AvrovaDonz commited on
Commit
14f63a3
·
verified ·
1 Parent(s): 7944460

Link the independent publish gate and state the record mix

Browse files
README.md CHANGED
@@ -39,6 +39,14 @@ configs:
39
  path: components/openai-math/data/dev.parquet
40
  - split: test
41
  path: components/openai-math/data/test.parquet
 
 
 
 
 
 
 
 
42
  ---
43
 
44
  # Cayley-Math
@@ -49,6 +57,7 @@ Cayley-Math is a collaboratively maintained collection of mathematical text for
49
  | --- | --- | --- |
50
  | Anholon Phase 1 mathematics mirror | [Phase 1 bundle](phase1/README.md) | Anholon data maintainers |
51
  | OpenAI mathematics manuscripts | [OpenAI Math component](components/openai-math/README.md) | Source extraction and OMP DeepSeek metadata |
 
52
  | Contributed mathematics collections | [Component guide](components/README.md) | Each component's contributors |
53
 
54
  ## Phase 1 Files
 
39
  path: components/openai-math/data/dev.parquet
40
  - split: test
41
  path: components/openai-math/data/test.parquet
42
+ - config_name: google-math
43
+ data_files:
44
+ - split: train
45
+ path: components/google-math/data/train.parquet
46
+ - split: validation
47
+ path: components/google-math/data/dev.parquet
48
+ - split: test
49
+ path: components/google-math/data/test.parquet
50
  ---
51
 
52
  # Cayley-Math
 
57
  | --- | --- | --- |
58
  | Anholon Phase 1 mathematics mirror | [Phase 1 bundle](phase1/README.md) | Anholon data maintainers |
59
  | OpenAI mathematics manuscripts | [OpenAI Math component](components/openai-math/README.md) | Source extraction and OMP DeepSeek metadata |
60
+ | Google math papers | [Google Math component](components/google-math/README.md) | Verbatim Lean, geometry, notebook and PDF sources plus a seeded generator sample |
61
  | Contributed mathematics collections | [Component guide](components/README.md) | Each component's contributors |
62
 
63
  ## Phase 1 Files
components/google-math/README.md ADDED
@@ -0,0 +1,421 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Google math papers contribution to Cayley-Math
2
+
3
+ An independently versioned component holding source-derived text from eight Google and Google DeepMind
4
+ publications about mathematics, formal mathematics and mathematical reasoning. It follows Cayley-Math's
5
+ `components/` convention and reuses the record schema, Parquet encoding, split policy and packed-array
6
+ format of `components/openai-math/`. The frozen Phase 1 mirror is unchanged.
7
+
8
+ ## What this is, and what it is not
9
+
10
+ These are **artifacts of specific Google math papers** (Lean libraries, geometry problem files, generator
11
+ output, paper notebooks and proof PDFs), not a survey of Google's mathematics output and not a
12
+ paper-complete reproduction of any one of them. The corpus is a curated projection: a bounded, seeded
13
+ sample for the generator source, a verbatim text projection for the file sources.
14
+
15
+ Nothing here establishes mathematical truth. **No Lean project was compiled, no proof was checked, no
16
+ `lake build` or Comparator run was performed, and no mathematical verification of any kind took place.**
17
+ `record_type` labels the artifact kind, not the validity of its content. Records that contain `sorry`
18
+ placeholders remain placeholders.
19
+
20
+ The `mathematics-dataset` subset is output of the Saxton/Grefenstette/Hill/Kohli ICLR 2018 generator at
21
+ school-level difficulty. Those pairs are **not research mathematics**, and the subset is a sampled
22
+ projection of the paper's curriculum rather than the paper's dataset. Every record carries
23
+ `content_origin` so consumers can separate verbatim source text from `generator_output`.
24
+
25
+ **This corpus is mostly seeded generator output, not research mathematics.** `generated_qa_pair` records
26
+ dominate the record count, the formal subset is a small minority, and **no Lean compilation and no proof
27
+ verification was performed on any subset**, so no record here has a checked proof or a checked statement.
28
+ `statistics.json` carries the measured counts, and the component README states them in full.
29
+
30
+ ## Included sources
31
+
32
+ Each source is pinned to a commit or dataset revision, and each ships math text under Apache-2.0
33
+ software terms with CC-BY 4.0 for non-software materials where the upstream states so.
34
+
35
+ | Source | Upstream at pinned revision | Records | Groups | Text characters | Record types |
36
+ | --- | --- | ---: | ---: | ---: | --- |
37
+ | `alphaproof-nexus` | [alphaproof-nexus-results](https://github.com/google-deepmind/alphaproof-nexus-results) `0647711a7118` | 83 | 51 | 2,229,771 | formal_lean_proof, natural_language_proof |
38
+ | `formal-conjectures` | [formal-conjectures](https://github.com/google-deepmind/formal-conjectures) `6751c0f04398` | 2,135 | 2,135 | 8,133,507 | formal_lean_statement |
39
+ | `formal-imo` | [formal-imo](https://github.com/google-deepmind/formal-imo) `a85d84b97c8e` | 261 | 261 | 420,948 | formal_lean_statement |
40
+ | `formal-putnam-like` | [formal-putnam-like](https://github.com/google-deepmind/formal-putnam-like) `636c2d46b3a7` | 96 | 96 | 117,201 | formal_lean_statement |
41
+ | `cryptids` | [cryptids](https://github.com/google-deepmind/cryptids) `3647718b5df2` | 464 | 464 | 31,638,592 | formal_lean_proof |
42
+ | `mathematics-dataset` | [mathematics_dataset](https://github.com/google-deepmind/mathematics_dataset) `427f45075f84` | 477,276 | 127 | 35,710,731 | generated_qa_pair |
43
+ | `alphageometry` | [alphageometry](https://github.com/google-deepmind/alphageometry) `6777cb586cbb` | 267 | 267 | 70,907 | formal_geometry_problem, geometry_rule_table |
44
+ | `mathematics-conjectures` | [mathematics_conjectures](https://github.com/google-deepmind/mathematics_conjectures) `8302f2910101` | 2 | 2 | 43,231 | research_notebook |
45
+
46
+ `alphaproof-nexus` keeps all 83 records of the earlier standalone build unchanged in content: 71 Lean
47
+ proof files plus 12 pypdf-extracted prose proof PDFs over 51 problem groups. Only its split assignment
48
+ changed, because splits are recomputed over the merged group set of all eight sources rather than over
49
+ its own 51 groups. `formal-conjectures` is the 1,861-file `FormalConjectures/` statement library plus
50
+ its 274 helper-library files. `formal-imo` and `formal-putnam-like` are Lean 4 problem statements.
51
+ `cryptids` is the whole Lean library, including the Busy Beaver holdout proofs, whose individual files
52
+ reach 1.4 MB of mechanically generated state traces. `alphageometry` contributes 265 geometry problems
53
+ in the AlphaGeometry language: 231 JGEX problems, 30 olympiad problems and 4 library examples, plus the
54
+ `defs.txt` and `rules.txt` rule tables. The two `mathematics-conjectures` notebooks accompany the Nature
55
+ 2021 knot-theory and representation-theory paper.
56
+
57
+ ## Excluded candidates and why
58
+
59
+ The widened scan is recorded in full in `sources.json` (`excluded_candidates`) and `audit/source-scan.json`.
60
+ The short version:
61
+
62
+ - `google-deepmind/eval_hub` (putnam_like and live_math) is excluded because every problem carries a
63
+ `CANARY_ID` benchmark-contamination canary and live_math is an actively maintained benchmark whose
64
+ multimodal problems depend on images absent from this text corpus.
65
+ - `deepmind/aqua_rat` accompanies a program-synthesis paper, not a mathematics paper.
66
+ - `deepmind/math_dataset` is a 40 GB pre-generated mirror of the generator already used here.
67
+ - `google-research/math-opt-benchmark`, `google/mathsteps`, `google-deepmind/alphatensor` and every
68
+ math-named software library (`tree-math`, `mathfu`, `vector_math.dart`, `opengrm`) ship code and no
69
+ math text.
70
+ - `google-deepmind/cartesian-frames` formalises an alignment blog sequence in HOL4 SML, not a
71
+ mathematics paper.
72
+ - `google-deepmind/miniF2F` is a fork of a non-Google corpus.
73
+ - `google-deepmind/alphageometry2` embeds 26 problems in Python dict literals, and AG1 already ships
74
+ the same olympiad problems verbatim in the AlphaGeometry language.
75
+ - `google/feabench`, `google/phasecoder` and `google-research/pointdit` are physics, audio and vision
76
+ papers whose titles happen to contain mathematical words.
77
+
78
+ ## Files and schema
79
+
80
+ - `data/{train,dev,test}.jsonl.gz`: source records with required fields text, group, license; unique id,
81
+ paper_id, split, record_type, source_url, and `source_id` naming the upstream.
82
+ - `data/{train,dev,test}.parquet`: identical records. Core fields
83
+ (id,text,group,license,paper_id,split,record_type,source_url) are string columns; all remaining JSON
84
+ fields are serialised in `metadata_json`. Merge `json.loads(metadata_json)` with the core columns to
85
+ reconstruct each JSONL record.
86
+ - `sources.json`: every included source with its pinned revision, license and measured counts, plus every
87
+ excluded candidate with its reason.
88
+ - `licenses/<source_id>/LICENSE`: one upstream license per source; `LICENSE` at the component root is the
89
+ Apache-2.0 text shared by all eight.
90
+ - `audit/source-files.jsonl`: path, sha256, bytes, source_id, kind and pinned source_url for every blob and
91
+ for the generated corpus file.
92
+ - `audit/group-splits.json`: the group assignment, the policy, and per-source group counts.
93
+ - `audit/extraction.jsonl`: per-PDF and per-notebook extraction metrics and quality flags.
94
+ - `audit/mathematics-dataset-generation.json`: generator commit, seed, per-regime quotas, module list, and
95
+ measured yield.
96
+ - `audit/excluded-duplicates.jsonl`, `audit/packing.json`, `audit/source-scan.json`.
97
+ - `audit/validation-evidence.json`: the independent decode and round-trip evidence described below.
98
+ - `audit/publish-check.json`: the second publish gate, measured from the delivered files after the build.
99
+ - `statistics.json`: measured records, groups, splits, characters and packed counts, per source.
100
+ - `manifest.json`: SHA256 and byte size for every delivered file except itself and validation.json.
101
+ - `packed/`: mmla-type2-packed-v1 arrays, manifest, row attribution and per-record token counts.
102
+ - `tools/`: the generator driver, the build script and the validation scripts actually run.
103
+
104
+ ## Loading
105
+
106
+ ~~~python
107
+ import gzip, json
108
+ from pathlib import Path
109
+ component = Path('components/google-math')
110
+ with gzip.open(component / 'data/train.jsonl.gz', 'rt', encoding='utf-8') as f:
111
+ first_record = json.loads(next(f))
112
+
113
+ import pyarrow.parquet as pq
114
+ row = pq.read_table(component / 'data/train.parquet').slice(0, 1).to_pylist()[0]
115
+ metadata = json.loads(row.pop('metadata_json'))
116
+ first_parquet_record = {**row, **metadata}
117
+ ~~~
118
+
119
+ ## Packed training arrays
120
+
121
+ `packed/manifest.json` uses `mmla-type2-packed-v1`: input_ids, labels and role_ids are int32 arrays of
122
+ shape [rows, 2048], pad=0, role=0, ignored labels=-100. Each record is encoded as
123
+ BOS + SentencePiece(text) + EOS with the included Cayley-Math tokenizer (32768 pieces); window stride is
124
+ 2048 with one-token lookahead, labels predict the following token, and targets that cross into a new
125
+ record's BOS or into padding are masked. Split-specific `rows.jsonl` maps every non-padding token back to
126
+ its record; `records.jsonl` stores token counts and hashes. This component does not reproduce Phase 1
127
+ quotas, packing or splits.
128
+
129
+ Measured packed counts: train 17691 rows /
130
+ 36230380 tokens /
131
+ 35751064 loss-bearing labels; dev
132
+ 51 / 102872 /
133
+ 101639; test
134
+ 91 / 185977 /
135
+ 185942.
136
+
137
+ ## Splits
138
+
139
+ Splits are assigned per group, never per record, so companions and families stay together: a Lean proof
140
+ and its prose proof, a conjecture file and its helper library, or one generated question family. Every
141
+ group key is namespaced by its `source_id`. Assignment sorts SHA256(group) as hex, reserves the first
142
+ ceil(1%) of groups for dev, the next ceil(1%) for test, and the rest for train.
143
+ `audit/group-splits.json` freezes every assignment: train
144
+ 479316 records over 3333 groups,
145
+ dev 1233 over 35,
146
+ test 35 over 35.
147
+
148
+ ## Text projection
149
+
150
+ Lean, geometry-problem and rule files are byte-identical to the pinned blob, license headers included.
151
+ Notebook records join the cell sources of each notebook with a newline; no cell text is altered.
152
+ `natural_language_proof` records come from pypdf page-text extraction and are **lossy**: math symbols
153
+ arrive as Unicode substitutions, ligature artifacts appear, and inline formulas lose their two-dimensional
154
+ layout. C0/C1 control characters were replaced and blank-line runs collapsed. Each PDF record carries
155
+ `extraction_quality = lossy_pdf_text_layer` with page and character counts, and `audit/extraction.jsonl`
156
+ holds the same per-file audit.
157
+
158
+ ## Verification evidence
159
+
160
+ `tools/validate_google_math.py` checks file hashes, source coverage, unique record IDs and unique text,
161
+ group split separation, lossless Parquet reconstruction, packed array shapes, causal shift, padding and
162
+ boundary masks, and re-encodes every record's token stream against `data/*.jsonl.gz`. It writes
163
+ `validation.json`; the last run passed all 13 checks
164
+ (`all_delivered_file_hashes_and_sizes`, `pinned_sources_metadata_and_excluded_candidates`,
165
+ `source_file_inventory_local_bytes`, `group_split_policy_recomputed_from_scratch`,
166
+ `lossless_jsonl_parquet_exact_text_uniqueness_and_record_provenance`, `group_split_separation`,
167
+ `verbatim_lean_source_equality_local_bytes`, `lean_and_generated_records_survive`,
168
+ `per_file_extraction_audit`, `dedup_audit_present`, `generator_audit_seed_and_quota`,
169
+ `all_packed_tokens_reencoded_source_shift_padding_masks_and_attribution`, `statistics_match_artifacts`).
170
+
171
+ `tools/verify_google_math.py` is the independent check against the delivered files, and its output is
172
+ [`audit/validation-evidence.json`](audit/validation-evidence.json). It decoded 96 packed rows, covering
173
+ 68 of the 70 train shards plus the single dev and test shards, and for each row (a) confirmed every row
174
+ slice equals the re-encoding of the corresponding span of `[BOS] + SentencePiece(text) + [EOS]` for its
175
+ attributed source record and (b) confirmed the decoded text equals the decoded source text modulo
176
+ whitespace. It recorded 8 literal
177
+ unified diffs of a decoded row against its source record; every diff had 0 lines, whitespace identical.
178
+ It round-tripped 3 Parquet rows (indices 0, 239,658 and 479,315) back to their JSONL records with exact
179
+ match, and confirmed a `formal_lean_proof`, a `formal_lean_statement`, a `generated_qa_pair`, a
180
+ `formal_geometry_problem`, a `natural_language_proof` and a `research_notebook` record all survive with
181
+ matching `text_sha256`. Split separation was recomputed from the data rather than trusted: 3,403 groups
182
+ in the audit and 3,403 in the data, no group spanning two splits, train/dev/test at 3,333/35/35 groups.
183
+
184
+ [`audit/publish-check.json`](audit/publish-check.json) is a second publish gate with its own code rather
185
+ than this component's tools. It agrees with every total above (480,584 records, 3,403 groups, 78,364,888
186
+ characters, and the packed counts for all three splits), decodes 12 packed rows across all three splits
187
+ and compares 263 attributed segments. All 263 row slices matched their canonical token slices by exact
188
+ identity, and all 248 segments that cover a whole record decoded to text identical to the source record
189
+ under whitespace normalisation and re-encoded to the identical token sequence. The other 15 segments are
190
+ row-boundary fragments that begin or end inside a SentencePiece piece, so they cannot re-encode
191
+ fragment-by-fragment; each one's decoded text was still located verbatim inside its attributed record.
192
+ It matched 5 of 5 Parquet round-trips exactly and recomputed group separation from `data/*.jsonl.gz`
193
+ directly, finding 0 groups spanning two splits.
194
+
195
+ The record mix is deliberately stated because it is skewed: `generated_qa_pair` records are 99.31% of
196
+ records but 45.57% of characters, while formal Lean records are 0.63% of records and 54.22% of characters,
197
+ and all non-generator records together are 0.69% of records. A consumer wanting formal or prose
198
+ mathematics should filter on `record_type` rather than read the corpus whole.
199
+
200
+ ## Publication
201
+
202
+ The component is delivered as Hugging Face dataset pull request
203
+ [MMLA-ORG/Cayley-Math discussion #3](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/discussions/3)
204
+ on ref `refs/pr/3`. It carries the whole component plus the top-level `README.md` config and Collections
205
+ update, and it has not been merged into `main`. Consult that discussion for the current merge state.
206
+
207
+ A token with `write` scope or org-admin rights is required to merge it. The token used for this build can
208
+ read the repository and open pull requests, but a direct commit to `main` is rejected with
209
+ `403 Forbidden: You have read access but not the required permissions for this operation`. Merging the
210
+ pull request, or re-running `hf upload` without `--create-pr` once such a token is available, completes
211
+ the submission.
212
+
213
+ ## Limitations
214
+
215
+ - No Lean compilation, proof checking, Comparator run or mathematical verification of any kind.
216
+ - The build ran in a sandbox-free subprocess only for the generator; `cryptids`, `formal-conjectures`,
217
+ `formal-imo`, `formal-putnam-like`, `alphageometry`, `alphaproof-nexus` and the notebooks were never
218
+ built or executed, only read as text.
