mikeumus-divincian commited on
Commit
9bcc2b3
·
verified ·
1 Parent(s): b5d3986

Add phase1_moe_svd.json

Browse files
Files changed (1) hide show
  1. phase1_moe_svd.json +1292 -0
phase1_moe_svd.json ADDED
@@ -0,0 +1,1292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0": {
3
+ "layer": 0,
4
+ "is_moe": true,
5
+ "routed_experts": {
6
+ "median_var64": 0.0848,
7
+ "q25_var64": 0.0844,
8
+ "q75_var64": 0.0852,
9
+ "mean_s0": 2.88,
10
+ "std_s0": 0.21,
11
+ "mean_s0_ratio": 1.02,
12
+ "n_experts": 256
13
+ },
14
+ "shared_expert": {
15
+ "var64": 0.1268,
16
+ "s0": 3.53,
17
+ "shape": [
18
+ 4096,
19
+ 2048
20
+ ]
21
+ },
22
+ "router": {
23
+ "var64": 0.5374,
24
+ "s0": 7.86,
25
+ "s0_s1": 1.15,
26
+ "shape": [
27
+ 256,
28
+ 4096
29
+ ]
30
+ }
31
+ },
32
+ "1": {
33
+ "layer": 1,
34
+ "is_moe": true,
35
+ "routed_experts": {
36
+ "median_var64": 0.0866,
37
+ "q25_var64": 0.0865,
38
+ "q75_var64": 0.0868,
39
+ "mean_s0": 2.84,
40
+ "std_s0": 0.09,
41
+ "mean_s0_ratio": 1.01,
42
+ "n_experts": 256
43
+ },
44
+ "shared_expert": {
45
+ "var64": 0.1906,
46
+ "s0": 3.8,
47
+ "shape": [
48
+ 4096,
49
+ 2048
50
+ ]
51
+ },
52
+ "router": {
53
+ "var64": 0.4951,
54
+ "s0": 4.92,
55
+ "s0_s1": 1.44,
56
+ "shape": [
57
+ 256,
58
+ 4096
59
+ ]
60
+ }
61
+ },
62
+ "2": {
63
+ "layer": 2,
64
+ "is_moe": true,
65
+ "routed_experts": {
66
+ "median_var64": 0.0865,
67
+ "q25_var64": 0.0863,
68
+ "q75_var64": 0.0867,
69
+ "mean_s0": 2.87,
70
+ "std_s0": 0.35,
71
+ "mean_s0_ratio": 1.02,
72
+ "n_experts": 256
73
+ },
74
+ "shared_expert": {
75
+ "var64": 0.2442,
76
+ "s0": 3.8,
77
+ "shape": [
78
+ 4096,
79
+ 2048
80
+ ]
81
+ },
82
+ "router": {
83
+ "var64": 0.4392,
84
+ "s0": 3.65,
85
+ "s0_s1": 1.07,
86
+ "shape": [
87
+ 256,
88
+ 4096
89
+ ]
90
+ }
91
+ },
92
+ "3": {
93
+ "layer": 3,
94
+ "is_moe": true,
95
+ "routed_experts": {
96
+ "median_var64": 0.0998,
97
+ "q25_var64": 0.095,
98
+ "q75_var64": 0.1042,
99
+ "mean_s0": 3.72,
100
+ "std_s0": 0.47,
101
+ "mean_s0_ratio": 1.11,
102
+ "n_experts": 256
103
+ },
104
+ "shared_expert": {
105
+ "var64": 0.0948,
106
+ "s0": 4.65,
107
+ "shape": [
108
+ 4096,
109
+ 2048
110
+ ]
111
+ },
112
+ "router": {
113
+ "var64": 0.4697,
114
+ "s0": 6.69,
115
+ "s0_s1": 1.09,
116
+ "shape": [
117
+ 256,
118
+ 4096
119
+ ]
120
+ }
121
+ },
122
+ "4": {
123
+ "layer": 4,
124
+ "is_moe": true,
125
+ "routed_experts": {
126
+ "median_var64": 0.1017,
127
+ "q25_var64": 0.097,
128
+ "q75_var64": 0.1068,
129
+ "mean_s0": 3.9,
130
+ "std_s0": 0.53,
131
+ "mean_s0_ratio": 1.12,
132
+ "n_experts": 256
133
+ },
134
+ "shared_expert": {
135
+ "var64": 0.1239,
136
+ "s0": 4.34,
137
+ "shape": [
138
+ 4096,
139
+ 2048
140
+ ]
141
+ },
142
+ "router": {
143
+ "var64": 0.4817,
144
+ "s0": 6.34,
145
+ "s0_s1": 1.15,
146
+ "shape": [
147
+ 256,
148
+ 4096
149
+ ]
150
+ }
151
+ },
152
+ "5": {
153
+ "layer": 5,
154
+ "is_moe": true,
155
+ "routed_experts": {
156
+ "median_var64": 0.1062,
