memlayers-data / validation_report.json
roieg's picture
data-v1: the frozen bundle (59 hashed files, data_sha256 bb1bfd40)
db21410 verified
Raw History Blame Contribute Delete
18.1 kB
{
"checks": [
{
"detail": "59 files hashed; 0 changed, 0 added, 0 removed",
"id": "A1",
"numbers": {
"changed": 0,
"changed_paths": [],
"files": 59
},
"seconds": 253.0,
"status": "PASS",
"title": "Rebuild is byte-identical"
},
{
"detail": "59/59 files verified against v1.json (0 mismatched, 0 missing)",
"id": "A2",
"numbers": {
"bad": [],
"missing": [],
"ok": 59
},
"seconds": 0.4,
"status": "PASS",
"title": "`data.fetch --verify-only` passes"
},
{
"detail": "every Pile subject exactly 27,000,000 in train.bin; every held-out half within 1% of 2,500,000 per subject",
"id": "A3",
"numbers": {
"bio": 15142237,
"heldout": {
"label": {
"FreeLaw": 2501133,
"Github": 2500326,
"Pile-CC": 2501942,
"PubMed Central": 2503922,
"USPTO Backgrounds": 2500388,
"Wikipedia (en)": 2500790
},
"test": {
"FreeLaw": 2502989,
"Github": 2502662,
"Pile-CC": 2501855,
"PubMed Central": 2502199,
"USPTO Backgrounds": 2500283,
"Wikipedia (en)": 2500584
}
},
"train": {
"FreeLaw": 27000000,
"Github": 27000000,
"Pile-CC": 27000000,
"PubMed Central": 27000000,
"Wikipedia (en)": 27000000
}
},
"seconds": 0.0,
"status": "PASS",
"title": "Token counts per subject and half"
},
{
"detail": "worst deviation 0.416 points across 598 overlapping windows (stride 250,000), tolerance 2.0",
"id": "A4",
"numbers": {
"n_windows": 598,
"worst": {
"FreeLaw": {
"at_token": 85000000,
"target_share": 0.179829,
"worst_deviation_points": 0.3938,
"worst_share": 0.175892
},
"Github": {
"at_token": 71750000,
"target_share": 0.179829,
"worst_deviation_points": 0.3518,
"worst_share": 0.176311
},
"Pile-CC": {
"at_token": 15500000,
"target_share": 0.179829,
"worst_deviation_points": 0.416,
"worst_share": 0.17567
},
"PubMed Central": {
"at_token": 110750000,
"target_share": 0.179829,
"worst_deviation_points": 0.4143,
"worst_share": 0.175687
},
"Wikipedia (en)": {
"at_token": 128000000,
"target_share": 0.179829,
"worst_deviation_points": 0.3908,
"worst_share": 0.183737
},
"bio": {
"at_token": 147000000,
"target_share": 0.100853,
"worst_deviation_points": 0.168,
"worst_share": 0.102533
}
}
},
"seconds": 2.9,
"status": "PASS",
"title": "Shares in every 1,000,000-token window"
},
{
"detail": "train.bin: 150,475,264 tokens, 332,886 segments, 333,027 EOT separator/pad tokens, max id 50256 | label.bin: 15,021,806 tokens, 13,306 segments, 13,305 EOT separator/pad tokens, max id 50256 | test.bin: 15,023,952 tokens, 13,381 segments, 13,380 EOT separator/pad tokens, max id 50256",
"id": "A5",
"numbers": {
"label.bin": {
"covered": 15008501,
"eot_gap_tokens": 13305,
"max_id": 50256,
"segments": 13306,
"tokens": 15021806
},
"test.bin": {
"covered": 15010572,
"eot_gap_tokens": 13380,
"max_id": 50256,
"segments": 13381,
"tokens": 15023952
},
"train.bin": {
"covered": 150142237,
"eot_gap_tokens": 333027,
"max_id": 50256,
"segments": 332886,
"tokens": 150475264
}
},
"seconds": 0.3,
"status": "PASS",
"title": "Format: dtype, vocab bound, EOT, index tiling"
},
{
"detail": "1,000 random segments checked token for token (820 of them whole documents, also compared as text); 0 mismatches",
"id": "A6",
"numbers": {
"mismatches": [],