219
+ - PDF prose proofs are noisy pypdf text; symbol and ligature fidelity is not guaranteed.
220
+ - The generator subset is a seeded sample of one curriculum, not the paper's full dataset, and its
221
+ questions are school-level rather than research mathematics. One module, `algebra__linear_1d`, stopped
222
+ at 3,055 of its 8,000 quota when upstream's own `assert entropy == 0` fired; nothing was repaired, and
223
+ the shortfall is recorded in `audit/mathematics-dataset-generation.json`.
224
+ - `formal-conjectures` upstream states that some content derives from third-party sources (Wikipedia,
225
+ MathOverflow and OEIS under CC-BY-SA 4.0). Per-file reference URLs are retained inside the Lean text; no
226
+ separate per-file license audit was performed.
227
+ - Near-duplicate detection, Phase 1 overlap analysis and editorial review were not performed. 1,629
228
+ exact duplicate texts were dropped and are listed in `audit/excluded-duplicates.jsonl`.
229
+ - The SentencePiece model normalises runs of spaces, so packed text decodes whitespace-normalised rather
230
+ than byte-identically to the sources.
231
+ - Only problems that upstream published are present, so several subsets are selection-biased. `cryptids`
232
+ contributes 31.6M of the 78.4M characters because a handful of Busy Beaver traces are enormous, so the
233
+ corpus is dominated by machine-generated proof script rather than mathematics prose.
234
+
235
+ ## Statistics
236
+
237
+ ~~~json
238
+ {
239
+ "component": "google-math",
240
+ "sources": 8,
241
+ "source_commits": {
242
+ "alphaproof-nexus": "0647711a71183c1ea492ad60860776617ce1ea88",
243
+ "formal-conjectures": "6751c0f043985ac6647db28a45a63afdf21acffe",
244
+ "formal-imo": "a85d84b97c8e88a352c136b1e29c355139609bea",
245
+ "formal-putnam-like": "636c2d46b3a7d7e77a9f58209b0c17816f8d172b",
246
+ "cryptids": "3647718b5df22e1a61462dec0376bd204526285a",
247
+ "mathematics-dataset": "427f45075f84b8b9774950196ad63867ca20ffb3",
248
+ "alphageometry": "6777cb586cbb46beed28db12dc72c69770b68337",
249
+ "mathematics-conjectures": "8302f29101019434e5a0a25807a4e42cd9d96414"
250
+ },
251
+ "records_before_deduplication": 482213,
252
+ "records": 480584,
253
+ "excluded_exact_duplicates": 1629,
254
+ "groups": 3403,
255
+ "record_types": {
256
+ "geometry_rule_table": 2,
257
+ "formal_geometry_problem": 265,
258
+ "formal_lean_proof": 535,
259
+ "natural_language_proof": 12,
260
+ "formal_lean_statement": 2492,
261
+ "research_notebook": 2,
262
+ "generated_qa_pair": 477276
263
+ },
264
+ "text_characters": 78364888,
265
+ "per_source": {
266
+ "alphaproof-nexus": {
267
+ "records": 83,
268
+ "groups": 51,
269
+ "record_types": {
270
+ "formal_lean_proof": 71,
271
+ "natural_language_proof": 12
272
+ },
273
+ "text_characters": 2229771,
274
+ "upstream_files": 83,
275
+ "split_records": {
276
+ "train": 83,
277
+ "dev": 0,
278
+ "test": 0
279
+ }
280
+ },
281
+ "formal-conjectures": {
282
+ "records": 2135,
283
+ "groups": 2135,
284
+ "record_types": {
285
+ "formal_lean_statement": 2135
286
+ },
287
+ "text_characters": 8133507,
288
+ "upstream_files": 2135,
289
+ "split_records": {
290
+ "train": 2088,
291
+ "dev": 23,
292
+ "test": 24
293
+ }
294
+ },
295
+ "formal-imo": {
296
+ "records": 261,
297
+ "groups": 261,
298
+ "record_types": {
299
+ "formal_lean_statement": 261
300
+ },
301
+ "text_characters": 420948,
302
+ "upstream_files": 261,
303
+ "split_records": {
304
+ "train": 252,
305
+ "dev": 6,
306
+ "test": 3
307
+ }
308
+ },
309
+ "formal-putnam-like": {
310
+ "records": 96,
311
+ "groups": 96,
312
+ "record_types": {
313
+ "formal_lean_statement": 96
314
+ },
315
+ "text_characters": 117201,
316
+ "upstream_files": 96,
317
+ "split_records": {
318
+ "train": 93,
319
+ "dev": 1,
320
+ "test": 2
321
+ }
322
+ },
323
+ "cryptids": {
324
+ "records": 464,
325
+ "groups": 464,
326
+ "record_types": {
327
+ "formal_lean_proof": 464
328
+ },
329
+ "text_characters": 31638592,
330
+ "upstream_files": 464,
331
+ "split_records": {
332
+ "train": 457,
333
+ "dev": 2,
334
+ "test": 5
335
+ }
336
+ },
337
+ "mathematics-dataset": {
338
+ "records": 477276,
339
+ "groups": 127,
340
+ "record_types": {
341
+ "generated_qa_pair": 477276
342
+ },
343
+ "text_characters": 35710731,
344
+ "upstream_files": 1,
345
+ "split_records": {
346
+ "train": 476076,
347
+ "dev": 1200,
348
+ "test": 0
349
+ }
350
+ },
351
+ "alphageometry": {
352
+ "records": 267,
353
+ "groups": 267,
354
+ "record_types": {
355
+ "geometry_rule_table": 2,
356
+ "formal_geometry_problem": 265
357
+ },
358
+ "text_characters": 70907,
359
+ "upstream_files": 5,
360
+ "split_records": {
361
+ "train": 265,
362
+ "dev": 1,
363
+ "test": 1
364
+ }
365
+ },
366
+ "mathematics-conjectures": {
367
+ "records": 2,
368
+ "groups": 2,
369
+ "record_types": {
370
+ "research_notebook": 2
371
+ },
372
+ "text_characters": 43231,
373
+ "upstream_files": 2,
374
+ "split_records": {
375
+ "train": 2,
376
+ "dev": 0,
377
+ "test": 0
378
+ }
379
+ }
380
+ },
381
+ "splits": {
382
+ "train": {
383
+ "records": 479316,
384
+ "groups": 3333,
385
+ "sources": 8,
386
+ "text_characters": 77705566
387
+ },
388
+ "dev": {
389
+ "records": 1233,
390
+ "groups": 35,
391
+ "sources": 6,
392
+ "text_characters": 233131
393
+ },
394
+ "test": {
395
+ "records": 35,
396
+ "groups": 35,
397
+ "sources": 5,
398
+ "text_characters": 426191
399
+ }
400
+ },
401
+ "packed": {
402
+ "train": {
403
+ "rows": 17691,
404
+ "tokens": 36230380,
405
+ "loss_bearing_labels": 35751064
406
+ },
407
+ "dev": {
408
+ "rows": 51,
409
+ "tokens": 102872,
410
+ "loss_bearing_labels": 101639
411
+ },
412
+ "test": {
413
+ "rows": 91,
414
+ "tokens": 185977,
415
+ "loss_bearing_labels": 185942
416
+ }
417
+ },
418
+ "mathematical_verification": "not_performed",
419
+ "lean_compilation": "not_performed"
420
+ }
421
+ ~~~
components/google-math/manifest.json ADDED
@@ -0,0 +1,1285 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "cayley-contribution-file-manifest-v1",
3
+ "excludes": [
4
+ ".DS_Store",
5
+ "manifest.json",
6
+ "validation.json"
7
+ ],
8
+ "files": [
9
+ {
10
+ "path": "LICENSE",
11
+ "bytes": 11358,
12
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
13
+ },
14
+ {
15
+ "path": "README.md",
16
+ "bytes": 20593,
17
+ "sha256": "b3c98cc1005b5263dbe2bad37aafba811bf69ac6f268a918bcc851fe6f83af6d"
18
+ },
19
+ {
20
+ "path": "audit/excluded-duplicates.jsonl",
21
+ "bytes": 558417,
22
+ "sha256": "14f34b30f41b238e3998a6d248905e19b0617a87423551ce776b8f1e05211781"
23
+ },
24
+ {
25
+ "path": "audit/extraction.jsonl",
26
+ "bytes": 6986,
27
+ "sha256": "c33145cc7ec9286da3c4225961eca177f9ffb0260bec543773e61c31a05cdb5f"
28
+ },
29
+ {
30
+ "path": "audit/group-splits.json",
31
+ "bytes": 256196,
32
+ "sha256": "ca0daf0f06a8e1825393734adeab265122a3179e2eb95d44c65112162ce8d7f1"
33
+ },
34
+ {
35
+ "path": "audit/mathematics-dataset-generation.json",
36
+ "bytes": 23626,
37
+ "sha256": "8c595f7d49017869116e9e5690ec648e4651b5c3738402fe80ef31234d465646"
38
+ },
39
+ {
40
+ "path": "audit/packing.json",
41
+ "bytes": 366,
42
+ "sha256": "37328e5f0c61c0ca8dc51ccc147ad5cf9c7a33e638c2d36e35e502205a803712"
43
+ },
44
+ {
45
+ "path": "audit/publish-check.json",
46
+ "bytes": 214178,
47
+ "sha256": "276b46afdb38becead1a5cbbb5f08a6cd78e56659692b29a45e46cf58d92b9eb"
48
+ },
49
+ {
50
+ "path": "audit/source-files.jsonl",
51
+ "bytes": 1085021,
52
+ "sha256": "94e48be18e67d57ba3ea25a63fd43b93c51d0c2f33cad43e6539bc09c6b2f79b"
53
+ },
54
+ {
55
+ "path": "audit/source-scan.json",
56
+ "bytes": 8517,
57
+ "sha256": "446e06f0880bc28a6c4d40e63bd70eadb4de60c48d2bd06cde98bae39d79f25d"
58
+ },
59
+ {
60
+ "path": "audit/validation-evidence.json",
61
+ "bytes": 62693,
62
+ "sha256": "4e74750160791bf43eee9c895e4d07999f6458317e459fee52ebfbb500d3c123"
63
+ },
64
+ {
65
+ "path": "data/dev.jsonl.gz",
66
+ "bytes": 138302,
67
+ "sha256": "4f7d79e58e1b20893f3f04a5a22273e38cea4e3425a63f15b611c19fda5a8d81"
68
+ },
69
+ {
70
+ "path": "data/dev.parquet",
71
+ "bytes": 173120,
72
+ "sha256": "ea45347fcab0d1c38ccdd73f0d8b7c3540efd8dccdac3d699653f9118eb7f846"
73
+ },
74
+ {
75
+ "path": "data/test.jsonl.gz",
76
+ "bytes": 85758,
77
+ "sha256": "64428a08480b28dd8c60357847dd0e23f7d14f210d041a471021a14b4f40de08"
78
+ },
79
+ {
80
+ "path": "data/test.parquet",
81
+ "bytes": 107242,
82
+ "sha256": "32826a8a6771cf14b009d631a7d2dcf9194d04afe16ba7a299647590bcd46649"
83
+ },
84
+ {
85
+ "path": "data/train.jsonl.gz",
86
+ "bytes": 50794406,
87
+ "sha256": "2b08f67728228b7262b4b51f6c675ca6408802f0a8ed2ebc72531c6e4fe6af67"
88
+ },
89
+ {
90
+ "path": "data/train.parquet",
91
+ "bytes": 62017527,
92
+ "sha256": "fa810ac44fca0b912e3dada3db9db5f2d25480f0a71c041a219007ad8556b505"
93
+ },
94
+ {
95
+ "path": "licenses/alphageometry/LICENSE",
96
+ "bytes": 11358,
97
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
98
+ },
99
+ {
100
+ "path": "licenses/alphaproof-nexus/LICENSE",
101
+ "bytes": 11358,
102
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
103
+ },
104
+ {
105
+ "path": "licenses/cryptids/LICENSE",
106
+ "bytes": 11358,
107
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
108
+ },
109
+ {
110
+ "path": "licenses/formal-conjectures/LICENSE",
111
+ "bytes": 11358,
112
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
113
+ },
114
+ {
115
+ "path": "licenses/formal-imo/LICENSE",
116
+ "bytes": 11358,
117
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
118
+ },
119
+ {
120
+ "path": "licenses/formal-putnam-like/LICENSE",
121
+ "bytes": 11357,
122
+ "sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4"
123
+ },
124
+ {
125
+ "path": "licenses/mathematics-conjectures/LICENSE",
126
+ "bytes": 11358,
127
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
128
+ },
129
+ {
130
+ "path": "licenses/mathematics-dataset/LICENSE",
131
+ "bytes": 11358,
132
+ "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
133
+ },
134
+ {
135
+ "path": "packed/dev-000000-input_ids.npy",
136
+ "bytes": 417920,
137
+ "sha256": "048ed57ac967dfef03c40c7de3d3cec24e9119a2b179d84a8c66b9ce89de0eaa"
138
+ },
139
+ {
140
+ "path": "packed/dev-000000-labels.npy",
141
+ "bytes": 417920,
142
+ "sha256": "e16d2b07ec13ffaa24052d169cf3420488e949c2e2dba29126874897addb3e91"
143
+ },
144
+ {
145
+ "path": "packed/dev-000000-role_ids.npy",
146
+ "bytes": 417920,
147
+ "sha256": "8de6e2693cad31951340cfa4bc3966dd33870c1149703bfdbba82736d09674e5"
148
+ },
149
+ {
150
+ "path": "packed/dev-records.jsonl",
151
+ "bytes": 264613,
152
+ "sha256": "d40b8751ecfdf2aec25756c31f08c0bbff15202676c3d03dc76f2d9918873f3a"
153
+ },
154
+ {
155
+ "path": "packed/dev-rows.jsonl",
156
+ "bytes": 251135,
157
+ "sha256": "d713121de44a92b3cbb13927b46f4d60a4b87190ac14aff5f93c62c4495cb40b"
158
+ },
159
+ {
160
+ "path": "packed/manifest.json",
161
+ "bytes": 73029,
162
+ "sha256": "3cc73dc38aa1281e1f6805c1d6b11b3967d62fbc07eee0ecd721f61a1ac3fb43"
163
+ },
164
+ {
165
+ "path": "packed/test-000000-input_ids.npy",
166
+ "bytes": 745600,
167
+ "sha256": "3113618e0c2fd5b6f4ec7d2378ee88b69a0bfe4f3b1f396f192e58da70d86a41"
168
+ },
169
+ {
170
+ "path": "packed/test-000000-labels.npy",
171
+ "bytes": 745600,
172
+ "sha256": "3addd041129a9b90eb324b9f052c4d95e89993a6b6fecce747c22ebfbc2f0a12"
173
+ },
174
+ {
175
+ "path": "packed/test-000000-role_ids.npy",
176
+ "bytes": 745600,
177
+ "sha256": "4c1b072b717384b67e2a97fb33129074169edbe0dedce19324520cdd7e1a8e6c"
178
+ },
179
+ {
180
+ "path": "packed/test-records.jsonl",
181
+ "bytes": 8442,
182
+ "sha256": "12905cc97a4b3dd00deb77d4434512e164ca03b07cc2860fba370209afe07257"
183
+ },
184
+ {
185
+ "path": "packed/test-rows.jsonl",
186
+ "bytes": 34909,
187
+ "sha256": "cc9950d781db6a5b2ae3b7c92ee2666a362fe9deea5c9646c18e13c52b079fae"
188
+ },
189
+ {
190
+ "path": "packed/train-000000-input_ids.npy",
191
+ "bytes": 2097280,
192
+ "sha256": "67a3fc74e41197b40ebbd57ba77f132d1a44f4384b7726c3634f0a69d07e5f89"
193
+ },
194
+ {
195
+ "path": "packed/train-000000-labels.npy",
196
+ "bytes": 2097280,
197
+ "sha256": "b87bfb10a4c9147b64e4f24ace6cf7becfb09007a9c73ca4ee8746ef7a61b883"
198
+ },
199
+ {
200
+ "path": "packed/train-000000-role_ids.npy",
201
+ "bytes": 2097280,
202
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
203
+ },
204
+ {
205
+ "path": "packed/train-000001-input_ids.npy",
206
+ "bytes": 2097280,
207
+ "sha256": "92ca6973edb703072e2c07abe2184dcdb2d41bba229393c16fe6a6a10c6320f8"
208
+ },
209
+ {
210
+ "path": "packed/train-000001-labels.npy",
211
+ "bytes": 2097280,
212
+ "sha256": "3ed7b86f967a8243a8b96376d76418eab0c879e58a9d382d023c165d6f6db267"
213
+ },
214
+ {
215
+ "path": "packed/train-000001-role_ids.npy",
216
+ "bytes": 2097280,
217
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
218
+ },
219
+ {
220
+ "path": "packed/train-000002-input_ids.npy",
221
+ "bytes": 2097280,
222
+ "sha256": "5ec248bfc437db9ea4ee641d207e4b7aa99be6f2686a773d44eaa2afad536239"
223
+ },
224
+ {
225
+ "path": "packed/train-000002-labels.npy",
226
+ "bytes": 2097280,
227
+ "sha256": "b459bfb5de67a6967f1f2a8c6d96c36080e4b4d0f455cd164b5b180c3eb2df32"
228
+ },
229
+ {
230
+ "path": "packed/train-000002-role_ids.npy",
231
+ "bytes": 2097280,
232
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
233
+ },
234
+ {
235
+ "path": "packed/train-000003-input_ids.npy",
236
+ "bytes": 2097280,
237
+ "sha256": "f3603091efa22deccd00a879ddc6d1982ad98408144e3d4822eb2dea2d014dbd"
238