157
+ "q25_var64": 0.0981,
158
+ "q75_var64": 0.1157,
159
+ "mean_s0": 4.2,
160
+ "std_s0": 0.65,
161
+ "mean_s0_ratio": 1.16,
162
+ "n_experts": 256
163
+ },
164
+ "shared_expert": {
165
+ "var64": 0.1554,
166
+ "s0": 3.83,
167
+ "shape": [
168
+ 4096,
169
+ 2048
170
+ ]
171
+ },
172
+ "router": {
173
+ "var64": 0.4402,
174
+ "s0": 6.17,
175
+ "s0_s1": 1.23,
176
+ "shape": [
177
+ 256,
178
+ 4096
179
+ ]
180
+ }
181
+ },
182
+ "6": {
183
+ "layer": 6,
184
+ "is_moe": true,
185
+ "routed_experts": {
186
+ "median_var64": 0.1017,
187
+ "q25_var64": 0.0963,
188
+ "q75_var64": 0.1079,
189
+ "mean_s0": 4.11,
190
+ "std_s0": 0.62,
191
+ "mean_s0_ratio": 1.16,
192
+ "n_experts": 256
193
+ },
194
+ "shared_expert": {
195
+ "var64": 0.1669,
196
+ "s0": 4.28,
197
+ "shape": [
198
+ 4096,
199
+ 2048
200
+ ]
201
+ },
202
+ "router": {
203
+ "var64": 0.437,
204
+ "s0": 5.39,
205
+ "s0_s1": 1.12,
206
+ "shape": [
207
+ 256,
208
+ 4096
209
+ ]
210
+ }
211
+ },
212
+ "7": {
213
+ "layer": 7,
214
+ "is_moe": true,
215
+ "routed_experts": {
216
+ "median_var64": 0.1055,
217
+ "q25_var64": 0.0982,
218
+ "q75_var64": 0.1137,
219
+ "mean_s0": 4.26,
220
+ "std_s0": 0.56,
221
+ "mean_s0_ratio": 1.19,
222
+ "n_experts": 256
223
+ },
224
+ "shared_expert": {
225
+ "var64": 0.1686,
226
+ "s0": 4.15,
227
+ "shape": [
228
+ 4096,
229
+ 2048
230
+ ]
231
+ },
232
+ "router": {
233
+ "var64": 0.4548,
234
+ "s0": 5.14,
235
+ "s0_s1": 1.07,
236
+ "shape": [
237
+ 256,
238
+ 4096
239
+ ]
240
+ }
241
+ },
242
+ "8": {
243
+ "layer": 8,
244
+ "is_moe": true,
245
+ "routed_experts": {
246
+ "median_var64": 0.104,
247
+ "q25_var64": 0.0992,
248
+ "q75_var64": 0.1135,
249
+ "mean_s0": 4.28,
250
+ "std_s0": 0.68,
251
+ "mean_s0_ratio": 1.19,
252
+ "n_experts": 256
253
+ },
254
+ "shared_expert": {
255
+ "var64": 0.194,
256
+ "s0": 4.24,
257
+ "shape": [
258
+ 4096,
259
+ 2048
260
+ ]
261
+ },
262
+ "router": {
263
+ "var64": 0.4371,
264
+ "s0": 5.56,
265
+ "s0_s1": 1.14,
266
+ "shape": [
267
+ 256,
268
+ 4096
269
+ ]
270
+ }
271
+ },
272
+ "9": {
273
+ "layer": 9,
274
+ "is_moe": true,
275
+ "routed_experts": {
276
+ "median_var64": 0.1122,
277
+ "q25_var64": 0.105,
278
+ "q75_var64": 0.1233,
279
+ "mean_s0": 4.62,
280
+ "std_s0": 0.8,
281
+ "mean_s0_ratio": 1.2,
282
+ "n_experts": 256
283
+ },
284
+ "shared_expert": {
285
+ "var64": 0.2007,
286
+ "s0": 4.39,
287
+ "shape": [
288
+ 4096,
289
+ 2048
290
+ ]
291
+ },
292
+ "router": {
293
+ "var64": 0.4314,
294
+ "s0": 6.46,
295
+ "s0_s1": 1.24,
296
+ "shape": [
297
+ 256,
298
+ 4096
299
+ ]
300
+ }
301
+ },
302
+ "10": {
303
+ "layer": 10,
304
+ "is_moe": true,
305
+ "routed_experts": {
306
+ "median_var64": 0.1172,
307
+ "q25_var64": 0.1082,
308
+ "q75_var64": 0.1314,
309
+ "mean_s0": 4.78,
310
+ "std_s0": 0.81,
311
+ "mean_s0_ratio": 1.21,
312
+ "n_experts": 256
313
+ },
314
+ "shared_expert": {
315
+ "var64": 0.1916,
316
+ "s0": 4.74,
317
+ "shape": [
318
+ 4096,
319