"sampled": 1000,
"whole_documents": 820
},
"seconds": 2.8,
"status": "PASS",
"title": "Decoding index segments reproduces the source"
},
{
"detail": "ids disjoint across train/label/test; 0 exact-hash matches; 0 documents over 50%; worst single document 50.0%; aggregates Pile-CC=0.64%, Github=14.35%, PubMed=2.93%, FreeLaw=9.21%, Wikipedia=2.14%, USPTO=0.00%",
"id": "A7",
"numbers": {
"documents_over_threshold": 0,
"exact_duplicates": 0,
"ids": {
"label": 13306,
"test": 13381,
"train": 113247
},
"ngram": {
"FreeLaw": {
"aggregate": 0.092085,
"ceiling": 0.15,
"max_document": 0.5,
"train_13grams": 28718183
},
"Github": {
"aggregate": 0.143485,
"ceiling": 0.25,
"max_document": 0.5,
"train_13grams": 22467753
},
"Pile-CC": {
"aggregate": 0.006409,
"ceiling": 0.05,
"max_document": 0.495238,
"train_13grams": 30818919
},
"PubMed Central": {
"aggregate": 0.029253,
"ceiling": 0.05,
"max_document": 0.492899,
"train_13grams": 30670380
},
"USPTO Backgrounds": {
"aggregate": 0.0,
"ceiling": 0.05,
"max_document": 0.0,
"note": "no training text"
},
"Wikipedia (en)": {
"aggregate": 0.021441,
"ceiling": 0.05,
"max_document": 0.5,
"train_13grams": 31062799
}
}
},
"seconds": 88.3,
"status": "PASS",
"title": "Leaks: ids, exact hashes, 13-gram overlap"
},
{
"detail": "halves share 0 documents; smallest half/subject is 708 documents (minimum 500); longest document 4096 tokens (cap 4096)",
"id": "A8",
"numbers": {
"documents": {
"label": {
"FreeLaw": 943,
"Github": 1762,
"Pile-CC": 3042,
"PubMed Central": 708,
"USPTO Backgrounds": 3029,
"Wikipedia (en)": 3822
},
"test": {
"FreeLaw": 995,
"Github": 1815,
"Pile-CC": 3087,
"PubMed Central": 713,
"USPTO Backgrounds": 3084,
"Wikipedia (en)": 3687
}
},
"shared": 0
},
"seconds": 0.0,
"status": "PASS",
"title": "Halves: disjoint, >=500 docs, <=4096 tokens"
},
{
"detail": "['USPTO Backgrounds'] appear in 0 train.bin segments by document id and 0 by subject label; train.bin carries ['FreeLaw', 'Github', 'Pile-CC', 'PubMed Central', 'Wikipedia (en)', 'bio']",
"id": "A9",
"numbers": {
"heldout_only": [
"USPTO Backgrounds"
],
"segments": 0
},
"seconds": 0.0,
"status": "PASS",
"title": "No held-out-only subject in train.bin"
},
{
"detail": "142,886 Pile segments: index subject == raw-archive meta.pile_set_name == the document id's slug; 190,000 biography segments all labelled `bio` with a known person id",
"id": "A10",
"numbers": {
"bio_segments": 190000,
"pile_segments": 142886,
"problems": [],
"raw_documents": 177718
},
"seconds": 4.9,
"status": "PASS",
"title": "Subject labels agree with the source"
},
{
"detail": "accuracy 94.87% on 13,381 test-half documents (threshold 90%); per subject Pile-CC=89.7%, Github=97.8%, PubMed=97.6%, FreeLaw=99.1%, Wikipedia=95.0%, USPTO=96.2%",
"id": "A11",
"numbers": {
"accuracy": 0.948733,
"confusion": [
[
2769,
14,
29,
18,
210,
47
],
[
13,
1775,
22,
2,
2,
1
],
[
12,
2,
696,
0,
3,
0
],
[
4,
3,
0,
986,
0,
2
],
[
86,
3,
40,
9,
3503,
46
],
[
25,
0,
91,
0,
2,
2966
]
],
"documents": 13381,
"per_subject_recall": {
"FreeLaw": 0.991,
"Github": 0.978,
"Pile-CC": 0.897,
"PubMed Central": 0.9762,
"USPTO Backgrounds": 0.9617,
"Wikipedia (en)": 0.9501
},
"subjects": [
"Pile-CC",
"Github",
"PubMed Central",
"FreeLaw",
"Wikipedia (en)",
"USPTO Backgrounds"
]
},
"seconds": 0.5,
"status": "PASS",