+ },
239
+ {
240
+ "path": "packed/train-000003-labels.npy",
241
+ "bytes": 2097280,
242
+ "sha256": "5f16021aff3c2498301a0ae7ad06795ea260d5c2d336b07f89d358e0745dec26"
243
+ },
244
+ {
245
+ "path": "packed/train-000003-role_ids.npy",
246
+ "bytes": 2097280,
247
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
248
+ },
249
+ {
250
+ "path": "packed/train-000004-input_ids.npy",
251
+ "bytes": 2097280,
252
+ "sha256": "a92114cf7849bd5b9a62deb6f417d444e82adfe90954f17a743828aab597427d"
253
+ },
254
+ {
255
+ "path": "packed/train-000004-labels.npy",
256
+ "bytes": 2097280,
257
+ "sha256": "75735298faa6d26bc009c752948e727dac697b5d5c7ca79f05cefb69e6d6b1a6"
258
+ },
259
+ {
260
+ "path": "packed/train-000004-role_ids.npy",
261
+ "bytes": 2097280,
262
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
263
+ },
264
+ {
265
+ "path": "packed/train-000005-input_ids.npy",
266
+ "bytes": 2097280,
267
+ "sha256": "a8a087408502409492951a1450e06c69adc026ca0b28c1fa9fed5fd38bf030d4"
268
+ },
269
+ {
270
+ "path": "packed/train-000005-labels.npy",
271
+ "bytes": 2097280,
272
+ "sha256": "88ad80409e1f35ec6889ebe7df11c6ae765c14594a56988aae6b719725dab65b"
273
+ },
274
+ {
275
+ "path": "packed/train-000005-role_ids.npy",
276
+ "bytes": 2097280,
277
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
278
+ },
279
+ {
280
+ "path": "packed/train-000006-input_ids.npy",
281
+ "bytes": 2097280,
282
+ "sha256": "74ea225057daac82209038cce61dd4beef2d5332b6ded3c08f36089455f15886"
283
+ },
284
+ {
285
+ "path": "packed/train-000006-labels.npy",
286
+ "bytes": 2097280,
287
+ "sha256": "0bb6f1ffb34d3948aa626ac0bd74e9e994231429f6d43541c1b309b545a9fed0"
288
+ },
289
+ {
290
+ "path": "packed/train-000006-role_ids.npy",
291
+ "bytes": 2097280,
292
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
293
+ },
294
+ {
295
+ "path": "packed/train-000007-input_ids.npy",
296
+ "bytes": 2097280,
297
+ "sha256": "15847a61802e7165b3365c0a7ba02b372fdda72f7f0ff620f440ecbc437f402a"
298
+ },
299
+ {
300
+ "path": "packed/train-000007-labels.npy",
301
+ "bytes": 2097280,
302
+ "sha256": "cf75c5701a6d98329e5e70e39572ba135958b966c26e8432079b2ca4f0cbc7c2"
303
+ },
304
+ {
305
+ "path": "packed/train-000007-role_ids.npy",
306
+ "bytes": 2097280,
307
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
308
+ },
309
+ {
310
+ "path": "packed/train-000008-input_ids.npy",
311
+ "bytes": 2097280,
312
+ "sha256": "5794a37cda50832e3854d9171c4c6346e2feb71676f7e979031aaef871696fcd"
313
+ },
314
+ {
315
+ "path": "packed/train-000008-labels.npy",
316
+ "bytes": 2097280,
317
+ "sha256": "c7d08bc3550adbc15f8c7de620ae87dd657cddeb36020ccd6406723d87daa5e0"
318
+ },
319
+ {
320
+ "path": "packed/train-000008-role_ids.npy",
321
+ "bytes": 2097280,
322
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
323
+ },
324
+ {
325
+ "path": "packed/train-000009-input_ids.npy",
326
+ "bytes": 2097280,
327
+ "sha256": "beb7fb7df5c92e6088aa4bc2077d9ab423fb6afba3f1ecbacd64bad093d8642e"
328
+ },
329
+ {
330
+ "path": "packed/train-000009-labels.npy",
331
+ "bytes": 2097280,
332
+ "sha256": "70ed458217ca3a76a372ac210e6bf45784dbcdf20c8c774d5b47fadc654b4629"
333
+ },
334
+ {
335
+ "path": "packed/train-000009-role_ids.npy",
336
+ "bytes": 2097280,
337
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
338
+ },
339
+ {
340
+ "path": "packed/train-000010-input_ids.npy",
341
+ "bytes": 2097280,
342
+ "sha256": "d43bef90029f63e679e6b2a0c8c7c98d52c0033b52b13fde754637f38ed8a352"
343
+ },
344
+ {
345
+ "path": "packed/train-000010-labels.npy",
346
+ "bytes": 2097280,
347
+ "sha256": "8c8a3341efbc137c65fdbc99c81eb71a25fcbe97853520165703a7e6b69c5e81"
348
+ },
349
+ {
350
+ "path": "packed/train-000010-role_ids.npy",
351
+ "bytes": 2097280,
352
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
353
+ },
354
+ {
355
+ "path": "packed/train-000011-input_ids.npy",
356
+ "bytes": 2097280,
357
+ "sha256": "df6cc218091484514e3e2fb6270edb428d035ac6b41b25c86b6c91a4fb36f659"
358
+ },
359
+ {
360
+ "path": "packed/train-000011-labels.npy",
361
+ "bytes": 2097280,
362
+ "sha256": "899f2be686f934b5456dd863d5e0a15798df4219e93d1e957daa8e42a77d694f"
363
+ },
364
+ {
365
+ "path": "packed/train-000011-role_ids.npy",
366
+ "bytes": 2097280,
367
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
368
+ },
369
+ {
370
+ "path": "packed/train-000012-input_ids.npy",
371
+ "bytes": 2097280,
372
+ "sha256": "5ace43c4acddd64c00ca1edc532bb7cb8593ed81416a61adb30425ebeb4ca38e"
373
+ },
374
+ {
375
+ "path": "packed/train-000012-labels.npy",
376
+ "bytes": 2097280,
377
+ "sha256": "ce15863449f784a5329a41377e5e22521aadcbe7a9a70461f457b6e9c5cf3514"
378
+ },
379
+ {
380
+ "path": "packed/train-000012-role_ids.npy",
381
+ "bytes": 2097280,
382
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
383
+ },
384
+ {
385
+ "path": "packed/train-000013-input_ids.npy",
386
+ "bytes": 2097280,
387
+ "sha256": "0b38bc5ccd5564fabbab5627a500ba466fa701a97812b168c38fe97e2ecf7f0f"
388
+ },
389
+ {
390
+ "path": "packed/train-000013-labels.npy",
391
+ "bytes": 2097280,
392
+ "sha256": "1f972889934c1f44aa6a5663c7e193559fdbd159ed6e3928ec5304ccce731369"
393
+ },
394
+ {
395
+ "path": "packed/train-000013-role_ids.npy",
396
+ "bytes": 2097280,
397
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
398
+ },
399
+ {
400
+ "path": "packed/train-000014-input_ids.npy",
401
+ "bytes": 2097280,
402
+ "sha256": "bae7e9820f78c3ea3d1bd2e97315a8b41aeb57a93b6f9e408d2daa721dfebca8"
403
+ },
404
+ {
405
+ "path": "packed/train-000014-labels.npy",
406
+ "bytes": 2097280,
407
+ "sha256": "35ee3e2bfe0464ffa0a62c6fb83dedfd216651badb39803afbb96f2253a02809"
408
+ },
409
+ {
410
+ "path": "packed/train-000014-role_ids.npy",
411
+ "bytes": 2097280,
412
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
413
+ },
414
+ {
415
+ "path": "packed/train-000015-input_ids.npy",
416
+ "bytes": 2097280,
417
+ "sha256": "af7b863da4adb1477edf52e97f9fff0e8af51345274f134f9e281366fad8b947"
418
+ },
419
+ {
420
+ "path": "packed/train-000015-labels.npy",
421
+ "bytes": 2097280,
422
+ "sha256": "08923fae4051d6fd86976fa67c127158078b1bc49978c1637283078be260f79a"
423
+ },
424
+ {
425
+ "path": "packed/train-000015-role_ids.npy",
426
+ "bytes": 2097280,
427
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
428
+ },
429
+ {
430
+ "path": "packed/train-000016-input_ids.npy",
431
+ "bytes": 2097280,
432
+ "sha256": "ff2f796edb333e540435cfe19f46ae58087fb2d67b3ae25ba3f3a1eff74cf4e2"
433
+ },
434
+ {
435
+ "path": "packed/train-000016-labels.npy",
436
+ "bytes": 2097280,
437
+ "sha256": "3f76693bba77a67c0a6d8f8fef28fd77251937c459cf63e50a404e1ffd0a8908"
438
+ },
439
+ {
440
+ "path": "packed/train-000016-role_ids.npy",
441
+ "bytes": 2097280,
442
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
443
+ },
444
+ {
445
+ "path": "packed/train-000017-input_ids.npy",
446
+ "bytes": 2097280,
447
+ "sha256": "4c9ce006e3e08a5a03ad92561250d42724ac914f7223f2ac1408ea7b2ff98ed3"
448
+ },
449
+ {
450
+ "path": "packed/train-000017-labels.npy",
451
+ "bytes": 2097280,
452
+ "sha256": "adff7be1924ab55845f6cc60f847a26a98cbb011b927019babfd45656128cb21"
453
+ },
454
+ {
455
+ "path": "packed/train-000017-role_ids.npy",
456
+ "bytes": 2097280,
457
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
458
+ },
459
+ {
460
+ "path": "packed/train-000018-input_ids.npy",
461
+ "bytes": 2097280,
462
+ "sha256": "8128ef4044063a7ed94766be978ccf9ac6a76f9871abe218c752fa14c4b92da4"
463
+ },
464
+ {
465
+ "path": "packed/train-000018-labels.npy",
466
+ "bytes": 2097280,
467
+ "sha256": "3e631a4858ee21f27292c9e4cea8852512d520aaf708fdfc8bbab3b18fc5817d"
468
+ },
469
+ {
470
+ "path": "packed/train-000018-role_ids.npy",
471
+ "bytes": 2097280,
472
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
473
+ },
474
+ {
475
+ "path": "packed/train-000019-input_ids.npy",
476
+ "bytes": 2097280,
477
+ "sha256": "41c7fba439b9e2838949d29a03c15c106bf3a715c39d7ecf731e86a7601232e5"
478
+ },
479
+ {
480
+ "path": "packed/train-000019-labels.npy",
481
+ "bytes": 2097280,
482
+ "sha256": "4f95df70f8b1df91a6215871919dba1a7867223b71951f47d9e64c7621e29b3a"
483
+ },
484
+ {
485
+ "path": "packed/train-000019-role_ids.npy",
486
+ "bytes": 2097280,
487
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
488
+ },
489
+ {
490
+ "path": "packed/train-000020-input_ids.npy",
491
+ "bytes": 2097280,
492
+ "sha256": "25a5c4bdb74c983a424c0e2ce5546c7017e3a883c71b0e7db63c62cce15bea40"
493
+ },
494
+ {
495
+ "path": "packed/train-000020-labels.npy",
496
+ "bytes": 2097280,
497
+ "sha256": "8ad4cdb467568516f05bd15c42f5c833b41a233a465c9836857bcc37a65414a5"
498
+ },
499
+ {
500
+ "path": "packed/train-000020-role_ids.npy",
501
+ "bytes": 2097280,
502
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
503
+ },
504
+ {
505
+ "path": "packed/train-000021-input_ids.npy",
506
+ "bytes": 2097280,
507
+ "sha256": "fba97f9955af982b7f39192f390041a035d3b0e6542c8559d6cf0ce5ff2a46e2"
508
+ },
509
+ {
510
+ "path": "packed/train-000021-labels.npy",
511
+ "bytes": 2097280,
512
+ "sha256": "0ced1222de8dc2d1f707ec6f69d81ff5638ce2012e1e1e0f5df0a00a94f7ad52"
513
+ },
514
+ {
515
+ "path": "packed/train-000021-role_ids.npy",
516
+ "bytes": 2097280,
517
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
518
+ },
519
+ {
520
+ "path": "packed/train-000022-input_ids.npy",
521
+ "bytes": 2097280,
522
+ "sha256": "8e68a48c63e0f471b3e07c6f600191b3dba89bef779862a4db080dd35fc16d7b"
523
+ },
524
+ {
525
+ "path": "packed/train-000022-labels.npy",
526
+ "bytes": 2097280,
527
+ "sha256": "70b04cb52c290d37995ef16f890ec3956942e2295e598a8054b439120072d852"
528
+ },
529
+ {
530
+ "path": "packed/train-000022-role_ids.npy",
531
+ "bytes": 2097280,
532
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
533
+ },
534
+ {
535
+ "path": "packed/train-000023-input_ids.npy",
536
+ "bytes": 2097280,
537
+ "sha256": "4aba6fc6870cb1b7e9174875e04d8a760889a6521b6dd5ae65e7933ded7af7b1"
538
+ },
539
+ {
540
+ "path": "packed/train-000023-labels.npy",
541
+ "bytes": 2097280,
542
+ "sha256": "bbd9c145b2dce65a4f99d1a3bd2d5c80278c75148947591ef6988d57ad1c5c91"
543
+ },
544
+ {
545
+ "path": "packed/train-000023-role_ids.npy",
546
+ "bytes": 2097280,
547
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
548
+ },
549
+ {
550
+ "path": "packed/train-000024-input_ids.npy",
551
+ "bytes": 2097280,
552
+ "sha256": "95334eaef896fa57231cad20c9e608bbb60ba22f126e6fd42193780487934bbe"
553
+ },
554
+ {
555
+ "path": "packed/train-000024-labels.npy",
556
+ "bytes": 2097280,
557
+ "sha256": "ec4ddc79fe710089a45b4b3cc0a4be2f66673b8b7169fcbcc80f9c88eea67982"
558
+ },
559
+ {
560
+ "path": "packed/train-000024-role_ids.npy",
561
+ "bytes": 2097280,
562
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
563
+ },
564
+ {
565
+ "path": "packed/train-000025-input_ids.npy",
566
+ "bytes": 2097280,
567
+ "sha256": "16ad874b80fce0e07c259ae151226bd7a047e0e4686f834893be2f223710f969"
568
+ },
569
+ {
570
+ "path": "packed/train-000025-labels.npy",
571
+ "bytes": 2097280,
572
+ "sha256": "981db3d9deb29f76c35438c7734650de0f65dafc29b7c8f986b161392c0824d8"
573
+ },
574
+ {
575
+ "path": "packed/train-000025-role_ids.npy",
576
+ "bytes": 2097280,
577
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
578
+ },
579
+ {
580
+ "path": "packed/train-000026-input_ids.npy",
581
+ "bytes": 2097280,
582
+ "sha256": "429d0ad049d9297785412272297af1f9002c7121fbe119eb639333343b54a2e3"
583
+ },
584
+ {
585
+ "path": "packed/train-000026-labels.npy",
586
+ "bytes": 2097280,
587
+ "sha256": "9f94680c036dc6aeba93b2ddee1dfb414de3b29f0ce78276b0182492e73edeba"
588
+ },
589
+ {
590
+ "path": "packed/train-000026-role_ids.npy",
591
+ "bytes": 2097280,
592
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
593
+ },
594
+ {
595
+ "path": "packed/train-000027-input_ids.npy",
596
+ "bytes": 2097280,
597
+ "sha256": "b8f9daa5f0225b1bf9f70c09728d51f9f1d207dadfa2c9f8efba007851c83bb1"
598
+ },
599
+ {
600
+ "path": "packed/train-000027-labels.npy",
601
+ "bytes": 2097280,
602
+ "sha256": "c3255681088fa15a812ea691c1284b75998dd6484b6bb7b82a2d697e9945d58a"
603
+ },
604
+ {
605
+ "path": "packed/train-000027-role_ids.npy",
606
+ "bytes": 2097280,
607
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
608
+ },
609
+ {
610
+ "path": "packed/train-000028-input_ids.npy",
611
+ "bytes": 2097280,
612
+ "sha256": "fb7cc7d62aea73f14115a33c2fb27a577444be99925da0926177ad0b92299e16"
613
+ },
614
+ {
615
+ "path": "packed/train-000028-labels.npy",
616
+ "bytes": 2097280,
617
+ "sha256": "2ef7e889fd50221c6cdca7ac4566f35426fe194ae818742750340ff6527e499f"
618
+ },
619
+ {
620
+ "path": "packed/train-000028-role_ids.npy",
621
+ "bytes": 2097280,
622
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
623
+ },
624
+ {
625
+ "path": "packed/train-000029-input_ids.npy",
626
+ "bytes": 2097280,
627
+ "sha256": "43aaf0846859c9560517703fde38443ea99868bdbc286fd7e3116c15bf39c350"
628
+ },
629
+ {
630
+ "path": "packed/train-000029-labels.npy",
631
+ "bytes": 2097280,
632
+ "sha256": "d5c945896a10d25dc8a447d481c1b999953a9552ac7dca7e3dfc1e6c37e8c467"
633
+ },
634
+ {
635
+ "path": "packed/train-000029-role_ids.npy",
636
+ "bytes": 2097280,
637
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
638
+ },
639
+ {
640
+ "path": "packed/train-000030-input_ids.npy",
641
+ "bytes": 2097280,
642
+ "sha256": "5acdbaad19838986d9f568b174cd7dd5bc8cc2dadd54ffffa93ea844da874459"
643
+ },
644
+ {
645
+ "path": "packed/train-000030-labels.npy",
