AbstractPhil commited on
Commit
42eb8ed
·
verified ·
1 Parent(s): 4023fb3

Frozen development lookup/parity: H16-seed1

Browse files
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/anchor_angles_off-queries.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0b4931a389baea83237de32de8150b5aeecced1d9e3aeafc510540e6d689b5f
3
+ size 122352
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/anchor_angles_off/dev-sts-predictions.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd35430d366331fe866d51567fca3738d95f6bd9afcfab17a1b57ae5d1696673
3
+ size 18514
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/intact-queries.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bae71974fee4922ed556a7295996290f14182cf055616eb5568cdd2c0674b5cf
3
+ size 122331
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/intact/dev-sts-predictions.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11c88c26d64d5e5a61f8238eab13e7b08455390706f2fe7c6b5fc6bbb2d14529
3
+ size 18514
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/report.json ADDED
@@ -0,0 +1,416 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "H16",
3
+ "seed": 1,
4
+ "config_id": "fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e",
5
+ "dataset_id": "659b21c1c6eb1074eeec7acdbfa9bd010965595a4669338690cf3ae38a4448d5",
6
+ "split": "chunk061 development",
7
+ "rows": 4096,
8
+ "row_seed": 20261106,
9
+ "row_ids_sha256": "e72547c65ba2f4d26131988027abd9004af0a350e3bd404242b0f843779eb8ed",
10
+ "precision": "fp32 TF32 off",
11
+ "measurements": {
12
+ "intact": {
13
+ "all": {
14
+ "queries": 10146,
15
+ "captions": 1935,
16
+ "record_mass": 0.02546592243015766,
17
+ "caption_mean_record_mass": 0.03954405295704696,
18
+ "median_record_mass": 0.010414542630314827,
19
+ "exact_record_top1": 0.015572639463828109,
20
+ "wrong_adjacent_mass": 0.011977557092905045,
21
+ "uniform_mass": 0.013713696040213108
22
+ },
23
+ "ending": {
24
+ "queries": 9483,
25
+ "captions": 1935,
26
+ "record_mass": 0.02607988379895687,
27
+ "caption_mean_record_mass": 0.040105033063934994,
28
+ "median_record_mass": 0.010789166204631329,
29
+ "exact_record_top1": 0.015712327322577244,
30
+ "wrong_adjacent_mass": 0.012126941233873367,
31
+ "uniform_mass": 0.013700349256396294
32
+ },
33
+ "non_ending": {
34
+ "queries": 663,
35
+ "captions": 299,
36
+ "record_mass": 0.01668434403836727,
37
+ "caption_mean_record_mass": 0.018355159964207314,
38
+ "median_record_mass": 0.006087308283895254,
39
+ "exact_record_top1": 0.013574660633484163,
40
+ "wrong_adjacent_mass": 0.009840885177254677,
41
+ "uniform_mass": 0.013904616236686707
42
+ },
43
+ "continuation": {
44
+ "queries": 663,
45
+ "captions": 299,
46
+ "record_mass": 0.021683087572455406,
47
+ "caption_mean_record_mass": 0.025753207746851314,
48
+ "median_record_mass": 0.0068626850843429565,
49
+ "exact_record_top1": 0.013574660633484163,
50
+ "wrong_adjacent_mass": 0.013456040062010288,
51
+ "uniform_mass": 0.013904616236686707
52
+ },
53
+ "above128": {
54
+ "queries": 3198,
55
+ "captions": 252,
56
+ "record_mass": 0.012743870727717876,
57
+ "caption_mean_record_mass": 0.012710957753745514,
58
+ "median_record_mass": 0.005444246344268322,
59
+ "exact_record_top1": 0.007817385866166354,
60
+ "wrong_adjacent_mass": 0.0054250722751021385,
61
+ "uniform_mass": 0.006118615623563528
62
+ }
63
+ },
64
+ "anchor_angles_off": {
65
+ "all": {
66
+ "queries": 10146,
