hackhackhack66666 commited on
Commit
f06a770
·
verified ·
1 Parent(s): 2fbe32f

Add measurements/calvin_s0_sr50_paired_fp16_3k_vs_bf16_5k.json

Browse files
measurements/calvin_s0_sr50_paired_fp16_3k_vs_bf16_5k.json ADDED
@@ -0,0 +1,479 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "label_a": "fp16@3000",
3
+ "label_b": "bf16_resume@5000",
4
+ "n_chains": 50,
5
+ "ckpt_a": "/home/askhabaliev_g/checkpoints/vla/calvin_s0/pins/step_3000/last.ckpt",
6
+ "ckpt_step_a": 3000,
7
+ "ckpt_b": "/home/askhabaliev_g/checkpoints/vla/calvin_s0_bf16_5k/latest/last.ckpt",
8
+ "ckpt_step_b": 5000,
9
+ "sr_a": [
10
+ 0.64,
11
+ 0.32,
12
+ 0.04,
13
+ 0.02,
14
+ 0.02
15
+ ],
16
+ "sr_b": [
17
+ 0.7,
18
+ 0.44,
19
+ 0.22,
20
+ 0.08,
21
+ 0.04
22
+ ],
23
+ "avg_len_a": 1.04,
24
+ "avg_len_b": 1.48,
25
+ "delta_avg_len": 0.43999999999999995,
26
+ "delta_avg_len_ci95": [
27
+ 0.0,
28
+ 0.86
29
+ ],
30
+ "delta_sr": [
31
+ 0.06,
32
+ 0.12,
33
+ 0.18,
34
+ 0.06,
35
+ 0.02
36
+ ],
37
+ "delta_sr_ci95": [
38
+ [
39
+ -0.12,
40
+ 0.24
41
+ ],
42
+ [
43
+ -0.040000000000000036,
44
+ 0.28
45
+ ],
46
+ [
47
+ 0.06,
48
+ 0.3
49
+ ],
50
+ [
51
+ -0.02,
52
+ 0.14
53
+ ],
54
+ [
55
+ -0.04,
56
+ 0.08
57
+ ]
58
+ ],
59
+ "sr1_discordant": {
60
+ "a_only": 9,
61
+ "b_only": 12
62
+ },
63
+ "p_delta_avg_len_le_0": 0.0307,
64
+ "n_boot": 10000,
65
+ "bootstrap_seed": 0,
66
+ "chains_improved": 22,
67
+ "chains_worsened": 11,
68
+ "chains_unchanged": 17,
69
+ "frac_delta_nonnegative": 0.78,
70
+ "per_chain_delta": [
71
+ {
72
+ "chain_id": 0,
73
+ "score_a": 0,
74
+ "score_b": 0,
75
+ "delta_score": 0
76
+ },
77
+ {
78
+ "chain_id": 1,
79
+ "score_a": 1,
80
+ "score_b": 1,
81
+ "delta_score": 0
82
+ },
83
+ {
84
+ "chain_id": 2,
85
+ "score_a": 0,
86
+ "score_b": 1,
87
+ "delta_score": 1
88
+ },
89
+ {
90
+ "chain_id": 3,
91
+ "score_a": 0,
92
+ "score_b": 1,
93
+ "delta_score": 1
94
+ },
95
+ {
96
+ "chain_id": 4,
97
+ "score_a": 2,
98
+ "score_b": 3,
99
+ "delta_score": 1
100
+ },
101
+ {
102
+ "chain_id": 5,
103
+ "score_a": 2,
104
+ "score_b": 0,
105
+ "delta_score": -2
106
+ },
107
+ {
108
+ "chain_id": 6,
109
+ "score_a": 1,
110
+ "score_b": 1,
111
+ "delta_score": 0
112
+ },
113
+ {
114
+ "chain_id": 7,
115
+ "score_a": 2,
116
+ "score_b": 5,
117
+ "delta_score": 3
118
+ },
119
+ {
120
+ "chain_id": 8,
121
+ "score_a": 1,
122
+ "score_b": 1,
123
+ "delta_score": 0
124
+ },
125
+ {
126
+ "chain_id": 9,
127
+ "score_a": 0,
128
+ "score_b": 3,
129
+ "delta_score": 3
130
+ },
131
+ {
132
+ "chain_id": 10,
133
+ "score_a": 0,
134
+ "score_b": 1,
135
+ "delta_score": 1
136
+ },
137
+ {
138
+ "chain_id": 11,
139
+ "score_a": 2,
140
+ "score_b": 2,
141
+ "delta_score": 0
142
+ },
143
+ {
144
+ "chain_id": 12,
145
+ "score_a": 1,
146
+ "score_b": 3,
147
+ "delta_score": 2
148
+ },
149
+ {
150
+ "chain_id": 13,
151
+ "score_a": 2,
152
+ "score_b": 2,
153