646
+ "bytes": 2097280,
647
+ "sha256": "e7f2b43c5f2d1d3bfee5ff7e1d0f7eb8594c93e635611ed39f5b160fb82b5732"
648
+ },
649
+ {
650
+ "path": "packed/train-000030-role_ids.npy",
651
+ "bytes": 2097280,
652
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
653
+ },
654
+ {
655
+ "path": "packed/train-000031-input_ids.npy",
656
+ "bytes": 2097280,
657
+ "sha256": "a4259d999894e0e8521f2a5df0ad1b1971b6d9a85b24d0b2602fe53d26079b09"
658
+ },
659
+ {
660
+ "path": "packed/train-000031-labels.npy",
661
+ "bytes": 2097280,
662
+ "sha256": "4135166083f4834f7f56aa303668285c006c5648223575539b2f3c7830062d3e"
663
+ },
664
+ {
665
+ "path": "packed/train-000031-role_ids.npy",
666
+ "bytes": 2097280,
667
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
668
+ },
669
+ {
670
+ "path": "packed/train-000032-input_ids.npy",
671
+ "bytes": 2097280,
672
+ "sha256": "93800c3e70177c4ffa6cb024674849b86a9ddbba2037650bc881dbe5bacf5ba7"
673
+ },
674
+ {
675
+ "path": "packed/train-000032-labels.npy",
676
+ "bytes": 2097280,
677
+ "sha256": "ec6ad678dd6c9b3800068a676918bb552c3e1542e382520932c42361ec1f750c"
678
+ },
679
+ {
680
+ "path": "packed/train-000032-role_ids.npy",
681
+ "bytes": 2097280,
682
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
683
+ },
684
+ {
685
+ "path": "packed/train-000033-input_ids.npy",
686
+ "bytes": 2097280,
687
+ "sha256": "589c7e36015ae45b44e2e9e261f6b869492210db222316a9b809cb2bf35201b9"
688
+ },
689
+ {
690
+ "path": "packed/train-000033-labels.npy",
691
+ "bytes": 2097280,
692
+ "sha256": "b1c36ea4c563437275e083e3350faabf6cdf2f7fd6415425bad419111a675f07"
693
+ },
694
+ {
695
+ "path": "packed/train-000033-role_ids.npy",
696
+ "bytes": 2097280,
697
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
698
+ },
699
+ {
700
+ "path": "packed/train-000034-input_ids.npy",
701
+ "bytes": 2097280,
702
+ "sha256": "e05c2bd95c1812922cb6f6c6ae660dee5e2665a81d77aa742b571713ed526da7"
703
+ },
704
+ {
705
+ "path": "packed/train-000034-labels.npy",
706
+ "bytes": 2097280,
707
+ "sha256": "3d64017781a0c113d4819cbb7ef8ab2dd8a09ef449c9d2ffea6bacbd3a3fa045"
708
+ },
709
+ {
710
+ "path": "packed/train-000034-role_ids.npy",
711
+ "bytes": 2097280,
712
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
713
+ },
714
+ {
715
+ "path": "packed/train-000035-input_ids.npy",
716
+ "bytes": 2097280,
717
+ "sha256": "41597052b2d82fde4b0c750345e29be463661d87ecd4781ac6fbf18f36ad8ad9"
718
+ },
719
+ {
720
+ "path": "packed/train-000035-labels.npy",
721
+ "bytes": 2097280,
722
+ "sha256": "ea0f0354482fd8a2d8ff12ac51cfd01ae7c5b98b6c75b8c202974f7ee429c7d8"
723
+ },
724
+ {
725
+ "path": "packed/train-000035-role_ids.npy",
726
+ "bytes": 2097280,
727
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
728
+ },
729
+ {
730
+ "path": "packed/train-000036-input_ids.npy",
731
+ "bytes": 2097280,
732
+ "sha256": "c6f77f37c2d46f6bb8b12a54080d381f63c3cea52a2dffddde0ac5f59ee29f59"
733
+ },
734
+ {
735
+ "path": "packed/train-000036-labels.npy",
736
+ "bytes": 2097280,
737
+ "sha256": "8b6cb52f1e9d317228da9cb59ad4d8c1786ccf88dbf7e10d9409afbdab98cecf"
738
+ },
739
+ {
740
+ "path": "packed/train-000036-role_ids.npy",
741
+ "bytes": 2097280,
742
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
743
+ },
744
+ {
745
+ "path": "packed/train-000037-input_ids.npy",
746
+ "bytes": 2097280,
747
+ "sha256": "af2e0101faa197ba7d7be8bef35e08de158ea09d319f5e709bd7ad899e89a540"
748
+ },
749
+ {
750
+ "path": "packed/train-000037-labels.npy",
751
+ "bytes": 2097280,
752
+ "sha256": "2df9fc4d1a20cba0675f1c090548510d4be92f76fdcbf7695101042736ac6088"
753
+ },
754
+ {
755
+ "path": "packed/train-000037-role_ids.npy",
756
+ "bytes": 2097280,
757
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
758
+ },
759
+ {
760
+ "path": "packed/train-000038-input_ids.npy",
761
+ "bytes": 2097280,
762
+ "sha256": "be988ef632ba39715c5088e292e66d59aef16f6511bbaec844ba9ce0f8c937ea"
763
+ },
764
+ {
765
+ "path": "packed/train-000038-labels.npy",
766
+ "bytes": 2097280,
767
+ "sha256": "32cc5d079f3cc3d94576ce456589d0082506c90023875720392fc4ab98555602"
768
+ },
769
+ {
770
+ "path": "packed/train-000038-role_ids.npy",
771
+ "bytes": 2097280,
772
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
773
+ },
774
+ {
775
+ "path": "packed/train-000039-input_ids.npy",
776
+ "bytes": 2097280,
777
+ "sha256": "22be43231accce4101ec2134ccde391016c2febb9a149ba38cc78c9755c51316"
778
+ },
779
+ {
780
+ "path": "packed/train-000039-labels.npy",
781
+ "bytes": 2097280,
782
+ "sha256": "cb8ec58a50a0b7affd18a016da1d9e5ed52de31f60a872ac5268a7d61805e84a"
783
+ },
784
+ {
785
+ "path": "packed/train-000039-role_ids.npy",
786
+ "bytes": 2097280,
787
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
788
+ },
789
+ {
790
+ "path": "packed/train-000040-input_ids.npy",
791
+ "bytes": 2097280,
792
+ "sha256": "4940ebee2e5cd46abd7140a673fe7887eaf59b7e93cc756ca958bdc64ce86ba5"
793
+ },
794
+ {
795
+ "path": "packed/train-000040-labels.npy",
796
+ "bytes": 2097280,
797
+ "sha256": "fd48fb9652d2f65e8cea3c2732805b81cc1a086c9768a21b34a5b61480af47a6"
798
+ },
799
+ {
800
+ "path": "packed/train-000040-role_ids.npy",
801
+ "bytes": 2097280,
802
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
803
+ },
804
+ {
805
+ "path": "packed/train-000041-input_ids.npy",
806
+ "bytes": 2097280,
807
+ "sha256": "446fdff487505d3081342132b588ddc35bbded8a369e01bb370e2b681c9cc041"
808
+ },
809
+ {
810
+ "path": "packed/train-000041-labels.npy",
811
+ "bytes": 2097280,
812
+ "sha256": "6fe6f9361c5ec507624bdb6eefc60700dd72e45f02019747106edc743f43d3bb"
813
+ },
814
+ {
815
+ "path": "packed/train-000041-role_ids.npy",
816
+ "bytes": 2097280,
817
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
818
+ },
819
+ {
820
+ "path": "packed/train-000042-input_ids.npy",
821
+ "bytes": 2097280,
822
+ "sha256": "63680585929747ac31b3192263feabce626826b2181cd93c149656ebec818440"
823
+ },
824
+ {
825
+ "path": "packed/train-000042-labels.npy",
826
+ "bytes": 2097280,
827
+ "sha256": "726027d7d221a2f6cfd6e8f2552d8c3bad5663d4108e9f3c8ee5c9fee4cb6dbc"
828
+ },
829
+ {
830
+ "path": "packed/train-000042-role_ids.npy",
831
+ "bytes": 2097280,
832
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
833
+ },
834
+ {
835
+ "path": "packed/train-000043-input_ids.npy",
836
+ "bytes": 2097280,
837
+ "sha256": "25dd15b2797b9f6d09e6e78632341a16299c94e2e79594fde0614d96c5860313"
838
+ },
839
+ {
840
+ "path": "packed/train-000043-labels.npy",
841
+ "bytes": 2097280,
842
+ "sha256": "ed9727af9d3e5d11d48388f2274bc07298f43be71f61022015e28ebc74ff3e25"
843
+ },
844
+ {
845
+ "path": "packed/train-000043-role_ids.npy",
846
+ "bytes": 2097280,
847
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
848
+ },
849
+ {
850
+ "path": "packed/train-000044-input_ids.npy",
851
+ "bytes": 2097280,
852
+ "sha256": "da874ad45be5ace45ce3d145099afec13087437b03102f714417f0be5a0dfe03"
853
+ },
854
+ {
855
+ "path": "packed/train-000044-labels.npy",
856
+ "bytes": 2097280,
857
+ "sha256": "b5c0761906ea7b7d5ecb55d510257d78c69f939a137f34c6f3cbbfdbb99cdbf1"
858
+ },
859
+ {
860
+ "path": "packed/train-000044-role_ids.npy",
861
+ "bytes": 2097280,
862
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
863
+ },
864
+ {
865
+ "path": "packed/train-000045-input_ids.npy",
866
+ "bytes": 2097280,
867
+ "sha256": "1bd80c45454952f73b8ffcc16c8e46fe3f00d1f5d99f31016a8f9e82e7d5c99c"
868
+ },
869
+ {
870
+ "path": "packed/train-000045-labels.npy",
871
+ "bytes": 2097280,
872
+ "sha256": "f88abf7de5c11322c52fc6fcd1d9f5358c6c13dc0820255d67d838de4b1a7990"
873
+ },
874
+ {
875
+ "path": "packed/train-000045-role_ids.npy",
876
+ "bytes": 2097280,
877
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
878
+ },
879
+ {
880
+ "path": "packed/train-000046-input_ids.npy",
881
+ "bytes": 2097280,
882
+ "sha256": "e5124005f12d4f9ba66a7a7c39b64c36e3496fe9c01b6eb87c4d20aff3aa82cc"
883
+ },
884
+ {
885
+ "path": "packed/train-000046-labels.npy",
886
+ "bytes": 2097280,
887
+ "sha256": "f3114fbd59a5c15b99bbb7fd7d342ba0c7665767a1f0c7bbcadacd72e5eae698"
888
+ },
889
+ {
890
+ "path": "packed/train-000046-role_ids.npy",
891
+ "bytes": 2097280,
892
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
893
+ },
894
+ {
895
+ "path": "packed/train-000047-input_ids.npy",
896
+ "bytes": 2097280,
897
+ "sha256": "3e2cce141aae1ab0767cfdc74256bd8c06b879c820d069bc8039ce880c09d912"
898
+ },
899
+ {
900
+ "path": "packed/train-000047-labels.npy",
901
+ "bytes": 2097280,
902
+ "sha256": "07e444973d506cd93e54dec91c546e097f7f600bf1114de1aaf77113746da321"
903
+ },
904
+ {
905
+ "path": "packed/train-000047-role_ids.npy",
906
+ "bytes": 2097280,
907
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
908
+ },
909
+ {
910
+ "path": "packed/train-000048-input_ids.npy",
911
+ "bytes": 2097280,
912
+ "sha256": "7a625331b5d84b9ac15bda3141779810c13dab69b544bf02883a51a4dcfbd28b"
913
+ },
914
+ {
915
+ "path": "packed/train-000048-labels.npy",
916
+ "bytes": 2097280,
917
+ "sha256": "c9882f15954c8c3cefac98f117c5412530520c2e4cbd81160effee96fead3e82"
918
+ },
919
+ {
920
+ "path": "packed/train-000048-role_ids.npy",
921
+ "bytes": 2097280,
922
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
923
+ },
924
+ {
925
+ "path": "packed/train-000049-input_ids.npy",
926
+ "bytes": 2097280,
927
+ "sha256": "f4e30c9e064892866493a55a92c0e8597fa456f49b4983fc4b1e21372b949ef5"
928
+ },
929
+ {
930
+ "path": "packed/train-000049-labels.npy",
931
+ "bytes": 2097280,
932
+ "sha256": "2f785cbee1858a93601eb7039582b853987172f5cfff8aa4686406b19af39713"
933
+ },
934
+ {
935
+ "path": "packed/train-000049-role_ids.npy",
936
+ "bytes": 2097280,
937
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
938
+ },
939
+ {
940
+ "path": "packed/train-000050-input_ids.npy",
941
+ "bytes": 2097280,
942
+ "sha256": "5dd2e1de773cee59108195c4c82e9ce71b6275d1d0568027826809ecf9a34711"
943
+ },
944
+ {
945
+ "path": "packed/train-000050-labels.npy",
946
+ "bytes": 2097280,
947
+ "sha256": "32e3374f97ad75b1825a1dc7911166367427d3632271c282cd2f846bad65ef93"
948
+ },
949
+ {
950
+ "path": "packed/train-000050-role_ids.npy",
951
+ "bytes": 2097280,
952
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
953
+ },
954
+ {
955
+ "path": "packed/train-000051-input_ids.npy",
956
+ "bytes": 2097280,
957
+ "sha256": "1e3499eec68612a9e94dca02eaa4979829e00dddcaa9dbcb97fa99fa0f1952e0"
958
+ },
959
+ {
960
+ "path": "packed/train-000051-labels.npy",
961
+ "bytes": 2097280,
962
+ "sha256": "1ab08e49815ccde21199510985d111497e0c350ffb8436974ac035efb045b293"
963
+ },
964
+ {
965
+ "path": "packed/train-000051-role_ids.npy",
966
+ "bytes": 2097280,
967
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
968
+ },
969
+ {
970
+ "path": "packed/train-000052-input_ids.npy",
971
+ "bytes": 2097280,
972
+ "sha256": "f5d1597708340ccad2879fcedcf0106e2b601712c773e78170710afa4622a9d4"
973
+ },
974
+ {
975
+ "path": "packed/train-000052-labels.npy",
976
+ "bytes": 2097280,
977
+ "sha256": "bb27e5f7dbc1722f934d9937654e2482aa28d23fd1618e94b823f98c89b5dc82"
978
+ },
979
+ {
980
+ "path": "packed/train-000052-role_ids.npy",
981
+ "bytes": 2097280,
982
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
983
+ },
984
+ {
985
+ "path": "packed/train-000053-input_ids.npy",
986
+ "bytes": 2097280,
987
+ "sha256": "ac0f2d3c56af78be929914893c21a5fdcc090a24a0982d5c7b452dda771c0b0d"
988
+ },
989
+ {
990
+ "path": "packed/train-000053-labels.npy",
991
+ "bytes": 2097280,
992
+ "sha256": "24dfc5e107b32c82c790385b6d1e7a4ced11cc0e6f43826bf12e525dd2cce5d3"
993
+ },
994
+ {
995
+ "path": "packed/train-000053-role_ids.npy",
996
+ "bytes": 2097280,
997
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
998
+ },
999
+ {
1000
+ "path": "packed/train-000054-input_ids.npy",
1001
+ "bytes": 2097280,
1002
+ "sha256": "32fc8e24ac0f37b566e66050cb96be3fa3292eca3af86ae0926658c8c6829571"
1003
+ },
1004
+ {
1005
+ "path": "packed/train-000054-labels.npy",
1006
+ "bytes": 2097280,
1007
+ "sha256": "e0ae7932cc20a16bb3efc58e1ed7d7619d2a0b43cdc370d60e8ed27b6fb8765b"
1008
+ },
1009
+ {
1010
+ "path": "packed/train-000054-role_ids.npy",
1011
+ "bytes": 2097280,
1012
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1013
+ },
1014
+ {
1015
+ "path": "packed/train-000055-input_ids.npy",
1016
+ "bytes": 2097280,
1017
+ "sha256": "0f4262abf457d5503cfa741eca3f51f6034981a82bfe56db902405e992073302"
1018
+ },
1019
+ {
1020
+ "path": "packed/train-000055-labels.npy",
1021
+ "bytes": 2097280,
1022
+ "sha256": "c6ca503dde0a9b7d964fa7c9fd61c998c08fab856348f6863516368b4dcd00e2"
1023
+ },
1024
+ {
1025
+ "path": "packed/train-000055-role_ids.npy",
1026
+ "bytes": 2097280,
1027
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1028
+ },
1029
+ {
1030
+ "path": "packed/train-000056-input_ids.npy",
1031
+ "bytes": 2097280,
1032
+ "sha256": "4b2e522f4d2638510036e6bbf08b7606b0d9676ff67a24127d906ea7321082d0"
1033
+ },
1034
+ {
1035
+ "path": "packed/train-000056-labels.npy",
1036
+ "bytes": 2097280,
1037
+ "sha256": "55ca218fb5cdbf740630d5e33ecc451350d8378c1f6daa40b26f48a01bc4121d"
1038
+ },
1039
+ {
1040
+ "path": "packed/train-000056-role_ids.npy",
1041
+ "bytes": 2097280,
1042
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1043
+ },
1044
+ {
1045
+ "path": "packed/train-000057-input_ids.npy",
1046
+ "bytes": 2097280,
1047
+ "sha256": "a33ab1bf70cf57937b1902dcb3bc4eec84c4845a08ffe07c77cbf88e671a18b4"
1048
+ },
1049
+ {
1050
