Laplace-Heterogeneous MoE Routing / bench_report.json
Failed on benchmark
1{
2 "bench_version": 1,
3 "track": "heterogeneous_regime_moe",
4 "model": "custom_sparse_moe",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.006,
10 "alpha": 0.0,
11 "aux": 0.05,
12 "beta": 0.9,
13 "wd": 0.0,
14 "epochs": 3
15 },
16 "sweep": [
17 {
18 "cfg": {
19 "lr": 0.001,
20 "alpha": 0.0,
21 "aux": 0.0,
22 "beta": 0.9,
23 "wd": 0.0,
24 "epochs": 3
25 },
26 "mean": 0.3828819803893566,
27 "per_seed": [
28 0.38268065452575684,
29 0.42182663083076477,
30 0.39980241656303406,
31 0.4181539714336395,
32 0.3709557354450226,
33 0.3404829204082489,
34 0.3406737446784973,
35 0.3884797692298889
36 ]
37 },
38 {
39 "cfg": {
40 "lr": 0.001,
41 "alpha": 0.0,
42 "aux": 0.02,
43 "beta": 0.9,
44 "wd": 0.0,
45 "epochs": 3
46 },
47 "mean": 0.38272077590227127,
48 "per_seed": [
49 0.38274234533309937,
50 0.42027127742767334,
51 0.40094804763793945,
52 0.41814735531806946,
53 0.37042883038520813,
54 0.34024134278297424,
55 0.34057343006134033,
56 0.38841357827186584
57 ]
58 },
59 {
60 "cfg": {
61 "lr": 0.001,
62 "alpha": 0.0,
63 "aux": 0.05,
64 "beta": 0.9,
65 "wd": 0.0,
66 "epochs": 3
67 },
68 "mean": 0.38203955814242363,
69 "per_seed": [
70 0.3827840983867645,
71 0.41726404428482056,
72 0.4006299376487732,
73 0.4182635545730591,
74 0.37063243985176086,
75 0.3401484489440918,
76 0.33957916498184204,
77 0.387014776468277
78 ]
79 },
80 {
81 "cfg": {
82 "lr": 0.001,
83 "alpha": 0.0,
84 "aux": 0.1,
85 "beta": 0.9,
86 "wd": 0.0,
87 "epochs": 3
88 },
89 "mean": 0.3819722458720207,
90 "per_seed": [
91 0.3825948238372803,
92 0.4162457585334778,
93 0.40236616134643555,
94 0.4189237058162689,
95 0.37040382623672485,
96 0.3401601314544678,
97 0.33924177289009094,
98 0.3858417868614197
99 ]
100 },
101 {
102 "cfg": {
103 "lr": 0.003,
104 "alpha": 0.0,
105 "aux": 0.0,
106 "beta": 0.9,
107 "wd": 0.0,
108 "epochs": 3
109 },
110 "mean": 0.322677806019783,
111 "per_seed": [
112 0.336835652589798,
113 0.3815375566482544,
114 0.31310850381851196,
115 0.36289894580841064,
116 0.2942163646221161,
117 0.27027586102485657,
118 0.2819417417049408,
119 0.34060782194137573
120 ]
121 },
122 {
123 "cfg": {
124 "lr": 0.003,
125 "alpha": 0.0,
126 "aux": 0.02,
127 "beta": 0.9,
128 "wd": 0.0,
129 "epochs": 3
130 },
131 "mean": 0.3223222754895687,
132 "per_seed": [
133 0.33747565746307373,
134 0.37728649377822876,
135 0.3187519311904907,
136 0.3634164333343506,
137 0.28765812516212463,
138 0.26715168356895447,
139 0.28378787636756897,
140 0.3430500030517578
141 ]
142 },
143 {
144 "cfg": {
145 "lr": 0.003,
146 "alpha": 0.0,
147 "aux": 0.05,
148 "beta": 0.9,
149 "wd": 0.0,
150 "epochs": 3
151 },
152 "mean": 0.3225990757346153,
153 "per_seed": [
154 0.3332405090332031,
155 0.37485089898109436,
156 0.3205052614212036,
157 0.36014488339424133,
158 0.2920096516609192,
159 0.2685995101928711,
160 0.28830453753471375,
161 0.34313735365867615
162 ]
163 },
164 {
165 "cfg": {
166 "lr": 0.003,
167 "alpha": 0.0,
168 "aux": 0.1,
169 "beta": 0.9,
170 "wd": 0.0,
171 "epochs": 3
172 },
173 "mean": 0.32372206822037697,
174 "per_seed": [
175 0.3341885805130005,
176 0.374936580657959,
177 0.3235602080821991,
178 0.3638923466205597,
179 0.29552894830703735,
180 0.26977014541625977,
