AbstractPhil commited on
Commit
e4cfa12
·
verified ·
1 Parent(s): de5efd8

Ablation H-HIGH-H5_batch_shared_svd-s1

Browse files
H/HIGH/H5_batch_shared_svd/seed1/epoch_1_checkpoint.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d12f6fcf9f8054da5f03eecd3acb6afc97e16e55d39ff5a5819c84d07486a22d
3
+ size 542831
H/HIGH/H5_batch_shared_svd/seed1/final_report.json ADDED
@@ -0,0 +1,575 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "group": "H",
4
+ "variant": "H5_batch_shared_svd",
5
+ "band": "HIGH",
6
+ "seed": 1,
7
+ "phase": 2,
8
+ "num_epochs": 1,
9
+ "batch_size": 256,
10
+ "overrides": {
11
+ "svd": "batch_shared"
12
+ },
13
+ "description": "H-HIGH-H5_batch_shared_svd-s1"
14
+ },
15
+ "run_config": {
16
+ "matrix_v": 32,
17
+ "D": 4,
18
+ "patch_size": 4,
19
+ "hidden": 64,
20
+ "depth": 1,
21
+ "n_cross_layers": 1,
22
+ "n_heads": 4,
23
+ "max_alpha": 0.2,
24
+ "alpha_init": -2.0,
25
+ "img_size": 64,
26
+ "batch_size": 128,
27
+ "lr": 0.0001,
28
+ "epochs": 1,
29
+ "weight_decay": 0.0,
30
+ "use_cv_ema": true,
31
+ "cv_ema_alpha": 0.01,
32
+ "cv_alignment_epochs": 0,
33
+ "cv_measure_every": 50,
34
+ "cv_sigma_scale": 0.3,
35
+ "boost": 0.5,
36
+ "cross_attn_clip": 0.5,
37
+ "allowed_types": [
38
+ 0,
39
+ 1,
40
+ 2,
41
+ 3,
42
+ 4,
43
+ 5,
44
+ 6,
45
+ 7,
46
+ 8,
47
+ 9,
48
+ 10,
49
+ 11,
50
+ 12,
51
+ 13,
52
+ 14,
53
+ 15
54
+ ],
55
+ "train_size": 1000000,
56
+ "val_size": 10000,
57
+ "num_workers": 2,
58
+ "report_every": 100,
59
+ "major_report_every": 10,
60
+ "save_every": 5,
61
+ "seed": 1,
62
+ "hf_repo": "AbstractPhil/geolip-svae-batteries",
63
+ "upload": false
64
+ },
65
+ "cv_ema_final": 0.9290479173888639,
66
+ "cv_last": 0.9397374568294355,
67
+ "predicted_band": "HIGH",
68
+ "expected_band": "HIGH",
69
+ "band_match": true,
70
+ "test_mse": 0.3020685613155365,
71
+ "recon_ema": 0.3030633640931148,
72
+ "S0": 57.71232986450195,
73
+ "SD": 35.24213409423828,
74
+ "ratio": 1.637594638673183,
75
+ "erank": 3.9268202781677246,
76
+ "observed_sphere_cv": 0.9809161725488351,
77
+ "uniform_sphere_cv_prediction": 0.9230044578859847,
78
+ "band_deviation": 0.05791171466285039,
79
+ "params_finite": true,
80
+ "num_epochs_run": 1,
81
+ "start_epoch": 0,
82
+ "per_epoch_metrics": [
83
+ {
84
+ "epoch": 1,
85
+ "test_mse": 0.3020685613155365,
86
+ "cv_ema": 0.9290479173888639,
87
+ "observed_sphere_cv": 0.9809161725488351,
88
+ "band_deviation": 0.05791171466285039,
89
+ "erank": 3.9268202781677246,
90
+ "params_finite": true,
91
+ "wallclock_seconds": 57.35291910171509,
92
+ "checkpoint_path": "/content/ablations_phase2/H/HIGH/H5_batch_shared_svd_s1/epoch_1_checkpoint.pt"
93
+ }
94
+ ],
95
+ "params_count": 40703,
96
+ "wallclock_seconds": 57.35416793823242,
97
+ "batches_completed": 3906,
98
+ "batch_limit": 3906,
99
+ "cv_trajectory": [
100
+ {