+ "delta_score": 0
154
+ },
155
+ {
156
+ "chain_id": 14,
157
+ "score_a": 0,
158
+ "score_b": 0,
159
+ "delta_score": 0
160
+ },
161
+ {
162
+ "chain_id": 15,
163
+ "score_a": 1,
164
+ "score_b": 5,
165
+ "delta_score": 4
166
+ },
167
+ {
168
+ "chain_id": 16,
169
+ "score_a": 1,
170
+ "score_b": 2,
171
+ "delta_score": 1
172
+ },
173
+ {
174
+ "chain_id": 17,
175
+ "score_a": 0,
176
+ "score_b": 2,
177
+ "delta_score": 2
178
+ },
179
+ {
180
+ "chain_id": 18,
181
+ "score_a": 1,
182
+ "score_b": 1,
183
+ "delta_score": 0
184
+ },
185
+ {
186
+ "chain_id": 19,
187
+ "score_a": 1,
188
+ "score_b": 1,
189
+ "delta_score": 0
190
+ },
191
+ {
192
+ "chain_id": 20,
193
+ "score_a": 5,
194
+ "score_b": 0,
195
+ "delta_score": -5
196
+ },
197
+ {
198
+ "chain_id": 21,
199
+ "score_a": 1,
200
+ "score_b": 0,
201
+ "delta_score": -1
202
+ },
203
+ {
204
+ "chain_id": 22,
205
+ "score_a": 0,
206
+ "score_b": 0,
207
+ "delta_score": 0
208
+ },
209
+ {
210
+ "chain_id": 23,
211
+ "score_a": 0,
212
+ "score_b": 3,
213
+ "delta_score": 3
214
+ },
215
+ {
216
+ "chain_id": 24,
217
+ "score_a": 1,
218
+ "score_b": 2,
219
+ "delta_score": 1
220
+ },
221
+ {
222
+ "chain_id": 25,
223
+ "score_a": 0,
224
+ "score_b": 1,
225
+ "delta_score": 1
226
+ },
227
+ {
228
+ "chain_id": 26,
229
+ "score_a": 1,
230
+ "score_b": 4,
231
+ "delta_score": 3
232
+ },
233
+ {
234
+ "chain_id": 27,
235
+ "score_a": 0,
236
+ "score_b": 2,
237
+ "delta_score": 2
238
+ },
239
+ {
240
+ "chain_id": 28,
241
+ "score_a": 3,
242
+ "score_b": 4,
243
+ "delta_score": 1
244
+ },
245
+ {
246
+ "chain_id": 29,
247
+ "score_a": 1,
248
+ "score_b": 0,
249
+ "delta_score": -1
250
+ },
251
+ {
252
+ "chain_id": 30,
253
+ "score_a": 0,
254
+ "score_b": 1,
255
+ "delta_score": 1
256
+ },
257
+ {
258
+ "chain_id": 31,
259
+ "score_a": 0,
260
+ "score_b": 2,
261
+ "delta_score": 2
262
+ },
263
+ {
264
+ "chain_id": 32,
265
+ "score_a": 0,
266
+ "score_b": 0,
267
+ "delta_score": 0
268
+ },
269
+ {
270
+ "chain_id": 33,
271
+ "score_a": 2,
272
+ "score_b": 2,
273
+ "delta_score": 0
274
+ },
275
+ {
276
+ "chain_id": 34,
277
+ "score_a": 1,
278
+ "score_b": 0,
279
+ "delta_score": -1
280
+ },
281
+ {
282
+ "chain_id": 35,
283
+ "score_a": 0,
284
+ "score_b": 0,
285
+ "delta_score": 0
286
+ },
287
+ {
288
+ "chain_id": 36,
289
+ "score_a": 2,
290
+ "score_b": 1,
291
+ "delta_score": -1
292
+ },
293
+ {
294
+ "chain_id": 37,
295
+ "score_a": 1,
296
+ "score_b": 0,
297
+ "delta_score": -1
298
+ },
299
+ {
300
+ "chain_id": 38,
301
+ "score_a": 2,
302
+ "score_b": 2,
303
+ "delta_score": 0
304
+ },
305
+ {
306
+ "chain_id": 39,
307
+ "score_a": 2,
308
+ "score_b": 0,
309
+ "delta_score": -2
310
+ },
311
+ {
312
+ "chain_id": 40,
313
+ "score_a": 2,
314
+ "score_b": 2,
315
+ "delta_score": 0
316
+ },
317
+ {
318
+ "chain_id": 41,
319
+ "score_a": 0,
320
+ "score_b": 0,
321
+ "delta_score": 0