+ "path": "packed/train-000057-labels.npy",
1051
+ "bytes": 2097280,
1052
+ "sha256": "aec8942f726c9647d89139a563e2fb6c9b64cc1b109238b6468f172e39e295d9"
1053
+ },
1054
+ {
1055
+ "path": "packed/train-000057-role_ids.npy",
1056
+ "bytes": 2097280,
1057
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1058
+ },
1059
+ {
1060
+ "path": "packed/train-000058-input_ids.npy",
1061
+ "bytes": 2097280,
1062
+ "sha256": "92ac5539461e028cdfeec78e1fd9e1390022602eba5f3ed3b784c7ec80940ae8"
1063
+ },
1064
+ {
1065
+ "path": "packed/train-000058-labels.npy",
1066
+ "bytes": 2097280,
1067
+ "sha256": "73a20209c9c3def7d91670eaf773c1741418364dae6e6212961aa5604e99eb6a"
1068
+ },
1069
+ {
1070
+ "path": "packed/train-000058-role_ids.npy",
1071
+ "bytes": 2097280,
1072
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1073
+ },
1074
+ {
1075
+ "path": "packed/train-000059-input_ids.npy",
1076
+ "bytes": 2097280,
1077
+ "sha256": "220a6fcf419fff6aeba473453f5a93911a64285a5c2a93d1555274958ae6d956"
1078
+ },
1079
+ {
1080
+ "path": "packed/train-000059-labels.npy",
1081
+ "bytes": 2097280,
1082
+ "sha256": "c6a2fc3b601f12bb7a8c17d13643df34fc24a4deb48fe4f18d8aa223818796d5"
1083
+ },
1084
+ {
1085
+ "path": "packed/train-000059-role_ids.npy",
1086
+ "bytes": 2097280,
1087
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1088
+ },
1089
+ {
1090
+ "path": "packed/train-000060-input_ids.npy",
1091
+ "bytes": 2097280,
1092
+ "sha256": "36fa1b3c8e924e9d838671bd6f80207f9c7cfe0259b1c2a2df1e9d31d8714b83"
1093
+ },
1094
+ {
1095
+ "path": "packed/train-000060-labels.npy",
1096
+ "bytes": 2097280,
1097
+ "sha256": "59da1eafff19846b203f1b8fa9957085f22d4cd8e8660544fbcbfc92befeb492"
1098
+ },
1099
+ {
1100
+ "path": "packed/train-000060-role_ids.npy",
1101
+ "bytes": 2097280,
1102
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1103
+ },
1104
+ {
1105
+ "path": "packed/train-000061-input_ids.npy",
1106
+ "bytes": 2097280,
1107
+ "sha256": "eea8e262e8a9fbc874c33849de2a273ef51355c7da342c0c2fe6b4de3c5cb358"
1108
+ },
1109
+ {
1110
+ "path": "packed/train-000061-labels.npy",
1111
+ "bytes": 2097280,
1112
+ "sha256": "d5dc71fa8970ea88b2664d4aef619e218222513fb624639219f849caedbd53c3"
1113
+ },
1114
+ {
1115
+ "path": "packed/train-000061-role_ids.npy",
1116
+ "bytes": 2097280,
1117
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1118
+ },
1119
+ {
1120
+ "path": "packed/train-000062-input_ids.npy",
1121
+ "bytes": 2097280,
1122
+ "sha256": "86929725ea778db63dbc4e7173552e88d16f99fe6bb3e33b88aacd0c3d4761a2"
1123
+ },
1124
+ {
1125
+ "path": "packed/train-000062-labels.npy",
1126
+ "bytes": 2097280,
1127
+ "sha256": "061af96ab76ecd7e2a7f9e6d9ecb46c77439e96007f5276658d69692b742ccb1"
1128
+ },
1129
+ {
1130
+ "path": "packed/train-000062-role_ids.npy",
1131
+ "bytes": 2097280,
1132
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1133
+ },
1134
+ {
1135
+ "path": "packed/train-000063-input_ids.npy",
1136
+ "bytes": 2097280,
1137
+ "sha256": "cd0d4948b093ec67f117dc67eb3677c4d53b7f9f5a1a13571d538394f6c2cd89"
1138
+ },
1139
+ {
1140
+ "path": "packed/train-000063-labels.npy",
1141
+ "bytes": 2097280,
1142
+ "sha256": "70bfaaa5a0ec9eb15dee44d4a7de2a0ee60c29b95c39b54265b1945c03702eee"
1143
+ },
1144
+ {
1145
+ "path": "packed/train-000063-role_ids.npy",
1146
+ "bytes": 2097280,
1147
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1148
+ },
1149
+ {
1150
+ "path": "packed/train-000064-input_ids.npy",
1151
+ "bytes": 2097280,
1152
+ "sha256": "f6737c2e565217512d61657e74c18a437ce19bec85f0835ca911271b476992fd"
1153
+ },
1154
+ {
1155
+ "path": "packed/train-000064-labels.npy",
1156
+ "bytes": 2097280,
1157
+ "sha256": "1c4176816d9fcf2fd24bf3a9cc2558af3d2d204775e034f1c5e9a511c1ca068f"
1158
+ },
1159
+ {
1160
+ "path": "packed/train-000064-role_ids.npy",
1161
+ "bytes": 2097280,
1162
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1163
+ },
1164
+ {
1165
+ "path": "packed/train-000065-input_ids.npy",
1166
+ "bytes": 2097280,
1167
+ "sha256": "93409b3cc645153bd21251245922a6520f04aef969931515f461a97406266468"
1168
+ },
1169
+ {
1170
+ "path": "packed/train-000065-labels.npy",
1171
+ "bytes": 2097280,
1172
+ "sha256": "144ae070409a543e7ededf8e0af09d77c3f34d994c6373687761d3060206ee87"
1173
+ },
1174
+ {
1175
+ "path": "packed/train-000065-role_ids.npy",
1176
+ "bytes": 2097280,
1177
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1178
+ },
1179
+ {
1180
+ "path": "packed/train-000066-input_ids.npy",
1181
+ "bytes": 2097280,
1182
+ "sha256": "40e8f59e6bc1b5f35726fc16a09ba662c93b8428c5fe67ccc513e79bbcad26c7"
1183
+ },
1184
+ {
1185
+ "path": "packed/train-000066-labels.npy",
1186
+ "bytes": 2097280,
1187
+ "sha256": "db2ad523b3e5357173bc780d33b4fc463e4dabf1399f65b066b51b6a5ea066f7"
1188
+ },
1189
+ {
1190
+ "path": "packed/train-000066-role_ids.npy",
1191
+ "bytes": 2097280,
1192
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1193
+ },
1194
+ {
1195
+ "path": "packed/train-000067-input_ids.npy",
1196
+ "bytes": 2097280,
1197
+ "sha256": "fc6f0197b503e74f6b16875b8e9712b002577bc5b20739527a240ec5104b24aa"
1198
+ },
1199
+ {
1200
+ "path": "packed/train-000067-labels.npy",
1201
+ "bytes": 2097280,
1202
+ "sha256": "d5397068dc3e8b9ebf55ce60256fe148b7cf133256f3c58a912c7dc10521f876"
1203
+ },
1204
+ {
1205
+ "path": "packed/train-000067-role_ids.npy",
1206
+ "bytes": 2097280,
1207
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1208
+ },
1209
+ {
1210
+ "path": "packed/train-000068-input_ids.npy",
1211
+ "bytes": 2097280,
1212
+ "sha256": "a36f854b9e96070665d3efad49d177b1ad440b97251b29a1017f83886845eb42"
1213
+ },
1214
+ {
1215
+ "path": "packed/train-000068-labels.npy",
1216
+ "bytes": 2097280,
1217
+ "sha256": "f0b90c34fb81ef32a025b1f92f49ffb54f2f219ad6f1250403b7f7640e1c7480"
1218
+ },
1219
+ {
1220
+ "path": "packed/train-000068-role_ids.npy",
1221
+ "bytes": 2097280,
1222
+ "sha256": "e8af22599bd2b950a4c4caafb09c863f5ab033a9ef09e030d08cce3c9e55506b"
1223
+ },
1224
+ {
1225
+ "path": "packed/train-000069-input_ids.npy",
1226
+ "bytes": 221312,
1227
+ "sha256": "48c46e4c77a6921611664188acca32427723a896121bbd3d8460a9ec04cdc9bb"
1228
+ },
1229
+ {
1230
+ "path": "packed/train-000069-labels.npy",
1231
+ "bytes": 221312,
1232
+ "sha256": "857ee31ddc5d71ab27b864d2fb05689af2c3173b839bc8ec24bb74210e460a90"
1233
+ },
1234
+ {
1235
+ "path": "packed/train-000069-role_ids.npy",
1236
+ "bytes": 221312,
1237
+ "sha256": "67e2cd22d28bcb409250df033260a0fa936a59a76893799ac171c73f0cafe062"
1238
+ },
1239
+ {
1240
+ "path": "packed/train-records.jsonl",
1241
+ "bytes": 100739210,
1242
+ "sha256": "05ad0c687f008c7a53525af094216ae6ea7ef0c3aa472fcb8da7a49e01cf39d4"
1243
+ },
1244
+ {
1245
+ "path": "packed/train-rows.jsonl",
1246
+ "bytes": 95002664,
1247
+ "sha256": "f61b4db0626dffe9d78b841092f41e17d08dff8a6c1470329634d4cd4a7498e6"
1248
+ },
1249
+ {
1250
+ "path": "sources.json",
1251
+ "bytes": 12145,
1252
+ "sha256": "325b92dc093211301c620b3213030caaf8b11be7896bb6ce3928bad0c52f93cc"
1253
+ },
1254
+ {
1255
+ "path": "statistics.json",
1256
+ "bytes": 4262,
1257
+ "sha256": "6f5c0bd4ef2301ed4b671ab9aac8deffeeae9af92ff7316570363cc32df6fbe4"
1258
+ },
1259
+ {
1260
+ "path": "tokenizer/tokenizer.model",
1261
+ "bytes": 503703,
1262
+ "sha256": "18b5aa749c6c791ab59c2fd1eb39dfbaa155247322119d3924d3cb9768622417"
1263
+ },
1264
+ {
1265
+ "path": "tools/build_google_math.py",
1266
+ "bytes": 61871,
1267
+ "sha256": "8e72001f8b24f5d188c73da7f4e9e29057300e5fa279e8d370f297375f6202d4"
1268
+ },
1269
+ {
1270
+ "path": "tools/generate_mathematics_dataset_sample.py",
1271
+ "bytes": 5319,
1272
+ "sha256": "8979306442033cc8cc6d71cb8aba069df6237ee93e1f07d13d86c095daf8cab1"
1273
+ },
1274
+ {
1275
+ "path": "tools/validate_google_math.py",
1276
+ "bytes": 22311,
1277
+ "sha256": "86fbfbfd0db16e8fcbff9476a38d1af92b4cddacdf975dcf6abab4f9d90b0aca"
1278
+ },
1279
+ {
1280
+ "path": "tools/verify_google_math.py",
1281
+ "bytes": 12954,
1282
+ "sha256": "9e1960faa56de9f6128cee519cf682b7e20d277326597de5bf8d235ef56a283a"
1283
+ }
1284
+ ]
1285
+ }
components/google-math/tools/build_google_math.py ADDED
@@ -0,0 +1,1141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Build components/google-math: Google math-paper artifacts in Cayley-Math format.
3
+
4
+ One component, eight upstream subsets. Mirrors components/openai-math exactly:
5
+ same record core fields, same Parquet encoding, same SHA256-ordered group split
6
+ policy, and the same mmla-type2-packed-v1 packing code.
7
+
8
+ Nothing here verifies mathematics. No Lean project is compiled, no proof is
9
+ checked, and generated QA pairs are generator output rather than research
10
+ mathematics.
11
+ """
12
+ from __future__ import annotations
13
+
14
+ import argparse
15
+ import gzip
16
+ import hashlib
17
+ import json
18
+ import math
19
+ import re
20
+ import shutil
21
+ import subprocess
22
+ from collections import Counter, deque
23
+ from pathlib import Path
24
+
25
+ COMPONENT = 'google-math'
26
+ CORE = ('id', 'text', 'group', 'license', 'paper_id', 'split', 'record_type', 'source_url')
27
+ SPLITS = ('train', 'dev', 'test')
28
+ SEQ_LEN = 2048
29
+ SHARD_ROWS = 256
30
+ IGNORED_LOCAL = {'.DS_Store', 'manifest.json', 'validation.json'}
31
+
32
+ # ---------------------------------------------------------------------------
33
+ # Source registry. Every included source is a Google or Google DeepMind
34
+ # publication that ships math text, and every entry pins a fixed revision.
35
+ # ---------------------------------------------------------------------------
36
+ SOURCES = {
37
+ 'alphaproof-nexus': {
38
+ 'name': 'AlphaProof Nexus results',
39
+ 'repository': 'https://github.com/google-deepmind/alphaproof-nexus-results',
40
+ 'commit': '0647711a71183c1ea492ad60860776617ce1ea88',
41
+ 'license': 'Apache-2.0',
42
+ 'creators': ['Google DeepMind'],
43
+ 'paper': 'Lean math proofs generated by AlphaProof Nexus and accompanying natural language prose proofs',
44
+ 'checkout': 'alphaproof-nexus-results',
45
+ 'dir': 'alphaproof-nexus-results',
46
+ },
47
+ 'formal-conjectures': {
48
+ 'name': 'Formal Conjectures',
49
+ 'repository': 'https://github.com/google-deepmind/formal-conjectures',
50
+ 'commit': '6751c0f043985ac6647db28a45a63afdf21acffe',
51
+ 'license': 'Apache-2.0',
52
+ 'creators': ['The Formal Conjectures Authors', 'Google LLC'],
53
+ 'paper': 'Formalised statements of mathematical conjectures in Lean, arXiv:2605.13171',
54
+ 'checkout': 'formal-conjectures',
55
+ 'dir': 'formal-conjectures',
56
+ },
57
+ 'formal-imo': {
58
+ 'name': 'formal-imo benchmark',
59
+ 'repository': 'https://github.com/google-deepmind/formal-imo',
60
+ 'commit': 'a85d84b97c8e88a352c136b1e29c355139609bea',
61
+ 'license': 'Apache-2.0',
62
+ 'creators': ['Google LLC'],
63
+ 'paper': 'Lean 4 formalizations of International Mathematical Olympiad problem statements',
64
+ 'checkout': 'formal-imo',
65
+ 'dir': 'formal-imo',
66
+ },
67
+ 'formal-putnam-like': {
68
+ 'name': 'Putnam-like problems in Lean',
69
+ 'repository': 'https://github.com/google-deepmind/formal-putnam-like',
70
+ 'commit': '636c2d46b3a7d7e77a9f58209b0c17816f8d172b',
71
+ 'license': 'Apache-2.0',
72
+ 'creators': ['Google DeepMind'],
73
+ 'paper': 'Lean 4 formalizations of Putnam-like problems (eval_hub putnam_like companion)',
74
+ 'checkout': 'formal-putnam-like',
75
+ 'dir': 'formal-putnam-like',
76
+ },
77
+ 'cryptids': {
78
+ 'name': 'Cryptids: conjecture-checking and Busy Beaver Turing machines',
79
+ 'repository': 'https://github.com/google-deepmind/cryptids',
80
+ 'commit': '3647718b5df22e1a61462dec0376bd204526285a',
81
+ 'commit_source': 'the checkout as of this build; recorded because the repository moves',
82
+ 'license': 'Apache-2.0',
83
+ 'creators': ['Google DeepMind'],
84
+ 'paper': 'Certified Turing-machine reductions and Busy Beaver proofs, arXiv:2610.22603',
85
+ 'checkout': 'cryptids',
86
+ 'dir': 'cryptids',
87
+ },
88
+ 'mathematics-dataset': {
89
+ 'name': 'Mathematics Dataset generator',
90
+ 'repository': 'https://github.com/google-deepmind/mathematics_dataset',
91
+ 'commit': '427f45075f84b8b9774950196ad63867ca20ffb3',
92
+ 'license': 'Apache-2.0',
93
+ 'creators': ['Google DeepMind'],
94
+ 'paper': 'Analysing Mathematical Reasoning Abilities of Neural Models (Saxton, Grefenstette, Hill, Kohli), ICLR 2018',
95
+ 'checkout': 'mathematics_dataset',
96
+ 'generated': 'mathematics_dataset_sample.jsonl',
97
+ },
98
+ 'alphageometry': {
99
+ 'name': 'AlphaGeometry geometry problem artifacts',
100
+ 'repository': 'https://github.com/google-deepmind/alphageometry',
101
+ 'commit': '6777cb586cbb46beed28db12dc72c69770b68337',
102
+ 'license': 'Apache-2.0',
103
+ 'creators': ['Google DeepMind'],
104
+ 'paper': 'Solving Olympiad Geometry without Human Demonstrations, Nature 2024',
105
+ 'checkout': 'alphageometry',
106
+ 'dir': 'alphageometry',
107
+ },
108
+ 'mathematics-conjectures': {
109
+ 'name': 'Mathematics conjectures companion notebooks',
110
+ 'repository': 'https://github.com/google-deepmind/mathematics_conjectures',
111
+ 'commit': '8302f29101019434e5a0a25807a4e42cd9d96414',
112
+ 'license': 'Apache-2.0',
113
+ 'creators': ['Google DeepMind'],
114
+ 'paper': 'Advancing mathematics by guiding human intuition with AI, Nature 2021',
115
+ 'checkout': 'mathematics_conjectures',
116
+ 'dir': 'mathematics_conjectures',
117
+ },
118
+ }
119
+
120
+ # Candidates found by the widened scan and deliberately not included.