"title": "Bag-of-words subject classifier (sanity)"
},
{
"detail": "2,500 people, names unique and absent from the Pile, genders {'female': 1250, 'male': 1250}, 5 facts each; first-token uniqueness holds per list (first two for birth_date, which has only 12 distinct first tokens); chi-square p birth_date=0.124, birth_city=0.877, university=0.183, major=0.940, employer=0.347",
"id": "A12",
"numbers": {
"by_gender": {
"female": 1250,
"male": 1250
},
"chisquare_p": {
"birth_city": 0.876588,
"birth_date": 0.123935,
"employer": 0.347052,
"major": 0.940421,
"university": 0.182991
},
"missing_values_in_rewordings": 0,
"names_in_pile": 0,
"people": 2500,
"tokens": {
"birth_city": {
"distinct_first": 200,
"prefix_tokens": 1,
"unique": true,
"values": 200
},
"birth_date": {
"distinct_first": 12,
"prefix_tokens": 2,
"unique": true,
"values": 200
},
"employer": {
"distinct_first": 250,
"prefix_tokens": 1,
"unique": true,
"values": 250
},
"major": {
"distinct_first": 100,
"prefix_tokens": 1,
"unique": true,
"values": 100
},
"university": {
"distinct_first": 300,
"prefix_tokens": 1,
"unique": true,
"values": 300
}
}
},
"seconds": 0.0,
"status": "PASS",
"title": "Biography facts"
},
{
"detail": "100,000 rows: **0** leak the value verbatim (0), **0** via the person's name (0), **0** via another of the person's values (0); 50,000 training prompts are exact prefixes of their biography; 50,000 paraphrase prompts, 200 of them searched in full against train.bin and absent; held-out markers inside biography text: 'Question:'=0, 'When it comes to'=0, 'is on record as'=0, 'reads as follows'=0 (the same markers occur 571 times in ordinary Pile text, which is not a leak); REPORTED: 48 rows whose ranking key is a TEMPLATE word (birth_city=Thessaloniki, Greece, employer=Heineken, major=Theology, university=Aalborg University)",
"id": "A13",
"numbers": {
"bio_marker_hits": {
"Question:": 0,
"When it comes to": 0,
"is on record as": 0,
"reads as follows": 0
},
"leak_value_verbatim": [],
"leak_via_name": [],
"leak_via_other_value": [],
"n_leak_value_verbatim": 0,
"n_leak_via_name": 0,
"n_leak_via_other_value": 0,
"paraphrase_prompts": 50000,
"pile_marker_hits": {
"Question:": 102,
"When it comes to": 275,
"is on record as": 2,
"reads as follows": 192
},
"rows": 100000,
"template_word_rows": 48,
"template_word_values": {
"birth_city=Thessaloniki, Greece": 6,
"employer=Heineken": 17,
"major=Theology": 20,
"university=Aalborg University": 5
},
"train_prompts": 50000
},
"seconds": 8.0,
"status": "PASS",
"title": "Fact prompts: no answer leaks"
},
{
"detail": "190,000 copies over 2,000 trained people, every count equal to its tier target and spread over the 4 rewordings; 0 cross a 512-token boundary; 0 of 500 control names appear in train text",
"id": "A14",
"numbers": {
"control_names_found": [],
"copies": 190000,
"crossing": 0,
"people_with_copies": 2000
},
"seconds": 3.4,
"status": "PASS",
"title": "Exposure counts and control isolation"
},
{
"detail": "random scorer: mean rank fraction 0.502, top-1 0.558% over 36,000 prompts; untrained model: 0.501, top-1 0.525% over 4,000; mean candidate list 210",
"id": "A15",
"numbers": {
"random": {
"candidates_mean": 210.0,
"mean_margin": -2.7352,
"mean_rank": 105.448,
"mean_rank_fraction": 0.502106,
"median_rank": 92.0,
"n": 36000,
