias_ptbr"
735
- ],
736
- "storage_cleanup_pre": {
737
- "n_files_removed": 0,
738
- "mb_freed": 0.0,
739
- "timestamp": "2026-08-08T03:02:07.249633"
740
- }
741
- }