67
+ "captions": 1935,
68
+ "record_mass": 0.03424637019634247,
69
+ "caption_mean_record_mass": 0.04698193805513047,
70
+ "median_record_mass": 0.013876909390091896,
71
+ "exact_record_top1": 0.026808594520007886,
72
+ "wrong_adjacent_mass": 0.013267986476421356,
73
+ "uniform_mass": 0.013713696040213108
74
+ },
75
+ "ending": {
76
+ "queries": 9483,
77
+ "captions": 1935,
78
+ "record_mass": 0.035177428275346756,
79
+ "caption_mean_record_mass": 0.04776747371852205,
80
+ "median_record_mass": 0.014343569055199623,
81
+ "exact_record_top1": 0.027628387641041863,
82
+ "wrong_adjacent_mass": 0.013516614213585854,
83
+ "uniform_mass": 0.013700349256396294
84
+ },
85
+ "non_ending": {
86
+ "queries": 663,
87
+ "captions": 299,
88
+ "record_mass": 0.020929310470819473,
89
+ "caption_mean_record_mass": 0.022358726953910852,
90
+ "median_record_mass": 0.008146652951836586,
91
+ "exact_record_top1": 0.015082956259426848,
92
+ "wrong_adjacent_mass": 0.009711815044283867,
93
+ "uniform_mass": 0.013904616236686707
94
+ },
95
+ "continuation": {
96
+ "queries": 663,
97
+ "captions": 299,
98
+ "record_mass": 0.03242213651537895,
99
+ "caption_mean_record_mass": 0.03743410519207244,
100
+ "median_record_mass": 0.011551249772310257,
101
+ "exact_record_top1": 0.027149321266968326,
102
+ "wrong_adjacent_mass": 0.01300971582531929,
103
+ "uniform_mass": 0.013904616236686707
104
+ },
105
+ "above128": {
106
+ "queries": 3198,
107
+ "captions": 252,
108
+ "record_mass": 0.021344158798456192,
109
+ "caption_mean_record_mass": 0.021484570800870035,
110
+ "median_record_mass": 0.00797557458281517,
111
+ "exact_record_top1": 0.01876172607879925,
112
+ "wrong_adjacent_mass": 0.0067310393787920475,
113
+ "uniform_mass": 0.006118615623563528
114
+ }
115
+ },
116
+ "value_addresses_constant": {
117
+ "all": {
118
+ "queries": 10146,
119
+ "captions": 1935,
120
+ "record_mass": 0.025496799498796463,
121
+ "caption_mean_record_mass": 0.039553816642194685,
122
+ "median_record_mass": 0.01042714063078165,
123
+ "exact_record_top1": 0.01537551744529864,
124
+ "wrong_adjacent_mass": 0.011988874524831772,
125
+ "uniform_mass": 0.013713696040213108
126
+ },
127
+ "ending": {
128
+ "queries": 9483,
129
+ "captions": 1935,
130
+ "record_mass": 0.026106538251042366,
131
+ "caption_mean_record_mass": 0.0401140916401891,
132
+ "median_record_mass": 0.010775180533528328,
133
+ "exact_record_top1": 0.015501423600126542,
134
+ "wrong_adjacent_mass": 0.012142317369580269,
135
+ "uniform_mass": 0.013700349256396294
136
+ },
137
+ "non_ending": {
138
+ "queries": 663,
139
+ "captions": 299,
140
+ "record_mass": 0.01677563413977623,
141
+ "caption_mean_record_mass": 0.01844043998930996,
142
+ "median_record_mass": 0.006101549603044987,
143
+ "exact_record_top1": 0.013574660633484163,
144
+ "wrong_adjacent_mass": 0.009794158861041069,
145
+ "uniform_mass": 0.013904616236686707
146
+ },
147
+ "continuation": {
148
+ "queries": 663,
149
+ "captions": 299,
150
+ "record_mass": 0.02177565172314644,
151
+ "caption_mean_record_mass": 0.025867332676380802,
152
+ "median_record_mass": 0.0068817432038486,
153
+ "exact_record_top1": 0.013574660633484163,
154
+ "wrong_adjacent_mass": 0.013361276127398014,
155
+ "uniform_mass": 0.013904616236686707
156
+ },
157
+ "above128": {
158
+ "queries": 3198,
159
+ "captions": 252,
160
+ "record_mass": 0.01276535727083683,
161
+ "caption_mean_record_mass": 0.012722745237480258,