121
+ EXCLUDED_CANDIDATES = [
122
+ {'repository': 'google-deepmind/eval_hub', 'area': 'eval_hub/putnam_like, eval_hub/live_math',
123
+ 'reason': 'Every problem directory carries a CANARY_ID benchmark-contamination canary '
124
+ '(96 putnam_like problems across Sets 1-8, 250 live_math problems across regular/difficult/multimodal). '
125
+ 'live_math is an actively maintained evaluation benchmark whose multimodal problems depend on images '
126
+ 'that are not part of this text corpus. Copying them into a pretraining corpus would poison a public '
127
+ 'benchmark, so the whole repository is excluded despite being Google-published and Apache-2.0.'},
128
+ {'repository': 'deepmind/aqua_rat', 'area': 'Hugging Face dataset',
129
+ 'reason': 'ACC-AQUA-RAT accompanies "Program Synthesis with Large Language Models", a code-generation paper, '
130
+ 'not a mathematics paper. The 97,467 rows are grade-school arithmetic word problems, so it fails the '
131
+ 'mathematics-paper test even though the license is Apache-2.0.'},
132
+ {'repository': 'deepmind/math_dataset', 'area': 'Hugging Face dataset, 40.0 GB',
133
+ 'reason': 'A pre-generated mirror of the same mathematics_dataset paper already represented here. Included '
134
+ 'through the official generator at a pinned commit instead, which keeps provenance exact and avoids '
135
+ 'pulling 40 GB over the wire.'},
136
+ {'repository': 'google-research/math-opt-benchmark',
137
+ 'area': 'C++ Bazel workspace',
138
+ 'reason': 'Ships no math text: only C++ sources plus download_data.sh, which fetches external instances at build '
139
+ 'time. Code-only utility with no shipped statements, proofs or generator.'},
140
+ {'repository': 'google/mathsteps', 'area': 'JavaScript library',
141
+ 'reason': 'A step-by-step equation solver library with 76 lib and 55 test files and no shipped math corpus. '
142
+ 'It is an equation-solving utility, not an artifact of a mathematics paper.'},
143
+ {'repository': 'google-deepmind/cartesian-frames', 'area': 'HOL4 .sml proof scripts',
144
+ 'reason': 'Formalises the Cartesian Frames sequence from the Alignment Forum, which is a philosophy-of-agency '
145
+ 'blog series rather than a mathematics paper, and it is HOL4 SML proof script rather than math text.'},
146
+ {'repository': 'google-deepmind/miniF2F', 'area': 'Lean 4 fork',
147
+ 'reason': 'A fork of openai/miniF2F. Only the Lean 4 adaptation is Google work; the corpus itself is not a '
148
+ 'Google publication, so including it would import a non-Google mirror.'},
149
+ {'repository': 'google-deepmind/alphageometry2', 'area': 'DDAR solver, 42 KB',
150
+ 'reason': 'The 26 olympiad problems live inside Python dict literals in test.py with point coordinates, so they '
151
+ 'are code text rather than a math dataset. alphageometry (AG1) already ships 261 olympiad geometry '
152
+ 'problems in the AG language verbatim, so AG2 would add solver code and no new math.'},
153
+ {'repository': 'google-deepmind/alphatensor', 'area': 'Python/NumPy',
154
+ 'reason': '24 entries of matrix-multiplication search code and README files. No math statements, proofs or '
155
+ 'text-bearing artifacts to project.'},
156
+ {'repository': 'google/feabench', 'area': 'Multiphysics benchmark',
157
+ 'reason': 'NeurIPS 2024 MATH-AI workshop benchmark for multiphysics reasoning, i.e. physics rather than '
158
+ 'mathematics, and its data ships separately.'},
159
+ {'repository': 'google/phasecoder', 'area': 'Audio research code',
160
+ 'reason': 'Microphone geometry for spatial audio. The word geometry refers to microphone arrays, not mathematics.'},
161
+ {'repository': 'google-research/pointdit', 'area': 'Diffusion model code',
162
+ 'reason': 'Monocular geometry estimation for images. Not a mathematics paper.'},
163
+ {'repository': 'google/tree-math, google/mathfu, google/vector_math.dart, google-research/opengrm',
164
+ 'area': 'Software libraries',
165
+ 'reason': 'Math-named software libraries (JAX pytree ops, a C++ game math library, a Dart vector package, n-gram '
166
+ 'libraries). Code-only utilities with no math text and no mathematics paper.'},
167
+ {'repository': 'google/research/mathlists, google-deepmind/isabelle_llm',
168
+ 'area': 'Probed repositories',
169
+ 'reason': 'Neither resolves: no repository exists at these paths, so there is nothing to include.'},
170
+ ]
171
+
172
+ PDF_ARTIFACT_RANGES = (
173
+ ('c0_c1_control', 0x00, 0x1F),
174
+ ('c1_control', 0x7F, 0x9F),
175
+ ('replacement_char', 0xFFFD, 0xFFFD),
176
+ ('private_use', 0xE000, 0xF8FF),
177
+ ('ligature_block', 0xFB00, 0xFB06),
178
+ ('combining_mark', 0x0300, 0x036F),
179
+ )
180
+
181
+
182
+ def canonical(value):
183
+ return json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(',', ':'))
184
+
185
+
186
+ def sha(path):
187
+ h = hashlib.sha256()
188
+ with Path(path).open('rb') as f:
189
+ for chunk in iter(lambda: f.read(1024 * 1024), b''):
190
+ h.update(chunk)
191
+ return h.hexdigest()
192
+
193
+
194
+ def write_json(path, obj):
195
+ Path(path).parent.mkdir(parents=True, exist_ok=True)
196
+ Path(path).write_text(json.dumps(obj, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
197
+
198
+
199
+ def write_jsonl(path, rows):
200
+ path = Path(path)
201
+ path.parent.mkdir(parents=True, exist_ok=True)
202
+ if path.suffix == '.gz':
203
+ with path.open('wb') as raw, gzip.GzipFile(filename='', mode='wb', fileobj=raw, mtime=0) as gz:
204
+ for row in rows:
205
+ gz.write((canonical(row) + '\n').encode('utf-8'))
206
+ else:
207
+ with path.open('w', encoding='utf-8') as f:
208
+ for row in rows:
209
+ f.write(canonical(row) + '\n')
210
+
211
+
212
+ def write_parquet(path, rows):
213
+ import pyarrow as pa
214
+ import pyarrow.parquet as pq
215
+ columns = {key: [str(row.get(key) or '') for row in rows] for key in CORE}
216
+ columns['metadata_json'] = [canonical({k: v for k, v in row.items() if k not in CORE}) for row in rows]
217
+ schema = pa.schema([(key, pa.string()) for key in (*CORE, 'metadata_json')])
218
+ pq.write_table(pa.Table.from_pydict(columns, schema=schema), path, compression='zstd', row_group_size=1024)
219
+
220
+
221
+ def split_groups(groups):
222
+ """Identical policy to components/openai-math/tools/finalize_openai_math.py."""
223
+ ordered = sorted(groups, key=lambda g: (hashlib.sha256(g.encode()).hexdigest(), g))
224
+ if len(ordered) < 3:
225
+ raise ValueError('Need at least 3 groups for nonempty leakage-safe splits')
226
+ holdout = max(1, math.ceil(len(ordered) * .01))
227
+ if 2 * holdout >= len(ordered):
228
+ holdout = 1
229
+ return {g: ('dev' if i < holdout else 'test' if i < 2 * holdout else 'train') for i, g in enumerate(ordered)}
230
+
231
+
232
+ def pack_split(root, split, records, sp, seq_len=SEQ_LEN, shard_rows=SHARD_ROWS):
233
+ """Copied verbatim in behaviour from components/openai-math/tools/finalize_openai_math.py."""
234
+ import numpy as np
235
+ packed = root / 'packed'
236
+ packed.mkdir(exist_ok=True)
237
+ shard_entries, pending_rows, pending_labels, pending_attrs = [], [], [], []
238
+ buffer, docs, stream_start, total_tokens, label_count, row_number = [], deque(), 0, 0, 0, 0
239
+ attr_file = (packed / f'{split}-rows.jsonl').open('w', encoding='utf-8')
240
+ token_file = (packed / f'{split}-records.jsonl').open('w', encoding='utf-8')
241
+
242
+ def flush():
243
+ nonlocal pending_rows, pending_labels, pending_attrs
244
+ if not pending_rows:
245
+ return
246
+ shard = len(shard_entries)
247
+ fields = {}
248
+ arrays = {'input_ids': np.asarray(pending_rows, dtype=np.int32),
249
+ 'labels': np.asarray(pending_labels, dtype=np.int32),
250
+ 'role_ids': np.zeros((len(pending_rows), seq_len), dtype=np.int32)}
251
+ for name, array in arrays.items():
252
+ path = packed / f'{split}-{shard:06d}-{name}.npy'
253
+ np.save(path, array, allow_pickle=False)
254
+ fields[name] = {'path': path.name, 'sha256': sha(path), 'bytes': path.stat().st_size,
255
+ 'dtype': 'int32', 'shape': list(array.shape)}
256
+ shard_entries.append({'count': len(pending_rows), 'fields': fields})
257
+ for index, attr in enumerate(pending_attrs):
258
+ attr.update(shard=shard, shard_row=index)
259
+ attr_file.write(canonical(attr) + '\n')
260
+ pending_rows, pending_labels, pending_attrs = [], [], []
261
+
262
+ def emit():
263
+ nonlocal buffer, stream_start, label_count, row_number
264
+ n = min(seq_len, len(buffer))
265
+ values = buffer[:n]
266
+ targets = buffer[1:min(n + 1, len(buffer))]
267
+ if len(targets) < n:
268
+ targets.append(-100)
269
+ targets = [-100 if t == 1 else t for t in targets]
270
+ label_count += sum(t != -100 for t in targets)
271
+ pending_rows.append(values + [0] * (seq_len - n))
272
+ pending_labels.append(targets + [-100] * (seq_len - n))
273
+ stop = stream_start + n
274
+ spans = []
275
+ for doc in docs:
276
+ lo, hi = max(stream_start, doc['start']), min(stop, doc['end'])
277
+ if lo < hi:
278
+ spans.append({'record_id': doc['id'], 'paper_id': doc['paper_id'],
279
+ 'row_start': lo - stream_start, 'row_end': hi - stream_start,
280
+ 'record_start': lo - doc['start'], 'record_end': hi - doc['start']})
281
+ pending_attrs.append({'row': row_number, 'stream_start': stream_start, 'length': n, 'segments': spans})
282
+ row_number += 1
283
+ stream_start = stop
284
+ del buffer[:n]
285
+ while docs and docs[0]['end'] <= stream_start:
286
+ docs.popleft()
287
+ if len(pending_rows) == shard_rows:
288
+ flush()
289
+
290
+ for record in records:
291
+ tokens = [1] + sp.encode(record['text'], out_type=int) + [2]
292
+ array = np.asarray(tokens, dtype='<i4')
293
+ token_file.write(canonical({'id': record['id'], 'paper_id': record['paper_id'], 'tokens': len(tokens),
294
+ 'token_sha256': hashlib.sha256(array.tobytes()).hexdigest()}) + '\n')
295
+ docs.append({'id': record['id'], 'paper_id': record['paper_id'], 'start': total_tokens,
296
+ 'end': total_tokens + len(tokens)})
297
+ total_tokens += len(tokens)
298
+ buffer.extend(tokens)
299
+ while len(buffer) >= seq_len + 1:
300
+ emit()
301
+ if buffer:
302
+ emit()
303
+ flush()
304
+ attr_file.close()
305
+ token_file.close()
306
+ return shard_entries, {'rows': row_number, 'tokens': total_tokens, 'loss_bearing_labels': label_count}
307
+
308
+
309
+ def write_manifest(root):
310
+ entries = []
311
+ for path in sorted(root.rglob('*')):
312
+ relative = path.relative_to(root).as_posix()
313
+ if path.is_file() and relative not in IGNORED_LOCAL and '.tmp' not in path.name:
314
+ entries.append({'path': relative, 'bytes': path.stat().st_size, 'sha256': sha(path)})
315
+ write_json(root / 'manifest.json', {'format': 'cayley-contribution-file-manifest-v1',
316
+ 'excludes': sorted(IGNORED_LOCAL), 'files': entries})
317
+
318
+
319
+ def extract_pdf(path):
320
+ import pypdf
321
+ reader = pypdf.PdfReader(str(path))
322
+ pages = [page.extract_text() or '' for page in reader.pages]
323
+ raw = '\n'.join(pages)
324
+ control = sum(1 for c in raw if (ord(c) < 32 and c not in '\n\t') or 0x7F <= ord(c) <= 0x9F)
325
+ text = ''.join(' ' if (ord(c) < 32 and c not in '\n\t') or 0x7F <= ord(c) <= 0x9F else c for c in raw)
326
+ text = '\n'.join(line.rstrip() for line in text.split('\n'))
327
+ text = re.sub(r'\n{3,}', '\n\n', text).strip() + '\n'
328
+ non_ascii = sum(1 for c in text if ord(c) > 127)
329
+ artifacts = Counter()
330
+ for name, lo, hi in PDF_ARTIFACT_RANGES:
331
+ artifacts[name] = sum(1 for c in text if lo <= ord(c) <= hi)
332
+ return {
333
+ 'text': text,
334
+ 'pages': len(pages),
335
+ 'raw_characters': len(raw),
336
+ 'characters': len(text),
337
+ 'page_characters': [len(page) for page in pages],
338
+ 'non_ascii_characters': non_ascii,
339
+ 'non_ascii_ratio': round(non_ascii / len(text), 6) if text else 0.0,
340
+ 'control_characters_replaced': control,
341
+ 'artifact_characters': {k: v for k, v in artifacts.items() if v},
342
+ }
343
+
344
+
345
+ # ---------------------------------------------------------------------------
346
+ # Per-source record builders.
347
+ # ---------------------------------------------------------------------------
348
+ def base_record(source_id, group, paper_id, record_type, source_url, path, digest, text, license_id):
349
+ return {
350
+ 'id': None,
351
+ 'text': text,
352
+ 'group': group,
353
+ 'license': license_id,
354
+ 'paper_id': paper_id,
355
+ 'split': None,
356
+ 'record_type': record_type,
357
+ 'source_url': source_url,
358
+ 'source_id': source_id,
359
+ 'source_path': path,
360
+ 'source_sha256': digest,
361
+ 'text_sha256': hashlib.sha256(text.encode('utf-8')).hexdigest(),
362
+ 'creators': SOURCES[source_id]['creators'],
363
+ 'verification_status': 'unverified_source_claim',
364
+ 'lean_verified': False,
365
+ 'mathematical_verification': 'not_performed',
366
+ }
367
+
368
+
369
+ def build_alphaproof_nexus(root, source_id, checkout, files, seen):
370
+ """The staged AlphaProof Nexus projection, reused unchanged (83 records)."""
371
+ source = SOURCES[source_id]
372
+ commit, repo = source['commit'], source['repository']
373
+ records = []
374
+ for rel in files:
375
+ path = checkout / rel
376
+ digest = sha(path)
377
+ stem = Path(rel).stem
378
+ if rel.endswith('.lean'):
379
+ text = path.read_text(encoding='utf-8')
380
+ group = lean_group(rel)
381
+ record_type, origin = 'formal_lean_proof', 'source_verbatim'
382
+ quality, transform = 'verbatim_source_text', 'identity_verbatim_bytes'
383
+ metrics = {'source_characters': len(text)}
384
+ elif rel.endswith('.pdf'):
385
+ extracted = extract_pdf(path)
386
+ group = pdf_group(rel)
387
+ text = extracted.pop('text')
388
+ record_type, origin = 'natural_language_proof', 'source_pdf_extraction'
389
+ quality = 'lossy_pdf_text_layer'
390
+ transform = 'pypdf_page_text_control_chars_replaced_blank_runs_collapsed'
391
+ metrics = extracted
392
+ global EXTRACTION_AUDIT
393
+ EXTRACTION_AUDIT.append({'source_id': source_id, 'source_path': rel, **metrics,
394
+ 'extraction_quality': quality,
395
+ 'source_url': f'{repo}/blob/{commit}/{rel}'})
396
+ else:
397
+ continue
398
+ kind = 'lean' if rel.endswith('.lean') else 'nl'
399
+ upstream_group = group
400
+ group = f'{source_id}::{upstream_group}'
401
+ record = base_record(source_id, group, upstream_group, record_type,
402
+ f'{repo}/blob/{commit}/{rel}', rel, digest, text, source['license'])
403
+ record['id'] = f'{COMPONENT}:{source_id}:{upstream_group}:{kind}:{stem}'
404
+ record['upstream_group'] = upstream_group
405
+ record['content_origin'] = origin
406
+ record['transform'] = transform
407
+ record['extraction_quality'] = quality
408
+ record['extraction_metrics'] = metrics
409
+ record['upstream_category'] = rel.split('/')[1]
410
+ record['source_commit'] = commit
411
+ records.append(record)
412
+ return records
413
+
414
+
415
+ def lean_group(rel):
416
+ """Problem-stem group for an APNOutputs Lean file (unchanged from the staged build)."""
417
+ parts = Path(rel).parts
418
+ category, stem = parts[1], Path(rel).stem
419
+ if category == 'ErdosProblems':
420
+ return 'erdos_' + stem.split('_')[1].split('.')[0]
421
+ if category == 'OEIS':
422
+ tokens = stem.split('_')
423
+ ident = tokens[1] if tokens[0].lower() == 'oeis' else tokens[0]
424
+ return 'oeis_' + ident.lower()
425
+ if category == 'StacksProject':
426
+ return 'stacks_project'
427
+ if category == 'AICollaborator':
428
+ subject = parts[2]
429
+ if stem.startswith('hilbert_functions'):
430
+ return 'ai_hilbert_functions'
431
+ if stem == 'bipartite_graph_reconstruction_conjecture_2':
432
+ return 'ai_graph_reconstruction'
433
+ return 'ai_' + subject.lower()
434
+ raise ValueError(f'Unknown Lean category: {rel}')
435
+
436
+
437
+ def pdf_group(rel):
438
+ """Problem-stem group for a NaturalLanguageProofs PDF (unchanged from the staged build)."""
439
+ category, stem = Path(rel).parts[1], Path(rel).stem
440
+ if category == 'ErdosProblems':
441
+ match = re.fullmatch(r'erdos(\d+)(i{1,2})?', stem)
442
+ if not match:
443
+ raise ValueError(f'Unparsed Erdos PDF name: {rel}')
444
+ return 'erdos_' + match.group(1)
445
+ if category == 'OEIS':
446
+ match = re.fullmatch(r'oeis_(\d+)', stem.lower())
447
+ if not match:
448
+ raise ValueError(f'Unparsed OEIS PDF name: {rel}')
449
+ return 'oeis_' + match.group(1)
450
+ if category == 'AICollaborator':
451
+ if stem == 'hilbert':
452
+ return 'ai_hilbert_functions'
453
+ if stem == 'graph_reconstruction':
454
+ return 'ai_graph_reconstruction'
455
+ raise ValueError(f'Unpaired PDF name: {rel}')
456
+
457
+
458
+ def build_lean_corpus(source_id, checkout, files, record_type):
459
+ """Verbatim projection of Lean source text, one record per file."""