"top1": 0.005583
},
"rank_fraction_by_attribute": {
"random/birth_city": 0.50306,
"random/birth_date": 0.504117,
"random/employer": 0.502423,
"random/major": 0.5004,
"random/university": 0.50053,
"untrained/birth_city": 0.505019,
"untrained/birth_date": 0.484719,
"untrained/employer": 0.499975,
"untrained/major": 0.519098,
"untrained/university": 0.49444
},
"untrained": {
"candidates_mean": 210.0,
"mean_margin": -1.1695,
"mean_rank": 104.635,
"mean_rank_fraction": 0.500638,
"median_rank": 92.0,
"n": 4000,
"top1": 0.00525
}
},
"seconds": 10.0,
"status": "PASS",
"title": "Probe floor: random and untrained sit mid-list"
},
{
"detail": "293,896 sequences of 512; same seed reproduces the order, a different seed does not, the order is a permutation, and a resume from (seed, position) matches an unbroken run",
"id": "A16",
"numbers": {
"n_sequences": 293896,
"resume_position": 12
},
"seconds": 0.1,
"status": "PASS",
"title": "Loader: order, resume, shapes"
},
{
"detail": "Pile-CC: model 6.7886 vs unigram 7.559; Github: model 5.7371 vs unigram 6.3888; PubMed: model 6.5792 vs unigram 7.4554; FreeLaw: model 5.8735 vs unigram 6.8493; Wikipedia: model 7.0163 vs unigram 7.72; USPTO: model 7.0021 vs unigram 7.9605",
"id": "A17",
"numbers": {
"FreeLaw": {
"counts_from": "own training text",
"smoke_run_loss": 5.8735,
"tokens": 2502989,
"unigram_cross_entropy": 6.8493
},
"Github": {
"counts_from": "own training text",
"smoke_run_loss": 5.7371,
"tokens": 2502662,
"unigram_cross_entropy": 6.3888
},
"Pile-CC": {
"counts_from": "own training text",
"smoke_run_loss": 6.7886,
"tokens": 2501855,
"unigram_cross_entropy": 7.559
},
"PubMed Central": {
"counts_from": "own training text",
"smoke_run_loss": 6.5792,
"tokens": 2502199,
"unigram_cross_entropy": 7.4554
},
"USPTO Backgrounds": {
"counts_from": "pooled 5 subjects",
"smoke_run_loss": 7.0021,
"tokens": 2500283,
"unigram_cross_entropy": 7.9605
},
"Wikipedia (en)": {
"counts_from": "own training text",
"smoke_run_loss": 7.0163,
"tokens": 2500584,
"unigram_cross_entropy": 7.72
}
},
"seconds": 7.0,
"status": "PASS",
"title": "Model beats the per-subject unigram baseline"
},
{
"detail": "5,830 steps / 96M tokens in 2.03 h of training. test top-1: frequent 40.5%, normal 3.1%, rare 0.8%, control 0.6% (frequent/control = 64.003x, needs 2x; monotone yes). paraphrase top-1: frequent 2.2%, control 0.4%",
"id": "A18",
"numbers": {
"steps": 5830,
"tokens_seen": 95518720,
"verdict": {
"checks": {
"frequent_at_least_twice_control": {
"control": 0.006333,
"frequent": 0.405333,
"pass": true,
"ratio": 64.003
},
"monotone_frequent_normal_rare": {
"frequent": 0.405333,
"normal": 0.031,
"pass": true,
"rare": 0.008
}
},
"paraphrase_top1": {
"control": 0.004,
"frequent": 0.021667,
"normal": 0.006833,
"rare": 0.006333
},
"pass": true,
"test_top1": {
"control": 0.006333,
"frequent": 0.405333,
"normal": 0.031,
"rare": 0.008
}
}
},
"seconds": 0.0,
"status": "PASS",
"title": "Biography facts are learnable"
}
],
"fingerprints": {
"bios": "7a8c48a96792d725205138dc31b1291033e6e18521603c6b6e987da2f2bb5ec4",
"pack": "15643c90477c31a7cbc5e394b1f353d795bb417ad6a5ef656b772ebbc1e91e81",
"pile": "3f4303d9b926b730d8d9bd9269084c6052df66b63aece15a07af100866aa2984"
},
"revision": "3be90335b66f24456a5d6659d9c8d208c0357119",
"version": "v1"
}