181 0.29074564576148987,
182 0.3371540904045105
183 ]
184 },
185 {
186 "cfg": {
187 "lr": 0.006,
188 "alpha": 0.0,
189 "aux": 0.0,
190 "beta": 0.9,
191 "wd": 0.0,
192 "epochs": 3
193 },
194 "mean": 0.3063250947743654,
195 "per_seed": [
196 0.340360552072525,
197 0.3520633578300476,
198 0.29556065797805786,
199 0.348251074552536,
200 0.26277485489845276,
201 0.275777131319046,
202 0.24570734798908234,
203 0.3301057815551758
204 ]
205 },
206 {
207 "cfg": {
208 "lr": 0.006,
209 "alpha": 0.0,
210 "aux": 0.02,
211 "beta": 0.9,
212 "wd": 0.0,
213 "epochs": 3
214 },
215 "mean": 0.3056254591792822,
216 "per_seed": [
217 0.3370864689350128,
218 0.3551115095615387,
219 0.29941967129707336,
220 0.33801084756851196,
221 0.26092731952667236,
222 0.2828543484210968,
223 0.23862560093402863,
224 0.3329679071903229
225 ]
226 },
227 {
228 "cfg": {
229 "lr": 0.006,
230 "alpha": 0.0,
231 "aux": 0.05,
232 "beta": 0.9,
233 "wd": 0.0,
234 "epochs": 3
235 },
236 "mean": 0.3030446618795395,
237 "per_seed": [
238 0.33402499556541443,
239 0.3469650149345398,
240 0.2999320328235626,
241 0.3352453112602234,
242 0.2639300525188446,
243 0.2693466544151306,
244 0.242201566696167,
245 0.33271166682243347
246 ]
247 },
248 {
249 "cfg": {
250 "lr": 0.006,
251 "alpha": 0.0,
252 "aux": 0.1,
253 "beta": 0.9,
254 "wd": 0.0,
255 "epochs": 3
256 },
257 "mean": 0.304129621013999,
258 "per_seed": [
259 0.33261367678642273,
260 0.34067946672439575,
261 0.2965738773345947,
262 0.3427000343799591,
263 0.2676871120929718,
264 0.26694804430007935,
265 0.2444242388010025,
266 0.3414105176925659
267 ]
268 }
269 ],
270 "full": {
271 "mean": 0.3030446618795395,
272 "std": 0.037387950188241835,
273 "per_seed": [
274 0.33402499556541443,
275 0.3469650149345398,
276 0.2999320328235626,
277 0.3352453112602234,
278 0.2639300525188446,
279 0.2693466544151306,
280 0.242201566696167,
281 0.33271166682243347
282 ],
283 "n": 8,
284 "signatures": [
285 {
286 "load_cv": 0.562938723734806,
287 "raw_cv": 0.16369719803202323,
288 "pressure": [
289 0.09628282487392426,
290 0.081390880048275,
291 0.08156248927116394,
292 0.09052031487226486,
293 0.108507439494133,
294 0.11428291350603104,
295 0.06951107084751129,
296 0.07551244646310806
297 ]
298 },
299 {
300 "load_cv": 0.5744562491404501,
301 "raw_cv": 0.1282355640205095,
302 "pressure": [
303 0.07845945656299591,
304 0.08906298130750656,
305 0.10167022794485092,
306 0.09873484820127487,
307 0.07295134663581848,
308 0.07633225619792938,
309 0.11173825711011887,
310 0.08862103521823883
311 ]
312 },
313 {
314 "load_cv": 0.6500000745006104,
315 "raw_cv": 0.18785403298311348,
316 "pressure": [
317 0.11149176955223083,
318 0.08400898426771164,
319 0.09686370939016342,
320 0.06456371396780014,
321 0.08030611276626587,
322 0.09944741427898407,
323 0.06263477355241776,
324 0.1182539090514183
325 ]
326 },
327 {
328 "load_cv": 0.617089927191566,
329 "raw_cv": 0.15858341208470625,
330 "pressure": [
331 0.07222633063793182,
332 0.06700322031974792,
333 0.09097043424844742,
334 0.0720442607998848,
335 0.1073877140879631,
336 0.09829375892877579,
337 0.0978299006819725,
338 0.11181474477052689
339 ]
340 },
341 {
342 "load_cv": 0.4471017718279356,
343 "raw_cv": 0.1292689442624241,
344 "pressure": [