460
+ source = SOURCES[source_id]
461
+ commit, repo = source['commit'], source['repository']
462
+ records = []
463
+ for rel in files:
464
+ path = checkout / rel
465
+ text = path.read_text(encoding='utf-8')
466
+ if not text.strip():
467
+ continue
468
+ group = f'{source_id}::{rel[:-len(".lean")]}'
469
+ record = base_record(source_id, group, group, record_type,
470
+ f'{repo}/blob/{commit}/{rel}', rel, sha(path), text, source['license'])
471
+ record['id'] = f'{COMPONENT}:{source_id}:{rel[:-len(".lean")]}'
472
+ record['content_origin'] = 'source_verbatim'
473
+ record['transform'] = 'identity_verbatim_bytes'
474
+ record['extraction_quality'] = 'verbatim_source_text'
475
+ record['source_commit'] = commit
476
+ record['source_characters'] = len(text)
477
+ records.append(record)
478
+ return records
479
+
480
+
481
+ def build_alphageometry(source_id, checkout):
482
+ """Verbatim problem statements from the AG language text artifacts."""
483
+ source = SOURCES[source_id]
484
+ commit, repo = source['commit'], source['repository']
485
+ records = []
486
+ pairs = {'examples.txt': 'library_example', 'imo_ag_30.txt': 'olympiad_geometry_problem',
487
+ 'jgex_ag_231.txt': 'jgex_geometry_problem'}
488
+ for name, kind in pairs.items():
489
+ path = checkout / name
490
+ lines = [line.strip() for line in path.read_text(encoding='utf-8').splitlines()]
491
+ body = [line for line in lines if line and not line.startswith('#')]
492
+ if len(body) % 2:
493
+ raise ValueError(f'{name}: odd number of problem lines ({len(body)})')
494
+ for index in range(0, len(body), 2):
495
+ problem_id, statement = body[index], body[index + 1]
496
+ text = f'{problem_id}\n{statement}\n'
497
+ group = f'{source_id}::{name}::{problem_id}'
498
+ record = base_record(source_id, group, group, 'formal_geometry_problem',
499
+ f'{repo}/blob/{commit}/{name}', name, sha(path), text, source['license'])
500
+ record['id'] = f'{COMPONENT}:{source_id}:{name}:{problem_id}'
501
+ record['content_origin'] = 'source_verbatim'
502
+ record['transform'] = 'identity_verbatim_lines'
503
+ record['extraction_quality'] = 'verbatim_source_text'
504
+ record['source_commit'] = commit
505
+ record['geometry_problem_id'] = problem_id
506
+ record['geometry_problem_kind'] = kind
507
+ record['source_line'] = index + 1
508
+ records.append(record)
509
+ for name in ('defs.txt', 'rules.txt'):
510
+ path = checkout / name
511
+ text = path.read_text(encoding='utf-8')
512
+ group = f'{source_id}::{name}'
513
+ record = base_record(source_id, group, group, 'geometry_rule_table',
514
+ f'{repo}/blob/{commit}/{name}', name, sha(path), text, source['license'])
515
+ record['id'] = f'{COMPONENT}:{source_id}:{name}'
516
+ record['content_origin'] = 'source_verbatim'
517
+ record['transform'] = 'identity_verbatim_bytes'
518
+ record['extraction_quality'] = 'verbatim_source_text'
519
+ record['source_commit'] = commit
520
+ records.append(record)
521
+ return records
522
+
523
+
524
+ def build_mathematics_dataset(source_id, generated, params):
525
+ """Seeded projection of the official generator's curriculum."""
526
+ source = SOURCES[source_id]
527
+ commit, repo = source['commit'], source['repository']
528
+ records = []
529
+ for row in generated:
530
+ group = f'{source_id}::{row["regime"]}::{row["module"]}'
531
+ text = row['question'] + '\n' + row['answer'] + '\n'
532
+ record = base_record(source_id, group, row['module'], 'generated_qa_pair',
533
+ f'{repo}/tree/{commit}', f'generated/{row["regime"]}/{row["module"]}',
534
+ None, text, source['license'])
535
+ record['id'] = f'{COMPONENT}:{source_id}:{row["regime"]}:{row["module"]}:{row["index"]:05d}'
536
+ record['content_origin'] = 'generator_output'
537
+ record['transform'] = 'identity_generator_question_and_answer_lines'
538
+ record['extraction_quality'] = 'verbatim_generator_output'
539
+ record['source_commit'] = commit
540
+ record['generator_regime'] = row['regime']
541
+ record['generator_module'] = row['module']
542
+ record['generator_index'] = row['index']
543
+ record['generator_seed'] = params['seed']
544
+ record['question'] = row['question']
545
+ record['answer'] = row['answer']
546
+ records.append(record)
547
+ return records
548
+
549
+
550
+ def build_notebooks(source_id, checkout):
551
+ """Notebook cell sources joined by newlines, verbatim."""
552
+ source = SOURCES[source_id]
553
+ commit, repo = source['commit'], source['repository']
554
+ records = []
555
+ for path in sorted(checkout.glob('*.ipynb')):
556
+ notebook = json.loads(path.read_text(encoding='utf-8'))
557
+ cells = [(cell['cell_type'], ''.join(cell['source'])) for cell in notebook['cells']]
558
+ text = '\n'.join(content for _, content in cells)
559
+ group = f'{source_id}::{path.name}'
560
+ record = base_record(source_id, group, path.stem, 'research_notebook',
561
+ f'{repo}/blob/{commit}/{path.name}', path.name, sha(path), text, source['license'])
562
+ record['id'] = f'{COMPONENT}:{source_id}:{path.stem}'
563
+ record['content_origin'] = 'source_notebook_projection'
564
+ record['transform'] = 'notebook_cell_sources_joined_by_newline'
565
+ record['extraction_quality'] = 'verbatim_cell_text'
566
+ record['source_commit'] = commit
567
+ record['cell_counts'] = dict(Counter(kind for kind, _ in cells))
568
+ EXTRACTION_AUDIT.append({'source_id': source_id, 'source_path': path.name,
569
+ 'cells': len(cells), 'characters': len(text),
570
+ 'markdown_cells': sum(1 for k, _ in cells if k == 'markdown'),
571
+ 'code_cells': sum(1 for k, _ in cells if k == 'code'),
572
+ 'extraction_quality': 'verbatim_cell_text',
573
+ 'source_url': f'{repo}/blob/{commit}/{path.name}'})
574
+ records.append(record)
575
+ return records
576
+
577
+
578
+ EXTRACTION_AUDIT = []
579
+
580
+
581
+ def main():
582
+ parser = argparse.ArgumentParser(description=__doc__)
583
+ parser.add_argument('--component', type=Path, default=Path(__file__).resolve().parents[1])
584
+ parser.add_argument('--source-root', type=Path, required=False,
585
+ help='Directory holding the pinned checkouts of every included repository')
586
+ parser.add_argument('--generated-root', type=Path, required=False,
587
+ help='Directory holding the pinned mathematics_dataset generator sample')
588
+ parser.add_argument('--tokenizer', type=Path,
589
+ default=Path('/Users/zxy/Cayley-Math/tokenizer/tokenizer.model'))
590
+ parser.add_argument('--pack', action='store_true')
591
+ parser.add_argument('--manifest-only', action='store_true',
592
+ help='Rewrite manifest.json from the current directory contents and exit')
593
+ args = parser.parse_args()
594
+ root = args.component.resolve()
595
+ if args.manifest_only:
596
+ stats = json.loads((root / 'statistics.json').read_text(encoding='utf-8'))
597
+ write_readme(root, stats, stats['per_source'])
598
+ write_manifest(root)
599
+ print(json.dumps({'manifest': 'rewritten',
600
+ 'files': len(json.loads((root / 'manifest.json').read_text())['files'])}, indent=2))
601
+ return
602
+ src = args.source_root.resolve()
603
+
604
+ # --- pin every checkout and refuse to build against a drifting revision ---
605
+ for source_id, source in SOURCES.items():
606
+ checkout = src / source['checkout']
607
+ if not checkout.is_dir():
608
+ raise ValueError(f'Missing pinned checkout for {source_id}: {checkout}')
609
+ if source.get('generated') or not source.get('commit'):
610
+ continue
611
+ live = subprocess.check_output(['git', '-C', str(checkout), 'rev-parse', 'HEAD'], text=True).strip()
612
+ if live != source['commit']:
613
+ raise ValueError(f'{source_id} checkout {live} differs from pinned {source["commit"]}')
614
+
615
+ generated_path = args.generated_root.resolve() / SOURCES['mathematics-dataset']['generated']
616
+ generated = [json.loads(line) for line in generated_path.read_text(encoding='utf-8').splitlines() if line]
617
+ gen_params = json.loads(Path(str(generated_path) + '.params.json').read_text(encoding='utf-8'))
618
+
619
+ all_files = []
620
+ records = []
621
+
622
+ # 1. AlphaProof Nexus (verbatim staged projection, 83 records)
623
+ apn = src / SOURCES['alphaproof-nexus']['checkout']
624
+ apn_files = sorted(p.relative_to(apn).as_posix() for p in apn.rglob('*')
625
+ if p.is_file() and '.git' not in p.parts and p.suffix in ('.lean', '.pdf'))
626
+ records += build_alphaproof_nexus(root, 'alphaproof-nexus', apn, apn_files, None)
627
+ all_files += [{'source_id': 'alphaproof-nexus', 'path': rel} for rel in apn_files]
628
+
629
+ # 2. Formal Conjectures library (statements only; tests and build scripts excluded)
630
+ fc = src / SOURCES['formal-conjectures']['checkout']
631
+ fc_files = sorted(p.relative_to(fc).as_posix() for p in fc.rglob('*.lean')
632
+ if '.git' not in p.parts
633
+ and p.relative_to(fc).parts[0] in ('FormalConjectures', 'FormalConjecturesForMathlib',
634
+ 'FormalConjecturesUtil'))
635
+ records += build_lean_corpus('formal-conjectures', fc, fc_files, 'formal_lean_statement')
636
+ all_files += [{'source_id': 'formal-conjectures', 'path': rel} for rel in fc_files]
637
+
638
+ # 3. formal-imo
639
+ imo = src / SOURCES['formal-imo']['checkout']
640
+ imo_files = sorted(p.relative_to(imo).as_posix() for p in imo.rglob('*.lean')
641
+ if '.git' not in p.parts and p.relative_to(imo).parts[0] == 'Imo')
642
+ records += build_lean_corpus('formal-imo', imo, imo_files, 'formal_lean_statement')
643
+ all_files += [{'source_id': 'formal-imo', 'path': rel} for rel in imo_files]
644
+
645
+ # 4. formal-putnam-like
646
+ pl = src / SOURCES['formal-putnam-like']['checkout']
647
+ pl_files = sorted(p.relative_to(pl).as_posix() for p in pl.rglob('*.lean')
648
+ if '.git' not in p.parts and p.relative_to(pl).parts[0] == 'PutnamLike')
649
+ records += build_lean_corpus('formal-putnam-like', pl, pl_files, 'formal_lean_statement')
650
+ all_files += [{'source_id': 'formal-putnam-like', 'path': rel} for rel in pl_files]
651
+
652
+ # 5. Cryptids (whole Lean library, including the Busy Beaver holdout proofs)
653
+ cr = src / SOURCES['cryptids']['checkout']
654
+ cr_files = sorted(p.relative_to(cr).as_posix() for p in cr.rglob('*.lean') if '.git' not in p.parts)
655
+ records += build_lean_corpus('cryptids', cr, cr_files, 'formal_lean_proof')
656
+ all_files += [{'source_id': 'cryptids', 'path': rel} for rel in cr_files]
657
+
658
+ # 6. Mathematics Dataset generator sample
659
+ records += build_mathematics_dataset('mathematics-dataset', generated, gen_params)
660
+ all_files.append({'source_id': 'mathematics-dataset', 'path': f'generated/{generated_path.name}',
661
+ 'generated_bytes': generated_path.stat().st_size,
662
+ 'generated_sha256': sha(generated_path)})
663
+
664
+ # 7. AlphaGeometry text artifacts
665
+ ag = src / SOURCES['alphageometry']['checkout']
666
+ ag_files = [name for name in ('examples.txt', 'imo_ag_30.txt', 'jgex_ag_231.txt', 'defs.txt', 'rules.txt')]
667
+ records += build_alphageometry('alphageometry', ag)
668
+ all_files += [{'source_id': 'alphageometry', 'path': rel} for rel in ag_files]
669
+
670
+ # 8. Mathematics conjectures notebooks
671
+ mc = src / SOURCES['mathematics-conjectures']['checkout']
672
+ mc_files = sorted(p.name for p in mc.glob('*.ipynb'))
673
+ records += build_notebooks('mathematics-conjectures', mc)
674
+ all_files += [{'source_id': 'mathematics-conjectures', 'path': rel} for rel in mc_files]
675
+
676
+ # --- exact text dedup ---
677
+ duplicates, seen = [], {}
678
+ kept = []
679
+ for record in records:
680
+ digest = record['text_sha256']
681
+ if digest in seen:
682
+ duplicates.append({'excluded_id': record['id'], 'kept_id': seen[digest],
683
+ 'source_id': record['source_id'], 'group': record['group'],
684
+ 'text_sha256': digest})
685
+ continue
686
+ seen[digest] = record['id']
687
+ kept.append(record)
688
+ records = kept
689
+ ids = [r['id'] for r in records]
690
+ if len(set(ids)) != len(ids):
691
+ raise ValueError('Duplicate record IDs')
692
+
693
+ # --- splits, assigned per group ---
694
+ groups = sorted({r['group'] for r in records})
695
+ split_of = split_groups(groups)
696
+ empty = [s for s in SPLITS if s not in split_of.values()]
697
+ if empty:
698
+ raise ValueError(f'Empty splits: {empty}')
699
+ for record in records:
700
+ record['split'] = split_of[record['group']]
701
+ # a group key is namespaced by source, so paper_id stays the upstream work unit
702
+ records.sort(key=lambda r: (r['source_id'], r['group'], r['record_type'], r['source_path'], r['id']))
703
+
704
+ split_records = {}
705
+ for split in SPLITS:
706
+ selected = [r for r in records if r['split'] == split]
707
+ if not selected:
708
+ raise ValueError(f'Empty {split} split')
709
+ write_jsonl(root / f'data/{split}.jsonl.gz', selected)
710
+ write_parquet(root / f'data/{split}.parquet', selected)
711
+ split_records[split] = selected
712
+
713
+ # --- licenses, one per source subdirectory plus an index ---
714
+ for source_id, source in SOURCES.items():
715
+ checkout = src / source['checkout']
716
+ destination = root / 'licenses' / source_id
717
+ destination.mkdir(parents=True, exist_ok=True)
718
+ license_file = checkout / 'LICENSE'
719
+ if license_file.is_file():
720
+ shutil.copy2(license_file, destination / 'LICENSE')
721
+ else:
722
+ raise ValueError(f'{source_id}: upstream LICENSE missing')
723
+ root_license = (root / 'licenses' / 'alphaproof-nexus' / 'LICENSE').read_text(encoding='utf-8')
724
+ (root / 'LICENSE').write_text(root_license, encoding='utf-8')
725
+
726
+ # --- audit ---
727
+ audit = root / 'audit'
728
+ audit.mkdir(parents=True, exist_ok=True)
729
+ source_file_rows = []
730
+ for item in all_files:
731
+ source = SOURCES[item['source_id']]
732
+ if 'generated_sha256' in item:
733
+ source_file_rows.append({
734
+ 'source_id': item['source_id'], 'path': item['path'],
735
+ 'sha256': item['generated_sha256'], 'bytes': item['generated_bytes'],
736
+ 'source_url': f'{source["repository"]}/tree/{source["commit"]}',
737
+ 'kind': 'generated_by_pinned_generator',
738
+ })
739
+ continue
740
+ path = (src / source['checkout'] / item['path'])
741
+ source_file_rows.append({
742
+ 'source_id': item['source_id'], 'path': item['path'], 'sha256': sha(path),
743
+ 'bytes': path.stat().st_size,
744
+ 'source_url': f'{source["repository"]}/blob/{source["commit"]}/{item["path"]}',
745
+ 'kind': 'upstream_blob',
746
+ })
747
+ write_jsonl(audit / 'source-files.jsonl', source_file_rows)
748
+ write_jsonl(audit / 'extraction.jsonl', EXTRACTION_AUDIT)
749
+ write_jsonl(audit / 'excluded-duplicates.jsonl', duplicates)
750
+ write_json(audit / 'group-splits.json', {
751
+ 'format': 'cayley-group-splits-v1',
752
+ 'policy': 'sorted SHA256(group), first ceil(1%) groups to dev, next ceil(1%) to test, rest train',
753
+ 'group_namespacing': 'every group key starts with its source_id so two upstreams never share a group',
754
+ 'splits': split_of,
755
+ 'groups_by_source': dict(Counter(g.split('::')[0] for g in groups)),
756
+ 'records_by_source': dict(Counter(r['source_id'] for r in records)),
757
+ })
758
+ write_json(audit / 'mathematics-dataset-generation.json', {
759
+ 'source_id': 'mathematics-dataset',
760
+ 'generator_repository': SOURCES['mathematics-dataset']['repository'],
761
+ 'generator_commit': SOURCES['mathematics-dataset']['commit'],
762
+ 'driver_script': 'tools/generate_mathematics_dataset_sample.py',
763
+ 'projection': 'sampled projection of the paper curriculum, not the paper-complete dataset',
764
+ 'measured': {
765
+ 'records': len(generated),
766
+ 'records_by_regime': dict(Counter(r['regime'] for r in generated)),
767
+ 'modules': len({r['module'] for r in generated}),
768
+ 'characters': sum(len(r['question']) + len(r['answer']) + 2 for r in generated),
769
+ 'records_retained_after_exact_dedup': sum(1 for r in records if r['source_id'] == 'mathematics-dataset'),
770
+ 'records_removed_as_exact_duplicates': sum(1 for d in duplicates
771
+ if d['source_id'] == 'mathematics-dataset'),
772
+ 'modules_stopped_early': [k for k, v in gen_params.get('module_yield', {}).items() if v['stopped_early']],
773
+ },
774
+ 'parameters': gen_params,
775
+ })
776
+ write_json(audit / 'source-scan.json', {
777
+ 'format': 'cayley-source-scan-v1',
778
+ 'queries': [
779
+ 'GitHub search repositories: org:google-deepmind + math|proof|theorem|formal|lean|geometry|olympiad|conjecture',
780
+ 'GitHub search repositories: org:google + math|theorem|lean|proof',
781