+ 2048
320
+ ]
321
+ },
322
+ "router": {
323
+ "var64": 0.4471,
324
+ "s0": 6.66,
325
+ "s0_s1": 1.3,
326
+ "shape": [
327
+ 256,
328
+ 4096
329
+ ]
330
+ }
331
+ },
332
+ "11": {
333
+ "layer": 11,
334
+ "is_moe": true,
335
+ "routed_experts": {
336
+ "median_var64": 0.1153,
337
+ "q25_var64": 0.1045,
338
+ "q75_var64": 0.1313,
339
+ "mean_s0": 4.62,
340
+ "std_s0": 1.02,
341
+ "mean_s0_ratio": 1.19,
342
+ "n_experts": 256
343
+ },
344
+ "shared_expert": {
345
+ "var64": 0.1849,
346
+ "s0": 4.12,
347
+ "shape": [
348
+ 4096,
349
+ 2048
350
+ ]
351
+ },
352
+ "router": {
353
+ "var64": 0.4657,
354
+ "s0": 6.93,
355
+ "s0_s1": 1.29,
356
+ "shape": [
357
+ 256,
358
+ 4096
359
+ ]
360
+ }
361
+ },
362
+ "12": {
363
+ "layer": 12,
364
+ "is_moe": true,
365
+ "routed_experts": {
366
+ "median_var64": 0.1208,
367
+ "q25_var64": 0.109,
368
+ "q75_var64": 0.1364,
369
+ "mean_s0": 4.9,
370
+ "std_s0": 0.9,
371
+ "mean_s0_ratio": 1.22,
372
+ "n_experts": 256
373
+ },
374
+ "shared_expert": {
375
+ "var64": 0.2135,
376
+ "s0": 4.62,
377
+ "shape": [
378
+ 4096,
379
+ 2048
380
+ ]
381
+ },
382
+ "router": {
383
+ "var64": 0.4562,
384
+ "s0": 6.02,
385
+ "s0_s1": 1.18,
386
+ "shape": [
387
+ 256,
388
+ 4096
389
+ ]
390
+ }
391
+ },
392
+ "13": {
393
+ "layer": 13,
394
+ "is_moe": true,
395
+ "routed_experts": {
396
+ "median_var64": 0.1286,
397
+ "q25_var64": 0.1127,
398
+ "q75_var64": 0.1449,
399
+ "mean_s0": 5.15,
400
+ "std_s0": 0.97,
401
+ "mean_s0_ratio": 1.26,
402
+ "n_experts": 256
403
+ },
404
+ "shared_expert": {
405
+ "var64": 0.2054,
406
+ "s0": 4.42,
407
+ "shape": [
408
+ 4096,
409
+ 2048
410
+ ]
411
+ },
412
+ "router": {
413
+ "var64": 0.4671,
414
+ "s0": 6.36,
415
+ "s0_s1": 1.25,
416
+ "shape": [
417
+ 256,
418
+ 4096
419
+ ]
420
+ }
421
+ },
422
+ "14": {
423
+ "layer": 14,
424
+ "is_moe": true,
425
+ "routed_experts": {
426
+ "median_var64": 0.1426,
427
+ "q25_var64": 0.1234,
428
+ "q75_var64": 0.1575,
429
+ "mean_s0": 5.46,
430
+ "std_s0": 1.03,
431
+ "mean_s0_ratio": 1.25,
432
+ "n_experts": 256
433
+ },
434
+ "shared_expert": {
435
+ "var64": 0.2041,
436
+ "s0": 4.94,
437
+ "shape": [
438
+ 4096,
439
+ 2048
440
+ ]
441
+ },
442
+ "router": {
443
+ "var64": 0.4733,
444
+ "s0": 6.23,
445
+ "s0_s1": 1.15,
446
+ "shape": [
447
+ 256,
448
+ 4096
449
+ ]
450
+ }
451
+ },
452
+ "15": {
453
+ "layer": 15,
454
+ "is_moe": true,
455
+ "routed_experts": {
456
+ "median_var64": 0.1413,
457
+ "q25_var64": 0.1179,
458
+ "q75_var64": 0.1633,
459
+ "mean_s0": 5.49,
460
+ "std_s0": 1.18,
461
+ "mean_s0_ratio": 1.24,
462
+ "n_experts": 256
463
+ },
464
+ "shared_expert": {
465
+ "var64": 0.2034,
466
+ "s0": 4.99,
467
+ "shape": [
468
+ 4096,
469
+ 2048
470
+ ]
471
+ },
472
+ "router": {
473
+ "var64": 0.4797,
474
+ "s0": 6.75,
475
+ "s0_s1": 1.22,
476
+ "shape": [
477
+ 256,
478
+ 4096
479
+ ]
480
+ }
481
+ },
482
+ "16": {
483
+ "layer": 16,
484
+ "is_moe": true,
485
+ "routed_experts": {
486