322
+ },
323
+ {
324
+ "chain_id": 42,
325
+ "score_a": 2,
326
+ "score_b": 2,
327
+ "delta_score": 0
328
+ },
329
+ {
330
+ "chain_id": 43,
331
+ "score_a": 2,
332
+ "score_b": 1,
333
+ "delta_score": -1
334
+ },
335
+ {
336
+ "chain_id": 44,
337
+ "score_a": 0,
338
+ "score_b": 1,
339
+ "delta_score": 1
340
+ },
341
+ {
342
+ "chain_id": 45,
343
+ "score_a": 0,
344
+ "score_b": 3,
345
+ "delta_score": 3
346
+ },
347
+ {
348
+ "chain_id": 46,
349
+ "score_a": 1,
350
+ "score_b": 3,
351
+ "delta_score": 2
352
+ },
353
+ {
354
+ "chain_id": 47,
355
+ "score_a": 1,
356
+ "score_b": 0,
357
+ "delta_score": -1
358
+ },
359
+ {
360
+ "chain_id": 48,
361
+ "score_a": 2,
362
+ "score_b": 3,
363
+ "delta_score": 1
364
+ },
365
+ {
366
+ "chain_id": 49,
367
+ "score_a": 2,
368
+ "score_b": 0,
369
+ "delta_score": -2
370
+ }
371
+ ],
372
+ "wilson95_a": [
373
+ {
374
+ "k": 1,
375
+ "successes": 32,
376
+ "n": 50,
377
+ "p": 0.64,
378
+ "wilson95": [
379
+ 0.501410168761127,
380
+ 0.7586125034815325
381
+ ]
382
+ },
383
+ {
384
+ "k": 2,
385
+ "successes": 16,
386
+ "n": 50,
387
+ "p": 0.32,
388
+ "wilson95": [
389
+ 0.20758216260656845,
390
+ 0.45810297308144077
391
+ ]
392
+ },
393
+ {
394
+ "k": 3,
395
+ "successes": 2,
396
+ "n": 50,
397
+ "p": 0.04,
398
+ "wilson95": [
399
+ 0.011038884327619805,
400
+ 0.1346009068750702
401
+ ]
402
+ },
403
+ {
404
+ "k": 4,
405
+ "successes": 1,
406
+ "n": 50,
407
+ "p": 0.02,
408
+ "wilson95": [
409
+ 0.003539259271646236,
410
+ 0.10495443589637815
411
+ ]
412
+ },
413
+ {
414
+ "k": 5,
415
+ "successes": 1,
416
+ "n": 50,
417
+ "p": 0.02,
418
+ "wilson95": [
419
+ 0.003539259271646236,
420
+ 0.10495443589637815
421
+ ]
422
+ }
423
+ ],
424
+ "wilson95_b": [
425
+ {
426
+ "k": 1,
427
+ "successes": 35,
428
+ "n": 50,
429
+ "p": 0.7,
430
+ "wilson95": [
431
+ 0.562496495355466,
432
+ 0.8089644649911906
433
+ ]
434
+ },
435
+ {
436
+ "k": 2,
437
+ "successes": 22,
438
+ "n": 50,
439
+ "p": 0.44,
440
+ "wilson95": [
441
+ 0.3116219921125717,
442
+ 0.5769397197834315
443
+ ]
444
+ },
445
+ {
446
+ "k": 3,
447
+ "successes": 11,
448
+ "n": 50,
449
+ "p": 0.22,
450
+ "wilson95": [
451
+ 0.1275391597021442,
452
+ 0.3524154958125367
453
+ ]
454
+ },
455
+ {
456
+ "k": 4,
457
+ "successes": 4,
458
+ "n": 50,
459
+ "p": 0.08,
460
+ "wilson95": [
461
+ 0.03154951408859298,
462
+ 0.18838246918342838
463
+ ]
464
+ },
465
+ {
466
+ "k": 5,
467
+ "successes": 2,
468
+ "n": 50,
469
+ "p": 0.04,
470
+ "wilson95": [
471
+ 0.011038884327619805,
472
+ 0.1346009068750702
473
+ ]
474
+ }
475
+ ],
476
+ "branch_a": "FP16 0-4k, stopped after repeated non-finite forwards. SR-50 is the step-3000 pin.",
477
+ "branch_b": "BF16 resume from the FP16 step-4000 pin to step 5000. Not a continuation of the FP16 curve and not the from-scratch BF16 seed.",
478
+ "verdict": "no significant difference (CI includes 0)"
479
+ }