101
+ "batch": 0,
102
+ "cv": 0.9333301926551878,
103
+ "cv_ema": 0.9333301926551878,
104
+ "recon": 1.7899079322814941
105
+ },
106
+ {
107
+ "batch": 50,
108
+ "cv": 0.9563751678141216,
109
+ "cv_ema": 0.933560642406777,
110
+ "recon": 1.473886489868164
111
+ },
112
+ {
113
+ "batch": 100,
114
+ "cv": 0.8415401523596465,
115
+ "cv_ema": 0.9326404375063057,
116
+ "recon": 1.1819517612457275
117
+ },
118
+ {
119
+ "batch": 150,
120
+ "cv": 0.9136874758016399,
121
+ "cv_ema": 0.932450907889259,
122
+ "recon": 1.1885793209075928
123
+ },
124
+ {
125
+ "batch": 200,
126
+ "cv": 0.9133888343507703,
127
+ "cv_ema": 0.9322602871538741,
128
+ "recon": 0.9235541224479675
129
+ },
130
+ {
131
+ "batch": 250,
132
+ "cv": 0.9158807786483318,
133
+ "cv_ema": 0.9320964920688187,
134
+ "recon": 1.0997004508972168
135
+ },
136
+ {
137
+ "batch": 300,
138
+ "cv": 1.002382532833677,
139
+ "cv_ema": 0.9327993524764673,
140
+ "recon": 0.7543506622314453
141
+ },
142
+ {
143
+ "batch": 350,
144
+ "cv": 0.8627157729849815,
145
+ "cv_ema": 0.9320985166815524,
146
+ "recon": 0.8052618503570557
147
+ },
148
+ {
149
+ "batch": 400,
150
+ "cv": 0.8363180793225564,
151
+ "cv_ema": 0.9311407123079624,
152
+ "recon": 0.8751272559165955
153
+ },
154
+ {
155
+ "batch": 450,
156
+ "cv": 0.8835276739969499,
157
+ "cv_ema": 0.9306645819248522,
158
+ "recon": 0.5888237357139587
159
+ },
160
+ {
161
+ "batch": 500,
162
+ "cv": 0.9103015315600811,
163
+ "cv_ema": 0.9304609514212044,
164
+ "recon": 0.8726667165756226
165
+ },
166
+ {
167
+ "batch": 550,
168
+ "cv": 0.9260731092632379,
169
+ "cv_ema": 0.9304170729996247,
170
+ "recon": 0.7402987480163574
171
+ },
172
+ {
173
+ "batch": 600,
174
+ "cv": 0.9342195771742599,
175
+ "cv_ema": 0.930455098041371,
176
+ "recon": 0.5168558955192566
177
+ },
178
+ {
179
+ "batch": 650,
180
+ "cv": 0.9310764350503995,
181
+ "cv_ema": 0.9304613114114614,
182
+ "recon": 0.6251877546310425
183
+ },
184
+ {
185
+ "batch": 700,
186
+ "cv": 0.8299733219711184,
187
+ "cv_ema": 0.929456431517058,
188
+ "recon": 0.6470041871070862
189
+ },
190
+ {
191
+ "batch": 750,
192
+ "cv": 0.9628889076769744,
193
+ "cv_ema": 0.9297907562786571,
194
+ "recon": 0.6045675277709961
195
+ },
196
+ {
197
+ "batch": 800,
198
+ "cv": 0.7448224392138737,
199
+ "cv_ema": 0.9279410731080093,
200
+ "recon": 0.6039523482322693
201
+ },
202
+ {
203
+ "batch": 850,
204
+ "cv": 0.903372662221209,
205
+ "cv_ema": 0.9276953889991414,
206
+ "recon": 0.5161391496658325
207
+ },
208
+ {
209
+ "batch": 900,
210
+ "cv": 1.0666779796632504,
211
+ "cv_ema": 0.9290852149057826,
212
+ "recon": 0.5709394812583923
213
+ },
214
+ {
215
+ "batch": 950,
216
+ "cv": 0.911388934695054,
217
+ "cv_ema": 0.9289082521036753,
218
+ "recon": 0.4927499294281006
219
+ },
220
+ {
221
+ "batch": 1000,
222
+ "cv": 1.010214405481214,