162
+ "median_record_mass": 0.00551917590200901,
163
+ "exact_record_top1": 0.007191994996873046,
164
+ "wrong_adjacent_mass": 0.0054713888093829155,
165
+ "uniform_mass": 0.006118615623563528
166
+ }
167
+ }
168
+ },
169
+ "sentence_measurements": {
170
+ "intact": {
171
+ "evaluation": {
172
+ "heldout": {
173
+ "n": 5000,
174
+ "erank": 95.62741489359365,
175
+ "self_cos": 0.003444665011627598,
176
+ "cos_to_target": 0.8080115914344788
177
+ },
178
+ "above_128_tokens": {
179
+ "n": 5000,
180
+ "erank": 73.03811752294244,
181
+ "self_cos": 0.011567466648996574,
182
+ "cos_to_target": 0.7362988591194153
183
+ },
184
+ "heldout_total_rows": 500000,
185
+ "above_128_total_rows": 29240
186
+ },
187
+ "development": {
188
+ "spearman": 0.6833117099948678,
189
+ "n": 1500,
190
+ "split": "validation",
191
+ "revision": "96943a16ea6a35129e253c659081cb59daf81b30",
192
+ "max_tokens": 64,
193
+ "used_for_selection": true
194
+ },
195
+ "precision": "legacy FP16 autocast evaluation; FP32 parameters; TF32 off"
196
+ },
197
+ "anchor_angles_off": {
198
+ "evaluation": {
199
+ "heldout": {
200
+ "n": 5000,
201
+ "erank": 88.9532024089657,
202
+ "self_cos": 0.009525986861605953,
203
+ "cos_to_target": 0.8086338043212891
204
+ },
205
+ "above_128_tokens": {
206
+ "n": 5000,
207
+ "erank": 60.776641035969774,
208
+ "self_cos": 0.1497647591973433,
209
+ "cos_to_target": 0.7148573994636536
210
+ },
211
+ "heldout_total_rows": 500000,
212
+ "above_128_total_rows": 29240
213
+ },
214
+ "development": {
215
+ "spearman": 0.6920672228800693,
216
+ "n": 1500,
217
+ "split": "validation",
218
+ "revision": "96943a16ea6a35129e253c659081cb59daf81b30",
219
+ "max_tokens": 64,
220
+ "used_for_selection": true
221
+ },
222
+ "precision": "legacy FP16 autocast evaluation; FP32 parameters; TF32 off"
223
+ },
224
+ "value_addresses_constant": {
225
+ "evaluation": {
226
+ "heldout": {
227
+ "n": 5000,
228
+ "erank": 95.62237220707621,
229
+ "self_cos": 0.003417102724543398,
230
+ "cos_to_target": 0.8079468607902527
231
+ },
232
+ "above_128_tokens": {
233
+ "n": 5000,
234
+ "erank": 72.97626574191753,
235
+ "self_cos": 0.011692005125623916,
236
+ "cos_to_target": 0.7362787127494812
237
+ },
238
+ "heldout_total_rows": 500000,
239
+ "above_128_total_rows": 29240
240
+ },
241
+ "development": {
242
+ "spearman": 0.6832543985674162,
243
+ "n": 1500,
244
+ "split": "validation",
245
+ "revision": "96943a16ea6a35129e253c659081cb59daf81b30",
246
+ "max_tokens": 64,
247
+ "used_for_selection": true
248
+ },
249
+ "precision": "legacy FP16 autocast evaluation; FP32 parameters; TF32 off"
250
+ }
251
+ },
252
+ "independent_checkpoint_auto_parity": {
253
+ "reference": "independent final state.pt model tensors",
254
+ "format": "model.safetensors",
255
+ "metadata_format": "pt",
256
+ "dtype": "fp32",
257
+ "state_tensors_equal": true,
258
+ "loading_info": {
259
+ "missing_keys": [],
260
+ "unexpected_keys": [],
261
+ "mismatched_keys": [],
262
+ "error_msgs": []
263
+ },
264
+ "cases": [
265
+ {
266
+ "case": "real_captions",
267
+ "input_shape": [
268
+ 2,
269
+ 106
270
+ ],
271
+ "embedding_shape": [
272
+ 2,
273
+ 768
274
+ ],
275
+ "hidden_shape": [
276
+ 2,
277
+ 106,
278
+ 512
279
+ ],
280
+ "max_embedding_delta": 0.0,
281
+ "max_hidden_delta": 0.0,
282
+ "seconds": 0.18057174799832865
283
+ },
284
+ {
285
+ "case": "all_padding",
286
+ "input_shape": [
287
+ 2,
288
+ 8