+ "median_var64": 0.1471,
487
+ "q25_var64": 0.1232,
488
+ "q75_var64": 0.1748,
489
+ "mean_s0": 5.65,
490
+ "std_s0": 1.25,
491
+ "mean_s0_ratio": 1.2,
492
+ "n_experts": 256
493
+ },
494
+ "shared_expert": {
495
+ "var64": 0.1954,
496
+ "s0": 5.52,
497
+ "shape": [
498
+ 4096,
499
+ 2048
500
+ ]
501
+ },
502
+ "router": {
503
+ "var64": 0.4886,
504
+ "s0": 6.54,
505
+ "s0_s1": 1.22,
506
+ "shape": [
507
+ 256,
508
+ 4096
509
+ ]
510
+ }
511
+ },
512
+ "17": {
513
+ "layer": 17,
514
+ "is_moe": true,
515
+ "routed_experts": {
516
+ "median_var64": 0.1523,
517
+ "q25_var64": 0.1307,
518
+ "q75_var64": 0.1797,
519
+ "mean_s0": 5.99,
520
+ "std_s0": 1.35,
521
+ "mean_s0_ratio": 1.22,
522
+ "n_experts": 256
523
+ },
524
+ "shared_expert": {
525
+ "var64": 0.1902,
526
+ "s0": 4.94,
527
+ "shape": [
528
+ 4096,
529
+ 2048
530
+ ]
531
+ },
532
+ "router": {
533
+ "var64": 0.4963,
534
+ "s0": 6.21,
535
+ "s0_s1": 1.26,
536
+ "shape": [
537
+ 256,
538
+ 4096
539
+ ]
540
+ }
541
+ },
542
+ "18": {
543
+ "layer": 18,
544
+ "is_moe": true,
545
+ "routed_experts": {
546
+ "median_var64": 0.1487,
547
+ "q25_var64": 0.1313,
548
+ "q75_var64": 0.17,
549
+ "mean_s0": 5.9,
550
+ "std_s0": 1.36,
551
+ "mean_s0_ratio": 1.19,
552
+ "n_experts": 256
553
+ },
554
+ "shared_expert": {
555
+ "var64": 0.1949,
556
+ "s0": 6.07,
557
+ "shape": [
558
+ 4096,
559
+ 2048
560
+ ]
561
+ },
562
+ "router": {
563
+ "var64": 0.4994,
564
+ "s0": 5.86,
565
+ "s0_s1": 1.17,
566
+ "shape": [
567
+ 256,
568
+ 4096
569
+ ]
570
+ }
571
+ },
572
+ "19": {
573
+ "layer": 19,
574
+ "is_moe": true,
575
+ "routed_experts": {
576
+ "median_var64": 0.1157,
577
+ "q25_var64": 0.1058,
578
+ "q75_var64": 0.1274,
579
+ "mean_s0": 4.61,
580
+ "std_s0": 0.89,
581
+ "mean_s0_ratio": 1.13,
582
+ "n_experts": 256
583
+ },
584
+ "shared_expert": {
585
+ "var64": 0.1279,
586
+ "s0": 4.84,
587
+ "shape": [
588
+ 4096,
589
+ 2048
590
+ ]
591
+ },
592
+ "router": {
593
+ "var64": 0.4813,
594
+ "s0": 8.17,
595
+ "s0_s1": 1.25,
596
+ "shape": [
597
+ 256,
598
+ 4096
599
+ ]
600
+ }
601
+ },
602
+ "20": {
603
+ "layer": 20,
604
+ "is_moe": true,
605
+ "routed_experts": {
606
+ "median_var64": 0.1415,
607
+ "q25_var64": 0.1242,
608
+ "q75_var64": 0.1635,
609
+ "mean_s0": 5.69,
610
+ "std_s0": 1.37,
611
+ "mean_s0_ratio": 1.17,
612
+ "n_experts": 256
613
+ },
614
+ "shared_expert": {
615
+ "var64": 0.215,
616
+ "s0": 6.22,
617
+ "shape": [
618
+ 4096,
619
+ 2048
620
+ ]
621
+ },
622
+ "router": {
623
+ "var64": 0.5111,
624
+ "s0": 6.1,
625
+ "s0_s1": 1.25,
626
+ "shape": [
627
+ 256,
628
+ 4096
629
+ ]
630
+ }
631
+ },
632
+ "21": {
633
+ "layer": 21,
634
+ "is_moe": true,
635
+ "routed_experts": {
636
+ "median_var64": 0.1292,
637
+ "q25_var64": 0.1037,
638
+ "q75_var64": 0.1529,
639
+ "mean_s0": 5.3,
640
+ "std_s0": 1.5,
641
+ "mean_s0_ratio": 1.17,
642
+ "n_experts": 256
643
+ },
644
+ "shared_expert": {
645
+ "var64": 0.2144,
646
+ "s0": 5.84,
647