223
+ "cv_ema": 0.9297213136374507,
224
+ "recon": 0.5622485876083374
225
+ },
226
+ {
227
+ "batch": 1050,
228
+ "cv": 0.9722861629546391,
229
+ "cv_ema": 0.9301469621306225,
230
+ "recon": 0.5008425116539001
231
+ },
232
+ {
233
+ "batch": 1100,
234
+ "cv": 0.8053090922699685,
235
+ "cv_ema": 0.928898583432016,
236
+ "recon": 0.502134382724762
237
+ },
238
+ {
239
+ "batch": 1150,
240
+ "cv": 0.9452159517872816,
241
+ "cv_ema": 0.9290617571155687,
242
+ "recon": 0.5883464217185974
243
+ },
244
+ {
245
+ "batch": 1200,
246
+ "cv": 0.9343141757990314,
247
+ "cv_ema": 0.9291142813024034,
248
+ "recon": 0.553665816783905
249
+ },
250
+ {
251
+ "batch": 1250,
252
+ "cv": 0.919280849032186,
253
+ "cv_ema": 0.9290159469797012,
254
+ "recon": 0.5350209474563599
255
+ },
256
+ {
257
+ "batch": 1300,
258
+ "cv": 0.9355246096627459,
259
+ "cv_ema": 0.9290810336065316,
260
+ "recon": 0.5733937621116638
261
+ },
262
+ {
263
+ "batch": 1350,
264
+ "cv": 0.9382969734992151,
265
+ "cv_ema": 0.9291731930054585,
266
+ "recon": 0.43988871574401855
267
+ },
268
+ {
269
+ "batch": 1400,
270
+ "cv": 0.985715220047369,
271
+ "cv_ema": 0.9297386132758776,
272
+ "recon": 0.417407751083374
273
+ },
274
+ {
275
+ "batch": 1450,
276
+ "cv": 0.8210612646786681,
277
+ "cv_ema": 0.9286518397899055,
278
+ "recon": 0.4636685848236084
279
+ },
280
+ {
281
+ "batch": 1500,
282
+ "cv": 0.9084032156729126,
283
+ "cv_ema": 0.9284493535487356,
284
+ "recon": 0.4234672486782074
285
+ },
286
+ {
287
+ "batch": 1550,
288
+ "cv": 0.9000399236806153,
289
+ "cv_ema": 0.9281652592500544,
290
+ "recon": 0.5699334144592285
291
+ },
292
+ {
293
+ "batch": 1600,
294
+ "cv": 0.9338885611925521,
295
+ "cv_ema": 0.9282224922694794,
296
+ "recon": 0.371613472700119
297
+ },
298
+ {
299
+ "batch": 1650,
300
+ "cv": 0.9190059792479747,
301
+ "cv_ema": 0.9281303271392644,
302
+ "recon": 0.38867494463920593
303
+ },
304
+ {
305
+ "batch": 1700,
306
+ "cv": 0.9118195425238086,
307
+ "cv_ema": 0.9279672192931098,
308
+ "recon": 0.4008999764919281
309
+ },
310
+ {
311
+ "batch": 1750,
312
+ "cv": 0.816273881154946,
313
+ "cv_ema": 0.9268502859117281,
314
+ "recon": 0.41784703731536865
315
+ },
316
+ {
317
+ "batch": 1800,
318
+ "cv": 1.0527232879895247,
319
+ "cv_ema": 0.928109015932506,
320
+ "recon": 0.4079371392726898
321
+ },
322
+ {
323
+ "batch": 1850,
324
+ "cv": 0.8682528906145742,
325
+ "cv_ema": 0.9275104546793267,
326
+ "recon": 0.42887037992477417
327
+ },
328
+ {
329
+ "batch": 1900,
330
+ "cv": 0.989046980683674,
331
+ "cv_ema": 0.9281258199393702,
332
+ "recon": 0.42127132415771484
333
+ },
334
+ {
335
+ "batch": 1950,
336
+ "cv": 0.9295171253767693,
337
+ "cv_ema": 0.9281397329937441,
338
+ "recon": 0.41422033309936523
339
+ },
340
+ {
341
+ "batch": 2000,
342
+ "cv": 0.8789478646723349,
343