345 0.09001918137073517,
346 0.10562513768672943,
347 0.0990309864282608,
348 0.06985172629356384,
349 0.09212449193000793,
350 0.09441789984703064,
351 0.09658448398113251,
352 0.06991647183895111
353 ]
354 },
355 {
356 "load_cv": 0.5402776598887137,
357 "raw_cv": 0.15026783943056055,
358 "pressure": [
359 0.0705578625202179,
360 0.10928625613451004,
361 0.08838503062725067,
362 0.09580148756504059,
363 0.10527030378580093,
364 0.06459853053092957,
365 0.11027625948190689,
366 0.07339463382959366
367 ]
368 },
369 {
370 "load_cv": 0.5783597575594265,
371 "raw_cv": 0.11876459419632313,
372 "pressure": [
373 0.0868268609046936,
374 0.09471505880355835,
375 0.11200853437185287,
376 0.08284948021173477,
377 0.08226083219051361,
378 0.08512511849403381,
379 0.09811943769454956,
380 0.07566508650779724
381 ]
382 },
383 {
384 "load_cv": 0.4230839014019498,
385 "raw_cv": 0.09526789188308794,
386 "pressure": [
387 0.08401855081319809,
388 0.09309402853250504,
389 0.08401310443878174,
390 0.08053044974803925,
391 0.0959775522351265,
392 0.10520738363265991,
393 0.09375947713851929,
394 0.08096984028816223
395 ]
396 }
397 ]
398 }
399 },
400 "idea": {
401 "mean": 0.30424891598522663,
402 "std": 0.049616770951017984,
403 "per_seed": [
404 0.34059908986091614,
405 0.3489361107349396,
406 0.2619922161102295,
407 0.3619314134120941,
408 0.27461159229278564,
409 0.2645150423049927,
410 0.22574420273303986,
411 0.35566166043281555
412 ],
413 "n": 8,
414 "signatures": [
415 {
416 "load_cv": 0.3511409759493393,
417 "raw_cv": 0.17809261381483654,
418 "pressure": [
419 0.09716491401195526,
420 0.0805320143699646,
421 0.08110848069190979,
422 0.08782943338155746,
423 0.10998763889074326,
424 0.11640482395887375,
425 0.07011502236127853,
426 0.07442808896303177
427 ]
428 },
429 {
430 "load_cv": 0.5395368337588062,
431 "raw_cv": 0.18354195356222183,
432 "pressure": [
433 0.07514716684818268,
434 0.08768722414970398,
435 0.10187951475381851,
436 0.10252103954553604,
437 0.07107587903738022,
438 0.07342103868722916,
439 0.11874773353338242,
440 0.08709078282117844
441 ]
442 },
443 {
444 "load_cv": 0.5327288508372604,
445 "raw_cv": 0.2340563237648359,
446 "pressure": [
447 0.11677199602127075,
448 0.07979930937290192,
449 0.09878582507371902,
450 0.062348250299692154,
451 0.07951708883047104,
452 0.09967383742332458,
453 0.05981113389134407,
454 0.1208629459142685
455 ]
456 },
457 {
458 "load_cv": 0.48010414838406806,
459 "raw_cv": 0.2041649222357629,
460 "pressure": [
461 0.07007212191820145,
462 0.06405238062143326,
463 0.09175959974527359,
464 0.07002781331539154,
465 0.11198381334543228,
466 0.09871535003185272,
467 0.09743785858154297,
468 0.11352145671844482
469 ]
470 },
471 {
472 "load_cv": 0.2900000214553521,
473 "raw_cv": 0.14455384015921344,
474 "pressure": [
475 0.09033729881048203,
476 0.10671012848615646,
477 0.09973443299531937,
478 0.06994510442018509,
479 0.09231198579072952,
480 0.0953044667840004,
481 0.09653972834348679,
482 0.06668726354837418
483 ]
484 },
485 {
486 "load_cv": 0.3588871657819647,
487 "raw_cv": 0.19309554993951816,
488 "pressure": [
489 0.0671059787273407,
490 0.11245114356279373,
491 0.08675403892993927,
492 0.09586981683969498,
493 0.10887758433818817,
494 0.06107066944241524,