+ "shape": [
648
+ 4096,
649
+ 2048
650
+ ]
651
+ },
652
+ "router": {
653
+ "var64": 0.4869,
654
+ "s0": 6.86,
655
+ "s0_s1": 1.1,
656
+ "shape": [
657
+ 256,
658
+ 4096
659
+ ]
660
+ }
661
+ },
662
+ "22": {
663
+ "layer": 22,
664
+ "is_moe": true,
665
+ "routed_experts": {
666
+ "median_var64": 0.1209,
667
+ "q25_var64": 0.0982,
668
+ "q75_var64": 0.1496,
669
+ "mean_s0": 5.0,
670
+ "std_s0": 1.21,
671
+ "mean_s0_ratio": 1.15,
672
+ "n_experts": 256
673
+ },
674
+ "shared_expert": {
675
+ "var64": 0.2101,
676
+ "s0": 4.86,
677
+ "shape": [
678
+ 4096,
679
+ 2048
680
+ ]
681
+ },
682
+ "router": {
683
+ "var64": 0.4821,
684
+ "s0": 7.01,
685
+ "s0_s1": 1.23,
686
+ "shape": [
687
+ 256,
688
+ 4096
689
+ ]
690
+ }
691
+ },
692
+ "23": {
693
+ "layer": 23,
694
+ "is_moe": true,
695
+ "routed_experts": {
696
+ "median_var64": 0.0888,
697
+ "q25_var64": 0.0873,
698
+ "q75_var64": 0.0948,
699
+ "mean_s0": 3.42,
700
+ "std_s0": 0.76,
701
+ "mean_s0_ratio": 1.08,
702
+ "n_experts": 256
703
+ },
704
+ "shared_expert": {
705
+ "var64": 0.1911,
706
+ "s0": 5.09,
707
+ "shape": [
708
+ 4096,
709
+ 2048
710
+ ]
711
+ },
712
+ "router": {
713
+ "var64": 0.484,
714
+ "s0": 9.02,
715
+ "s0_s1": 1.1,
716
+ "shape": [
717
+ 256,
718
+ 4096
719
+ ]
720
+ }
721
+ },
722
+ "24": {
723
+ "layer": 24,
724
+ "is_moe": true,
725
+ "routed_experts": {
726
+ "median_var64": 0.0916,
727
+ "q25_var64": 0.0878,
728
+ "q75_var64": 0.1033,
729
+ "mean_s0": 3.65,
730
+ "std_s0": 0.9,
731
+ "mean_s0_ratio": 1.1,
732
+ "n_experts": 256
733
+ },
734
+ "shared_expert": {
735
+ "var64": 0.1699,
736
+ "s0": 4.09,
737
+ "shape": [
738
+ 4096,
739
+ 2048
740
+ ]
741
+ },
742
+ "router": {
743
+ "var64": 0.4696,
744
+ "s0": 7.31,
745
+ "s0_s1": 1.13,
746
+ "shape": [
747
+ 256,
748
+ 4096
749
+ ]
750
+ }
751
+ },
752
+ "25": {
753
+ "layer": 25,
754
+ "is_moe": true,
755
+ "routed_experts": {
756
+ "median_var64": 0.0953,
757
+ "q25_var64": 0.0903,
758
+ "q75_var64": 0.109,
759
+ "mean_s0": 3.96,
760
+ "std_s0": 1.16,
761
+ "mean_s0_ratio": 1.14,
762
+ "n_experts": 256
763
+ },
764
+ "shared_expert": {
765
+ "var64": 0.2279,
766
+ "s0": 4.9,
767
+ "shape": [
768
+ 4096,
769
+ 2048
770
+ ]
771
+ },
772
+ "router": {
773
+ "var64": 0.4568,
774
+ "s0": 7.36,
775
+ "s0_s1": 1.05,
776
+ "shape": [
777
+ 256,
778
+ 4096
779
+ ]
780
+ }
781
+ },
782
+ "26": {
783
+ "layer": 26,
784
+ "is_moe": true,
785
+ "routed_experts": {
786
+ "median_var64": 0.0954,
787
+ "q25_var64": 0.0915,
788
+ "q75_var64": 0.1058,
789
+ "mean_s0": 3.86,
790
+ "std_s0": 0.99,
791
+ "mean_s0_ratio": 1.11,
792
+ "n_experts": 256
793
+ },
794
+ "shared_expert": {
795
+ "var64": 0.199,
796
+ "s0": 4.62,
797
+ "shape": [
798
+ 4096,
799
+ 2048
800
+ ]
801
+ },
802
+ "router": {
803
+ "var64": 0.4361,
804
+ "s0": 6.16,
805
+ "s0_s1": 1.18,
806
+ "shape": [
807
+ 256,
808
+ 4096
809
+ ]
810
+ }
811
+ },
812
+ "27": {
813
+ "layer": 27,
814
+ "is_moe": true,
815