+ "cv_ema": 0.9276478143105299,
344
+ "recon": 0.2949117124080658
345
+ },
346
+ {
347
+ "batch": 2050,
348
+ "cv": 0.9899760338351095,
349
+ "cv_ema": 0.9282710965057757,
350
+ "recon": 0.39275845885276794
351
+ },
352
+ {
353
+ "batch": 2100,
354
+ "cv": 0.8669715198694775,
355
+ "cv_ema": 0.9276581007394127,
356
+ "recon": 0.3933781385421753
357
+ },
358
+ {
359
+ "batch": 2150,
360
+ "cv": 0.9366930246937433,
361
+ "cv_ema": 0.927748449978956,
362
+ "recon": 0.3658299148082733
363
+ },
364
+ {
365
+ "batch": 2200,
366
+ "cv": 0.8703927587002757,
367
+ "cv_ema": 0.9271748930661692,
368
+ "recon": 0.3182758688926697
369
+ },
370
+ {
371
+ "batch": 2250,
372
+ "cv": 1.064160325436516,
373
+ "cv_ema": 0.9285447473898726,
374
+ "recon": 0.3113817572593689
375
+ },
376
+ {
377
+ "batch": 2300,
378
+ "cv": 0.8522905967342862,
379
+ "cv_ema": 0.9277822058833167,
380
+ "recon": 0.33392569422721863
381
+ },
382
+ {
383
+ "batch": 2350,
384
+ "cv": 0.8737673313735863,
385
+ "cv_ema": 0.9272420571382194,
386
+ "recon": 0.4219721257686615
387
+ },
388
+ {
389
+ "batch": 2400,
390
+ "cv": 0.8823974164914996,
391
+ "cv_ema": 0.9267936107317521,
392
+ "recon": 0.4592185914516449
393
+ },
394
+ {
395
+ "batch": 2450,
396
+ "cv": 0.8914412984113109,
397
+ "cv_ema": 0.9264400876085477,
398
+ "recon": 0.2934716045856476
399
+ },
400
+ {
401
+ "batch": 2500,
402
+ "cv": 0.9089993453663091,
403
+ "cv_ema": 0.9262656801861253,
404
+ "recon": 0.37836241722106934
405
+ },
406
+ {
407
+ "batch": 2550,
408
+ "cv": 0.9552886810800134,
409
+ "cv_ema": 0.9265559101950642,
410
+ "recon": 0.2832300662994385
411
+ },
412
+ {
413
+ "batch": 2600,
414
+ "cv": 0.9572819161452084,
415
+ "cv_ema": 0.9268631702545655,
416
+ "recon": 0.3185213506221771
417
+ },
418
+ {
419
+ "batch": 2650,
420
+ "cv": 0.9138946195274691,
421
+ "cv_ema": 0.9267334847472946,
422
+ "recon": 0.3585360050201416
423
+ },
424
+ {
425
+ "batch": 2700,
426
+ "cv": 0.9294784368267298,
427
+ "cv_ema": 0.9267609342680888,
428
+ "recon": 0.2951648235321045
429
+ },
430
+ {
431
+ "batch": 2750,
432
+ "cv": 0.8613526418599315,
433
+ "cv_ema": 0.9261068513440073,
434
+ "recon": 0.35388848185539246
435
+ },
436
+ {
437
+ "batch": 2800,
438
+ "cv": 1.0970791576864953,
439
+ "cv_ema": 0.9278165744074321,
440
+ "recon": 0.2782197594642639
441
+ },
442
+ {
443
+ "batch": 2850,
444
+ "cv": 0.9628872289225974,
445
+ "cv_ema": 0.9281672809525837,
446
+ "recon": 0.3193906247615814
447
+ },
448
+ {
449
+ "batch": 2900,
450
+ "cv": 0.9693682582875168,
451
+ "cv_ema": 0.928579290725933,
452
+ "recon": 0.283214271068573
453
+ },
454
+ {
455
+ "batch": 2950,
456
+ "cv": 0.980442300307494,
457
+ "cv_ema": 0.9290979208217486,
458
+ "recon": 0.28262457251548767
459
+ },
460
+ {
461
+ "batch": 3000,
462
+ "cv": 1.003134060627183,
463
+ "cv_ema": 0.9298382822198029,