495 0.11346158385276794,
496 0.07197955995798111
497 ]
498 },
499 {
500 "load_cv": 0.5058655738790097,
501 "raw_cv": 0.1606991684840679,
502 "pressure": [
503 0.08636239171028137,
504 0.09632236510515213,
505 0.11678074300289154,
506 0.08211132884025574,
507 0.08124763518571854,
508 0.08318377286195755,
509 0.10013129562139511,
510 0.07143087685108185
511 ]
512 },
513 {
514 "load_cv": 0.34161379933083946,
515 "raw_cv": 0.12798243760960235,
516 "pressure": [
517 0.08367805182933807,
518 0.09391549974679947,
519 0.08226265013217926,
520 0.0783558338880539,
521 0.09979941695928574,
522 0.10702546685934067,
523 0.09416698664426804,
524 0.07836645841598511
525 ]
526 }
527 ],
528 "best_cfg": {
529 "lr": 0.006,
530 "alpha": 2.0,
531 "aux": 0.0,
532 "beta": 0.9,
533 "wd": 0.0,
534 "epochs": 3
535 },
536 "sweep": [
537 {
538 "cfg": {
539 "lr": 0.001,
540 "alpha": 0.5,
541 "aux": 0.0,
542 "beta": 0.9,
543 "wd": 0.0,
544 "epochs": 3
545 },
546 "mean": 0.3829677514731884
547 },
548 {
549 "cfg": {
550 "lr": 0.001,
551 "alpha": 1.0,
552 "aux": 0.0,
553 "beta": 0.9,
554 "wd": 0.0,
555 "epochs": 3
556 },
557 "mean": 0.38300157710909843
558 },
559 {
560 "cfg": {
561 "lr": 0.001,
562 "alpha": 2.0,
563 "aux": 0.0,
564 "beta": 0.9,
565 "wd": 0.0,
566 "epochs": 3
567 },
568 "mean": 0.3818103075027466
569 },
570 {
571 "cfg": {
572 "lr": 0.003,
573 "alpha": 0.5,
574 "aux": 0.0,
575 "beta": 0.9,
576 "wd": 0.0,
577 "epochs": 3
578 },
579 "mean": 0.3212699890136719
580 },
581 {
582 "cfg": {
583 "lr": 0.003,
584 "alpha": 1.0,
585 "aux": 0.0,
586 "beta": 0.9,
587 "wd": 0.0,
588 "epochs": 3
589 },
590 "mean": 0.32134268060326576
591 },
592 {
593 "cfg": {
594 "lr": 0.003,
595 "alpha": 2.0,
596 "aux": 0.0,
597 "beta": 0.9,
598 "wd": 0.0,
599 "epochs": 3
600 },
601 "mean": 0.32290053740143776
602 },
603 {
604 "cfg": {
605 "lr": 0.006,
606 "alpha": 0.5,
607 "aux": 0.0,
608 "beta": 0.9,
609 "wd": 0.0,
610 "epochs": 3
611 },
612 "mean": 0.3090713191777468
613 },
614 {
615 "cfg": {
616 "lr": 0.006,
617 "alpha": 1.0,
618 "aux": 0.0,
619 "beta": 0.9,
620 "wd": 0.0,
621 "epochs": 3
622 },
623 "mean": 0.3074618801474571
624 },
625 {
626 "cfg": {
627 "lr": 0.006,
628 "alpha": 2.0,
629 "aux": 0.0,
630 "beta": 0.9,
631 "wd": 0.0,
632 "epochs": 3
633 },
634 "mean": 0.30424891598522663
635 }
636 ]
637 },
638 "comparison": {
639 "delta_mean": 0.0012042541056871414,
640 "idea_wins": 3,
641 "n_pairs": 8,
642 "per_seed_diffs": [
643 0.006574094295501709,
644 0.0019710958003997803,
645 -0.03793981671333313,
646 0.026686102151870728,
647 0.01068153977394104,
648 -0.0048316121101379395,
649 -0.016457363963127136,
650 0.02294999361038208
651 ],
652 "p_value": 0.8799,
653 "mde": 0.01765867919995847,
654 "mde_rel_pct": 5.827088024067492,
655 "verdict": "no measurable effect",
656 "system_worked": false,
657 "paired_delta_mean": 0.0012042541056871414,
658 "permutation_pvalue": 0.8799
659 },
660 "mechanism_signature": {
661 "predicted_raw_pressure_cv_reduction": 0.17827335119625737,
662 "observed_trained_routed_load_cv": 0.42498467117208,
663 "confirmed": false
664 },
665 "custom_track": {
666 "name": "heterogeneous_regime_moe",
667 "file": "stage2_bench.py",
668 "domain": "moe-routing"
669 },
670 "runtime_sec": 99.08687734603882
671}