+ "routed_experts": {
816
+ "median_var64": 0.0981,
817
+ "q25_var64": 0.0943,
818
+ "q75_var64": 0.1058,
819
+ "mean_s0": 3.96,
820
+ "std_s0": 1.23,
821
+ "mean_s0_ratio": 1.13,
822
+ "n_experts": 256
823
+ },
824
+ "shared_expert": {
825
+ "var64": 0.3269,
826
+ "s0": 7.17,
827
+ "shape": [
828
+ 4096,
829
+ 2048
830
+ ]
831
+ },
832
+ "router": {
833
+ "var64": 0.444,
834
+ "s0": 7.06,
835
+ "s0_s1": 1.13,
836
+ "shape": [
837
+ 256,
838
+ 4096
839
+ ]
840
+ }
841
+ },
842
+ "28": {
843
+ "layer": 28,
844
+ "is_moe": true,
845
+ "routed_experts": {
846
+ "median_var64": 0.0938,
847
+ "q25_var64": 0.09,
848
+ "q75_var64": 0.1045,
849
+ "mean_s0": 3.95,
850
+ "std_s0": 1.19,
851
+ "mean_s0_ratio": 1.14,
852
+ "n_experts": 256
853
+ },
854
+ "shared_expert": {
855
+ "var64": 0.2398,
856
+ "s0": 4.73,
857
+ "shape": [
858
+ 4096,
859
+ 2048
860
+ ]
861
+ },
862
+ "router": {
863
+ "var64": 0.4454,
864
+ "s0": 6.37,
865
+ "s0_s1": 1.19,
866
+ "shape": [
867
+ 256,
868
+ 4096
869
+ ]
870
+ }
871
+ },
872
+ "29": {
873
+ "layer": 29,
874
+ "is_moe": true,
875
+ "routed_experts": {
876
+ "median_var64": 0.0937,
877
+ "q25_var64": 0.089,
878
+ "q75_var64": 0.11,
879
+ "mean_s0": 3.95,
880
+ "std_s0": 1.2,
881
+ "mean_s0_ratio": 1.12,
882
+ "n_experts": 256
883
+ },
884
+ "shared_expert": {
885
+ "var64": 0.2463,
886
+ "s0": 5.85,
887
+ "shape": [
888
+ 4096,
889
+ 2048
890
+ ]
891
+ },
892
+ "router": {
893
+ "var64": 0.4549,
894
+ "s0": 6.69,
895
+ "s0_s1": 1.14,
896
+ "shape": [
897
+ 256,
898
+ 4096
899
+ ]
900
+ }
901
+ },
902
+ "30": {
903
+ "layer": 30,
904
+ "is_moe": true,
905
+ "routed_experts": {
906
+ "median_var64": 0.0968,
907
+ "q25_var64": 0.0905,
908
+ "q75_var64": 0.1132,
909
+ "mean_s0": 4.16,
910
+ "std_s0": 1.32,
911
+ "mean_s0_ratio": 1.14,
912
+ "n_experts": 256
913
+ },
914
+ "shared_expert": {
915
+ "var64": 0.2351,
916
+ "s0": 6.71,
917
+ "shape": [
918
+ 4096,
919
+ 2048
920
+ ]
921
+ },
922
+ "router": {
923
+ "var64": 0.4442,
924
+ "s0": 6.39,
925
+ "s0_s1": 1.17,
926
+ "shape": [
927
+ 256,
928
+ 4096
929
+ ]
930
+ }
931
+ },
932
+ "31": {
933
+ "layer": 31,
934
+ "is_moe": true,
935
+ "routed_experts": {
936
+ "median_var64": 0.0947,
937
+ "q25_var64": 0.0906,
938
+ "q75_var64": 0.1092,
939
+ "mean_s0": 4.03,
940
+ "std_s0": 1.18,
941
+ "mean_s0_ratio": 1.14,
942
+ "n_experts": 256
943
+ },
944
+ "shared_expert": {
945
+ "var64": 0.2778,
946
+ "s0": 7.04,
947
+ "shape": [
948
+ 4096,
949
+ 2048
950
+ ]
951
+ },
952
+ "router": {
953
+ "var64": 0.4478,
954
+ "s0": 6.75,
955
+ "s0_s1": 1.14,
956
+ "shape": [
957
+ 256,
958
+ 4096
959
+ ]
960
+ }
961
+ },
962
+ "32": {
963
+ "layer": 32,
964
+ "is_moe": true,
965
+ "routed_experts": {
966
+ "median_var64": 0.092,
967
+ "q25_var64": 0.0892,
968
+ "q75_var64": 0.101,
969
+ "mean_s0": 3.83,
970
+ "std_s0": 1.11,
971
+ "mean_s0_ratio": 1.11,
972
+ "n_experts": 256
973
+ },
974
+ "shared_expert": {
975
+ "var64": 0.2878,
976