464
+ "recon": 0.33111342787742615
465
+ },
466
+ {
467
+ "batch": 3050,
468
+ "cv": 0.9698476143480744,
469
+ "cv_ema": 0.9302383755410856,
470
+ "recon": 0.30264902114868164
471
+ },
472
+ {
473
+ "batch": 3100,
474
+ "cv": 0.8839880526123687,
475
+ "cv_ema": 0.9297758723117985,
476
+ "recon": 0.31580156087875366
477
+ },
478
+ {
479
+ "batch": 3150,
480
+ "cv": 0.9426064679496433,
481
+ "cv_ema": 0.929904178268177,
482
+ "recon": 0.27825242280960083
483
+ },
484
+ {
485
+ "batch": 3200,
486
+ "cv": 0.8379216607953662,
487
+ "cv_ema": 0.9289843530934488,
488
+ "recon": 0.3000723123550415
489
+ },
490
+ {
491
+ "batch": 3250,
492
+ "cv": 0.8817049598379773,
493
+ "cv_ema": 0.9285115591608941,
494
+ "recon": 0.2769181430339813
495
+ },
496
+ {
497
+ "batch": 3300,
498
+ "cv": 0.981134621493685,
499
+ "cv_ema": 0.929037789784222,
500
+ "recon": 0.3098016381263733
501
+ },
502
+ {
503
+ "batch": 3350,
504
+ "cv": 1.0128105593859875,
505
+ "cv_ema": 0.9298755174802397,
506
+ "recon": 0.2997617721557617
507
+ },
508
+ {
509
+ "batch": 3400,
510
+ "cv": 0.8006982136208997,
511
+ "cv_ema": 0.9285837444416463,
512
+ "recon": 0.29007261991500854
513
+ },
514
+ {
515
+ "batch": 3450,
516
+ "cv": 1.0136063187487903,
517
+ "cv_ema": 0.9294339701847177,
518
+ "recon": 0.2872808277606964
519
+ },
520
+ {
521
+ "batch": 3500,
522
+ "cv": 0.9124230631763295,
523
+ "cv_ema": 0.9292638611146338,
524
+ "recon": 0.3082427978515625
525
+ },
526
+ {
527
+ "batch": 3550,
528
+ "cv": 0.898009051933911,
529
+ "cv_ema": 0.9289513130228265,
530
+ "recon": 0.24971339106559753
531
+ },
532
+ {
533
+ "batch": 3600,
534
+ "cv": 0.9270032718532364,
535
+ "cv_ema": 0.9289318326111305,
536
+ "recon": 0.34069371223449707
537
+ },
538
+ {
539
+ "batch": 3650,
540
+ "cv": 0.8342856516303838,
541
+ "cv_ema": 0.927985370801323,
542
+ "recon": 0.257284939289093
543
+ },
544
+ {
545
+ "batch": 3700,
546
+ "cv": 1.00547444062475,
547
+ "cv_ema": 0.9287602614995574,
548
+ "recon": 0.2841836214065552
549
+ },
550
+ {
551
+ "batch": 3750,
552
+ "cv": 1.0121257858200436,
553
+ "cv_ema": 0.9295939167427621,
554
+ "recon": 0.3320108950138092
555
+ },
556
+ {
557
+ "batch": 3800,
558
+ "cv": 0.9257378684643812,
559
+ "cv_ema": 0.9295553562599783,
560
+ "recon": 0.3151949644088745
561
+ },
562
+ {
563
+ "batch": 3850,
564
+ "cv": 0.8680139545621077,
565
+ "cv_ema": 0.9289399422429996,
566
+ "recon": 0.289948433637619
567
+ },
568
+ {
569
+ "batch": 3900,
570
+ "cv": 0.9397374568294355,
571
+ "cv_ema": 0.9290479173888639,
572
+ "recon": 0.3325439393520355
573
+ }
574
+ ]
575
+ }
H/HIGH/H5_batch_shared_svd/seed1/tensorboard/events.out.tfevents.1776798055.52b79b2c7612.4109.74 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22c4e8f7612a0ee5b7cf4dae7dc79ed0d1d05c0ae61986287458e16d1a6d6d1e
3
+ size 612125