289
+ ],
290
+ "embedding_shape": [
291
+ 2,
292
+ 768
293
+ ],
294
+ "hidden_shape": [
295
+ 2,
296
+ 8,
297
+ 512
298
+ ],
299
+ "max_embedding_delta": 0.0,
300
+ "max_hidden_delta": 0.0,
301
+ "seconds": 0.05103075899751275
302
+ },
303
+ {
304
+ "case": "synthetic_length_1",
305
+ "input_shape": [
306
+ 1,
307
+ 1
308
+ ],
309
+ "embedding_shape": [
310
+ 1,
311
+ 768
312
+ ],
313
+ "hidden_shape": [
314
+ 1,
315
+ 1,
316
+ 512
317
+ ],
318
+ "max_embedding_delta": 0.0,
319
+ "max_hidden_delta": 0.0,
320
+ "seconds": 0.01124924000032479
321
+ },
322
+ {
323
+ "case": "synthetic_length_17",
324
+ "input_shape": [
325
+ 1,
326
+ 17
327
+ ],
328
+ "embedding_shape": [
329
+ 1,
330
+ 768
331
+ ],
332
+ "hidden_shape": [
333
+ 1,
334
+ 17,
335
+ 512
336
+ ],
337
+ "max_embedding_delta": 0.0,
338
+ "max_hidden_delta": 0.0,
339
+ "seconds": 0.008814799999527168
340
+ },
341
+ {
342
+ "case": "synthetic_length_128",
343
+ "input_shape": [
344
+ 1,
345
+ 128
346
+ ],
347
+ "embedding_shape": [
348
+ 1,
349
+ 768
350
+ ],
351
+ "hidden_shape": [
352
+ 1,
353
+ 128,
354
+ 512
355
+ ],
356
+ "max_embedding_delta": 0.0,
357
+ "max_hidden_delta": 0.0,
358
+ "seconds": 0.008224659999541473
359
+ },
360
+ {
361
+ "case": "synthetic_length_512",
362
+ "input_shape": [
363
+ 1,
364
+ 512
365
+ ],
366
+ "embedding_shape": [
367
+ 1,
368
+ 768
369
+ ],
370
+ "hidden_shape": [
371
+ 1,
372
+ 512,
373
+ 512
374
+ ],
375
+ "max_embedding_delta": 0.0,
376
+ "max_hidden_delta": 0.0,
377
+ "seconds": 0.009231739997630939
378
+ },
379
+ {
380
+ "case": "synthetic_length_8192",
381
+ "input_shape": [
382
+ 1,
383
+ 8192
384
+ ],
385
+ "embedding_shape": [
386
+ 1,
387
+ 768
388
+ ],
389
+ "hidden_shape": [
390
+ 1,
391
+ 8192,
392
+ 512
393
+ ],
394
+ "max_embedding_delta": 0.0,
395
+ "max_hidden_delta": 0.0,
396
+ "seconds": 0.17464379699958954
397
+ }
398
+ ]
399
+ },
400
+ "initialization_control": null,
401
+ "probe_source_sha256": "c778258cabc69e1985c03304af0ebf2e8e4be3ad64aba680d9ccaec4bdd5d03a",
402
+ "environment": {
403
+ "python": "3.13.16",
404
+ "torch": "2.15.0+cu132",
405
+ "cuda": "13.2",
406
+ "gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition"
407
+ },
408
+ "memory": {
409
+ "peak_allocated_gb": 0.761546752,
410
+ "peak_reserved_gb": 1.088421888,
411
+ "max_observed_device_gb": 1.842348032,
412
+ "total_device_limit_gb": 92.0
413
+ },
414
+ "selection_use": false,
415
+ "training_performed": false
416
+ }
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/row-ids.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:553ab8315c50cd610d501ef29f60ad46d2d639677ee0cc1329a893e574c87fe8
3
+ size 32896
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/tensorboard/events.out.tfevents.1793995846.79fe39615b31.111574.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7baf37c0e6ee4aaa6a1b235ff380f148d9135286fe1c12e42d0a37b4f4be8386
3
+ size 15662
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/value_addresses_constant-queries.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13f9a21ddcfdffbd5e1505dbca880370c7c19894dff184270a2c090413a31394
3
+ size 122268
experiments/fad07b2a05857a7abef278004986cacc1e3b8eb89b70943668505ea90498525e/H16-seed1/frozen-followup-59ec95344958/value_addresses_constant/dev-sts-predictions.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b0fd189d2fa2a66ebb2b9546d1cf7c67e1f8ab785c934bae06568f7e4e76d86
3
+ size 18514