+ "s0": 6.94,
977
+ "shape": [
978
+ 4096,
979
+ 2048
980
+ ]
981
+ },
982
+ "router": {
983
+ "var64": 0.4465,
984
+ "s0": 5.95,
985
+ "s0_s1": 1.14,
986
+ "shape": [
987
+ 256,
988
+ 4096
989
+ ]
990
+ }
991
+ },
992
+ "33": {
993
+ "layer": 33,
994
+ "is_moe": true,
995
+ "routed_experts": {
996
+ "median_var64": 0.0925,
997
+ "q25_var64": 0.0898,
998
+ "q75_var64": 0.1035,
999
+ "mean_s0": 3.79,
1000
+ "std_s0": 1.06,
1001
+ "mean_s0_ratio": 1.11,
1002
+ "n_experts": 256
1003
+ },
1004
+ "shared_expert": {
1005
+ "var64": 0.2721,
1006
+ "s0": 5.62,
1007
+ "shape": [
1008
+ 4096,
1009
+ 2048
1010
+ ]
1011
+ },
1012
+ "router": {
1013
+ "var64": 0.4406,
1014
+ "s0": 5.59,
1015
+ "s0_s1": 1.05,
1016
+ "shape": [
1017
+ 256,
1018
+ 4096
1019
+ ]
1020
+ }
1021
+ },
1022
+ "34": {
1023
+ "layer": 34,
1024
+ "is_moe": true,
1025
+ "routed_experts": {
1026
+ "median_var64": 0.0922,
1027
+ "q25_var64": 0.09,
1028
+ "q75_var64": 0.1012,
1029
+ "mean_s0": 3.71,
1030
+ "std_s0": 1.17,
1031
+ "mean_s0_ratio": 1.09,
1032
+ "n_experts": 256
1033
+ },
1034
+ "shared_expert": {
1035
+ "var64": 0.2554,
1036
+ "s0": 6.2,
1037
+ "shape": [
1038
+ 4096,
1039
+ 2048
1040
+ ]
1041
+ },
1042
+ "router": {
1043
+ "var64": 0.4352,
1044
+ "s0": 5.0,
1045
+ "s0_s1": 1.08,
1046
+ "shape": [
1047
+ 256,
1048
+ 4096
1049
+ ]
1050
+ }
1051
+ },
1052
+ "35": {
1053
+ "layer": 35,
1054
+ "is_moe": true,
1055
+ "routed_experts": {
1056
+ "median_var64": 0.1022,
1057
+ "q25_var64": 0.0955,
1058
+ "q75_var64": 0.1166,
1059
+ "mean_s0": 4.32,
1060
+ "std_s0": 1.84,
1061
+ "mean_s0_ratio": 1.13,
1062
+ "n_experts": 256
1063
+ },
1064
+ "shared_expert": {
1065
+ "var64": 0.2325,
1066
+ "s0": 6.58,
1067
+ "shape": [
1068
+ 4096,
1069
+ 2048
1070
+ ]
1071
+ },
1072
+ "router": {
1073
+ "var64": 0.4464,
1074
+ "s0": 5.85,
1075
+ "s0_s1": 1.13,
1076
+ "shape": [
1077
+ 256,
1078
+ 4096
1079
+ ]
1080
+ }
1081
+ },
1082
+ "36": {
1083
+ "layer": 36,
1084
+ "is_moe": true,
1085
+ "routed_experts": {
1086
+ "median_var64": 0.0958,
1087
+ "q25_var64": 0.0918,
1088
+ "q75_var64": 0.1045,
1089
+ "mean_s0": 3.78,
1090
+ "std_s0": 1.0,
1091
+ "mean_s0_ratio": 1.1,
1092
+ "n_experts": 256
1093
+ },
1094
+ "shared_expert": {
1095
+ "var64": 0.1882,
1096
+ "s0": 7.17,
1097
+ "shape": [
1098
+ 4096,
1099
+ 2048
1100
+ ]
1101
+ },
1102
+ "router": {
1103
+ "var64": 0.452,
1104
+ "s0": 4.96,
1105
+ "s0_s1": 1.2,
1106
+ "shape": [
1107
+ 256,
1108
+ 4096
1109
+ ]
1110
+ }
1111
+ },
1112
+ "37": {
1113
+ "layer": 37,
1114
+ "is_moe": true,
1115
+ "routed_experts": {
1116
+ "median_var64": 0.0949,
1117
+ "q25_var64": 0.0904,
1118
+ "q75_var64": 0.1048,
1119
+ "mean_s0": 4.02,
1120
+ "std_s0": 1.65,
1121
+ "mean_s0_ratio": 1.13,
1122
+ "n_experts": 256
1123
+ },
1124
+ "shared_expert": {
1125
+ "var64": 0.2981,
1126
+ "s0": 8.46,
1127
+ "shape": [
1128
+ 4096,
1129
+ 2048
1130
+ ]
1131
+ },
1132
+ "router": {
1133
+ "var64": 0.4593,
1134
+ "s0": 4.93,
1135
+ "s0_s1": 1.09,
1136
+ "shape": [
1137
+ 256,
1138
+ 4096
1139
+ ]
1140
+ }
1141
+ },
1142
+ "38": {
1143
+ "layer": 38,
1144
+ "is_moe": true,
1145
+ "routed_experts": {
1146
+ "median_var64": 0.098,
1147
+ "q25_var64": 0.0917,
1148
+ "q75_var64": 0.1083,
1149
+ "mean_s0": 4.0,
1150
+ "std_s0": 1.14,
1151
+ "mean_s0_ratio": 1.1,
1152
+ "n_experts": 256
1153
+ },
1154
+ "shared_expert": {
1155
+ "var64": 0.2166,
1156
+ "s0": 8.46,
1157
+ "shape": [
1158
+ 4096,
1159
+ 2048
1160
+ ]
1161
+ },
1162
+ "router": {
1163
+ "var64": 0.4608,
1164
+ "s0": 4.84,
1165
+ "s0_s1": 1.09,
1166
+ "shape": [
1167
+ 256,
1168
+ 4096
1169
+ ]
1170
+ }
1171
+ },
1172
+ "39": {
1173
+ "layer": 39,
1174
+ "is_moe": true,
1175
+ "routed_experts": {
1176
+ "median_var64": 0.1116,
1177
+ "q25_var64": 0.1027,
1178
+ "q75_var64": 0.1299,
1179
+ "mean_s0": 5.53,
1180
+ "std_s0": 2.34,
1181
+ "mean_s0_ratio": 1.25,
1182
+ "n_experts": 256
1183
+ },
1184
+ "shared_expert": {
1185
+ "var64": 0.2764,
1186
+ "s0": 8.83,
1187
+ "shape": [
1188
+ 4096,
1189
+ 2048
1190
+ ]
1191
+ },
1192
+ "router": {
1193
+ "var64": 0.4941,
1194
+ "s0": 4.11,
1195
+ "s0_s1": 1.06,
1196
+ "shape": [
1197
+ 256,
1198
+ 4096
1199
+ ]
1200
+ }
1201
+ },
1202
+ "40": {
1203
+ "layer": 40,
1204
+ "is_moe": true,
1205
+ "routed_experts": {
1206
+ "median_var64": 0.1027,
1207
+ "q25_var64": 0.0943,
1208
+ "q75_var64": 0.1168,
1209
+ "mean_s0": 5.01,
1210
+ "std_s0": 1.97,
1211
+ "mean_s0_ratio": 1.19,
1212
+ "n_experts": 256
1213
+ },
1214
+ "shared_expert": {
1215
+ "var64": 0.3241,
1216
+ "s0": 9.94,
1217
+ "shape": [
1218
+ 4096,
1219
+ 2048
1220
+ ]
1221
+ },
1222
+ "router": {
1223
+ "var64": 0.4982,
1224
+ "s0": 4.8,
1225
+ "s0_s1": 1.15,
1226
+ "shape": [
1227
+ 256,
1228
+ 4096
1229
+ ]
1230
+ }
1231
+ },
1232
+ "41": {
1233
+ "layer": 41,
1234
+ "is_moe": true,
1235
+ "routed_experts": {
1236
+ "median_var64": 0.0988,
1237
+ "q25_var64": 0.0921,
1238
+ "q75_var64": 0.1127,
1239
+ "mean_s0": 5.16,
1240
+ "std_s0": 2.69,
1241
+ "mean_s0_ratio": 1.16,
1242
+ "n_experts": 256
1243
+ },
1244
+ "shared_expert": {
1245
+ "var64": 0.2952,
1246
+ "s0": 8.3,
1247
+ "shape": [
1248
+ 4096,
1249
+ 2048
1250
+ ]
1251
+ },
1252
+ "router": {
1253
+ "var64": 0.49,
1254
+ "s0": 5.52,
1255
+ "s0_s1": 1.22,
1256
+ "shape": [
1257
+ 256,
1258
+ 4096
1259
+ ]
1260
+ }
1261
+ },
1262
+ "42": {
1263
+ "layer": 42,
1264
+ "is_moe": true,
1265
+ "routed_experts": {
1266
+ "median_var64": 0.0997,
1267
+ "q25_var64": 0.0938,
1268
+ "q75_var64": 0.1085,
1269
+ "mean_s0": 5.85,
1270
+ "std_s0": 4.75,
1271
+ "mean_s0_ratio": 1.18,
1272
+ "n_experts": 256
1273
+ },
1274
+ "shared_expert": {
1275
+ "var64": 0.2863,
1276
+ "s0": 16.44,
1277
+ "shape": [
1278
+ 4096,
1279
+ 2048
1280
+ ]
1281
+ },
1282
+ "router": {
1283
+ "var64": 0.5181,
1284
+ "s0": 6.69,
1285
+ "s0_s1": 1.35,
1286
+ "shape": [
1287
+ 256,
1288
+ 4096
1289
+ ]
1290
+ }
1291
+ }
1292
+ }