+ 'GitHub search repositories: org:google-research + math|proof|theorem|lean|geometry|olympiad|formal',
782
+ 'Direct probes of every candidate named in the brief, plus every google-deepmind repository page 1-2 '
783
+ 'filtered on math/theorem/proof/formal/lean/geometry/olympiad/conjecture/arithmetic/algebra keywords',
784
+ 'Hugging Face dataset listing: authors google (76 datasets), google-research-datasets (37), deepmind (6)',
785
+ ],
786
+ 'included': [{'source_id': sid, **{k: s[k] for k in ('name', 'repository', 'commit', 'license',
787
+ 'paper', 'creators')}} for sid, s in SOURCES.items()],
788
+ 'excluded_candidates': EXCLUDED_CANDIDATES,
789
+ })
790
+
791
+ # --- packed arrays ---
792
+ packed_stats = {}
793
+ if args.pack:
794
+ import sentencepiece as spm
795
+ sp = spm.SentencePieceProcessor(model_file=str(args.tokenizer))
796
+ if (sp.pad_id(), sp.bos_id(), sp.eos_id(), sp.unk_id(), sp.vocab_size()) != (0, 1, 2, 3, 32768):
797
+ raise ValueError('Unexpected tokenizer special IDs or vocabulary size')
798
+ tokenizer_dir = root / 'tokenizer'
799
+ tokenizer_dir.mkdir(exist_ok=True)
800
+ shutil.copy2(args.tokenizer, tokenizer_dir / 'tokenizer.model')
801
+ packed = root / 'packed'
802
+ packed.mkdir(exist_ok=True)
803
+ for stale in packed.glob('*'):
804
+ if stale.is_file():
805
+ stale.unlink()
806
+ shards = {}
807
+ for split in SPLITS:
808
+ shards[split], packed_stats[split] = pack_split(root, split, split_records[split], sp)
809
+ write_json(packed / 'manifest.json', {
810
+ 'format': 'mmla-type2-packed-v1', 'phase': 'pretrain', 'layout': 'lm', 'seq_len': SEQ_LEN,
811
+ 'tokenizer': {'kind': 'sentencepiece', 'vocab_size': 32768, 'sha256': sha(args.tokenizer),
812
+ 'path': '../tokenizer/tokenizer.model',
813
+ 'ids': {'pad': 0, 'bos': 1, 'eos': 2, 'unk': 3, 'boundary': 4}},
814
+ 'quota_units': 'tokens', 'source': COMPONENT, 'splits': shards, 'statistics': packed_stats,
815
+ 'packing': {'stride': SEQ_LEN, 'bos_per_record': True, 'eos_per_record': True,
816
+ 'mask_target_bos': True, 'mask_final_target': True, 'loss_mask': -100,
817
+ 'row_attribution': '{split}-rows.jsonl', 'record_tokens': '{split}-records.jsonl'},
818
+ })
819
+ write_json(audit / 'packing.json', {
820
+ 'packer': 'tools/build_google_math.py:pack_split (copied from components/openai-math)',
821
+ 'tokenizer_sha256': sha(tokenizer_dir / 'tokenizer.model'),
822
+ 'tokenizer_note': 'SentencePiece decode is whitespace-normalising for runs of spaces; every packed row is '
823
+ 'verified by re-encoding.',
824
+ 'sequence_length': SEQ_LEN, 'shard_rows': SHARD_ROWS,
825
+ })
826
+
827
+ # --- statistics ---
828
+ source_stats = {}
829
+ for source_id in SOURCES:
830
+ rows = [r for r in records if r['source_id'] == source_id]
831
+ packed_rows = [r for r in rows if r['split'] == 'train']
832
+ source_stats[source_id] = {
833
+ 'records': len(rows),
834
+ 'groups': len({r['group'] for r in rows}),
835
+ 'record_types': dict(Counter(r['record_type'] for r in rows)),
836
+ 'text_characters': sum(len(r['text']) for r in rows),
837
+ 'upstream_files': sum(1 for x in source_file_rows if x['source_id'] == source_id),
838
+ 'split_records': {s: sum(1 for r in rows if r['split'] == s) for s in SPLITS},
839
+ }
840
+ stats = {
841
+ 'component': COMPONENT,
842
+ 'sources': len(SOURCES),
843
+ 'source_commits': {sid: s['commit'] for sid, s in SOURCES.items()},
844
+ 'records_before_deduplication': len(records) + len(duplicates),
845
+ 'records': len(records),
846
+ 'excluded_exact_duplicates': len(duplicates),
847
+ 'groups': len(groups),
848
+ 'record_types': dict(Counter(r['record_type'] for r in records)),
849
+ 'text_characters': sum(len(r['text']) for r in records),
850
+ 'per_source': source_stats,
851
+ 'splits': {s: {'records': len(v), 'groups': len({r['group'] for r in v}),
852
+ 'sources': len({r['source_id'] for r in v}),
853
+ 'text_characters': sum(len(r['text']) for r in v)} for s, v in split_records.items()},
854
+ 'packed': packed_stats,
855
+ 'mathematical_verification': 'not_performed',
856
+ 'lean_compilation': 'not_performed',
857
+ }
858
+ write_json(root / 'statistics.json', stats)
859
+
860
+ sources_index = {
861
+ 'component': COMPONENT,
862
+ 'format': 'cayley-sources-v1',
863
+ 'all_records_license': 'per-record',
864
+ 'license_index': {sid: {'license': s['license'],
865
+ 'license_file': f'licenses/{sid}/LICENSE',
866
+ 'creators': s['creators']} for sid, s in SOURCES.items()},
867
+ 'included': {sid: {
868
+ 'name': s['name'],
869
+ 'repository': s['repository'],
870
+ 'commit': s['commit'],
871
+ 'fixed_revision_url': f'{s["repository"]}/tree/{s["commit"]}',
872
+ 'license': s['license'],
873
+ 'license_file': f'licenses/{sid}/LICENSE',
874
+ 'creators': s['creators'],
875
+ 'paper': s['paper'],
876
+ 'records': source_stats[sid]['records'],
877
+ 'text_characters': source_stats[sid]['text_characters'],
878
+ } for sid, s in SOURCES.items()},
879
+ 'excluded_candidates': EXCLUDED_CANDIDATES,
880
+ 'changes': [
881
+ 'Verbatim text projection of Lean sources, geometry problem language, notebooks and PDFs; no rewriting',
882
+ 'Deterministically seeded sample from the official mathematics_dataset generator at a pinned commit',
883
+ 'Group keys namespaced per source so a proof and its prose companion and a generated QA family stay together',
884
+ 'Exact text deduplication and deterministic SHA256-ordered group split assignment',
885
+ 'Lossless JSONL/Parquet export; SentencePiece causal-window packing in the existing openai-math format',
886
+ ],
887
+ 'record_changes': [
888
+ 'No text was generated, summarised, translated or repaired beyond PDF control-character handling',
889
+ 'PDF text control characters were replaced and blank-line runs collapsed; original bytes stay identified by sha256',
890
+ 'No Lean compilation, proof checking or mathematical verification was performed',
891
+ ],
892
+ 'mathematical_verification': 'not_performed',
893
+ }
894
+ write_json(root / 'sources.json', sources_index)
895
+
896
+ # --- README ---
897
+ write_readme(root, stats, source_stats)
898
+ write_manifest(root)
899
+ print(json.dumps(stats, ensure_ascii=False, indent=2))
900
+
901
+
902
+ def write_readme(root, stats, source_stats):
903
+ per_source_table = '\n'.join(
904
+ f'| `{sid}` | [{SOURCES[sid]["repository"].rsplit("/", 1)[-1]}]({SOURCES[sid]["repository"]}) '
905
+ f'`{SOURCES[sid]["commit"][:12]}` | {src["records"]:,} | {src["groups"]:,} | '
906
+ f'{src["text_characters"]:,} | {", ".join(sorted(src["record_types"]))} |'
907
+ for sid, src in source_stats.items())
908
+ stats_json = json.dumps(stats, indent=2)
909
+ readme = f"""# Google math papers contribution to Cayley-Math
910
+
911
+ An independently versioned component holding source-derived text from eight Google and Google DeepMind
912
+ publications about mathematics, formal mathematics and mathematical reasoning. It follows Cayley-Math's
913
+ `components/` convention and reuses the record schema, Parquet encoding, split policy and packed-array
914
+ format of `components/openai-math/`. The frozen Phase 1 mirror is unchanged.
915
+
916
+ ## What this is, and what it is not
917
+
918
+ These are **artifacts of specific Google math papers** (Lean libraries, geometry problem files, generator
919
+ output, paper notebooks and proof PDFs), not a survey of Google's mathematics output and not a
920
+ paper-complete reproduction of any one of them. The corpus is a curated projection: a bounded, seeded
921
+ sample for the generator source, a verbatim text projection for the file sources.
922
+
923
+ Nothing here establishes mathematical truth. **No Lean project was compiled, no proof was checked, no
924
+ `lake build` or Comparator run was performed, and no mathematical verification of any kind took place.**
925
+ `record_type` labels the artifact kind, not the validity of its content. Records that contain `sorry`
926
+ placeholders remain placeholders.
927
+
928
+ The `mathematics-dataset` subset is output of the Saxton/Grefenstette/Hill/Kohli ICLR 2018 generator at
929
+ school-level difficulty. Those pairs are **not research mathematics**, and the subset is a sampled
930
+ projection of the paper's curriculum rather than the paper's dataset. Every record carries
931
+ `content_origin` so consumers can separate verbatim source text from `generator_output`.
932
+
933
+ **This corpus is mostly seeded generator output, not research mathematics.** `generated_qa_pair` records
934
+ dominate the record count, the formal subset is a small minority, and **no Lean compilation and no proof
935
+ verification was performed on any subset**, so no record here has a checked proof or a checked statement.
936
+ `statistics.json` carries the measured counts, and the component README states them in full.
937
+
938
+ ## Included sources
939
+
940
+ Each source is pinned to a commit or dataset revision, and each ships math text under Apache-2.0
941
+ software terms with CC-BY 4.0 for non-software materials where the upstream states so.
942
+
943
+ | Source | Upstream at pinned revision | Records | Groups | Text characters | Record types |
944
+ | --- | --- | ---: | ---: | ---: | --- |
945
+ {per_source_table}
946
+
947
+ `alphaproof-nexus` keeps all 83 records of the earlier standalone build unchanged in content: 71 Lean
948
+ proof files plus 12 pypdf-extracted prose proof PDFs over 51 problem groups. Only its split assignment
949
+ changed, because splits are recomputed over the merged group set of all eight sources rather than over
950
+ its own 51 groups. `formal-conjectures` is the 1,861-file `FormalConjectures/` statement library plus
951
+ its 274 helper-library files. `formal-imo` and `formal-putnam-like` are Lean 4 problem statements.
952
+ `cryptids` is the whole Lean library, including the Busy Beaver holdout proofs, whose individual files
953
+ reach 1.4 MB of mechanically generated state traces. `alphageometry` contributes 265 geometry problems
954
+ in the AlphaGeometry language: 231 JGEX problems, 30 olympiad problems and 4 library examples, plus the
955
+ `defs.txt` and `rules.txt` rule tables. The two `mathematics-conjectures` notebooks accompany the Nature
956
+ 2021 knot-theory and representation-theory paper.
957
+
958
+ ## Excluded candidates and why
959
+
960
+ The widened scan is recorded in full in `sources.json` (`excluded_candidates`) and `audit/source-scan.json`.
961
+ The short version:
962
+
963
+ - `google-deepmind/eval_hub` (putnam_like and live_math) is excluded because every problem carries a
964
+ `CANARY_ID` benchmark-contamination canary and live_math is an actively maintained benchmark whose
965
+ multimodal problems depend on images absent from this text corpus.
966
+ - `deepmind/aqua_rat` accompanies a program-synthesis paper, not a mathematics paper.
967
+ - `deepmind/math_dataset` is a 40 GB pre-generated mirror of the generator already used here.
968
+ - `google-research/math-opt-benchmark`, `google/mathsteps`, `google-deepmind/alphatensor` and every
969
+ math-named software library (`tree-math`, `mathfu`, `vector_math.dart`, `opengrm`) ship code and no
970
+ math text.
971
+ - `google-deepmind/cartesian-frames` formalises an alignment blog sequence in HOL4 SML, not a
972
+ mathematics paper.
973
+ - `google-deepmind/miniF2F` is a fork of a non-Google corpus.
974
+ - `google-deepmind/alphageometry2` embeds 26 problems in Python dict literals, and AG1 already ships
975
+ the same olympiad problems verbatim in the AlphaGeometry language.
976
+ - `google/feabench`, `google/phasecoder` and `google-research/pointdit` are physics, audio and vision
977
+ papers whose titles happen to contain mathematical words.
978
+
979
+ ## Files and schema
980
+
981
+ - `data/{{train,dev,test}}.jsonl.gz`: source records with required fields text, group, license; unique id,
982
+ paper_id, split, record_type, source_url, and `source_id` naming the upstream.
983
+ - `data/{{train,dev,test}}.parquet`: identical records. Core fields
984
+ (id,text,group,license,paper_id,split,record_type,source_url) are string columns; all remaining JSON
985
+ fields are serialised in `metadata_json`. Merge `json.loads(metadata_json)` with the core columns to
986
+ reconstruct each JSONL record.
987
+ - `sources.json`: every included source with its pinned revision, license and measured counts, plus every
988
+ excluded candidate with its reason.
989
+ - `licenses/<source_id>/LICENSE`: one upstream license per source; `LICENSE` at the component root is the
990
+ Apache-2.0 text shared by all eight.
991
+ - `audit/source-files.jsonl`: path, sha256, bytes, source_id, kind and pinned source_url for every blob and
992
+ for the generated corpus file.
993
+ - `audit/group-splits.json`: the group assignment, the policy, and per-source group counts.
994
+ - `audit/extraction.jsonl`: per-PDF and per-notebook extraction metrics and quality flags.
995
+ - `audit/mathematics-dataset-generation.json`: generator commit, seed, per-regime quotas, module list, and
996
+ measured yield.
997
+ - `audit/excluded-duplicates.jsonl`, `audit/packing.json`, `audit/source-scan.json`.
998
+ - `audit/validation-evidence.json`: the independent decode and round-trip evidence described below.
999
+ - `statistics.json`: measured records, groups, splits, characters and packed counts, per source.
1000
+ - `manifest.json`: SHA256 and byte size for every delivered file except itself and validation.json.
1001
+ - `packed/`: mmla-type2-packed-v1 arrays, manifest, row attribution and per-record token counts.
1002
+ - `tools/`: the generator driver, the build script and the validation scripts actually run.
1003
+
1004
+ ## Loading
1005
+
1006
+ ~~~python
1007
+ import gzip, json
1008
+ from pathlib import Path
1009
+ component = Path('components/google-math')
1010
+ with gzip.open(component / 'data/train.jsonl.gz', 'rt', encoding='utf-8') as f:
1011
+ first_record = json.loads(next(f))
1012
+
1013
+ import pyarrow.parquet as pq
1014
+ row = pq.read_table(component / 'data/train.parquet').slice(0, 1).to_pylist()[0]
1015
+ metadata = json.loads(row.pop('metadata_json'))
1016
+ first_parquet_record = {{**row, **metadata}}
1017
+ ~~~
1018
+
1019
+ ## Packed training arrays
1020
+
1021
+ `packed/manifest.json` uses `mmla-type2-packed-v1`: input_ids, labels and role_ids are int32 arrays of
1022
+ shape [rows, {SEQ_LEN}], pad=0, role=0, ignored labels=-100. Each record is encoded as
1023
+ BOS + SentencePiece(text) + EOS with the included Cayley-Math tokenizer (32768 pieces); window stride is
1024
+ {SEQ_LEN} with one-token lookahead, labels predict the following token, and targets that cross into a new
1025
+ record's BOS or into padding are masked. Split-specific `rows.jsonl` maps every non-padding token back to
1026
+ its record; `records.jsonl` stores token counts and hashes. This component does not reproduce Phase 1
1027
+ quotas, packing or splits.
1028
+
1029
+ Measured packed counts: train {stats['packed'].get('train', {}).get('rows', 0)} rows /
1030
+ {stats['packed'].get('train', {}).get('tokens', 0)} tokens /
1031
+ {stats['packed'].get('train', {}).get('loss_bearing_labels', 0)} loss-bearing labels; dev
1032
+ {stats['packed'].get('dev', {}).get('rows', 0)} / {stats['packed'].get('dev', {}).get('tokens', 0)} /
1033
+ {stats['packed'].get('dev', {}).get('loss_bearing_labels', 0)}; test
1034
+ {stats['packed'].get('test', {}).get('rows', 0)} / {stats['packed'].get('test', {}).get('tokens', 0)} /
1035
+ {stats['packed'].get('test', {}).get('loss_bearing_labels', 0)}.
1036
+
1037
+ ## Splits
1038
+
1039
+ Splits are assigned per group, never per record, so companions and families stay together: a Lean proof
1040
+ and its prose proof, a conjecture file and its helper library, or one generated question family. Every
1041
+ group key is namespaced by its `source_id`. Assignment sorts SHA256(group) as hex, reserves the first
1042
+ ceil(1%) of groups for dev, the next ceil(1%) for test, and the rest for train.
1043
+ `audit/group-splits.json` freezes every assignment: train
1044
+ {stats['splits']['train']['records']} records over {stats['splits']['train']['groups']} groups,
1045
+ dev {stats['splits']['dev']['records']} over {stats['splits']['dev']['groups']},
1046
+ test {stats['splits']['test']['records']} over {stats['splits']['test']['groups']}.
1047
+
1048
+ ## Text projection
1049
+
1050
+ Lean, geometry-problem and rule files are byte-identical to the pinned blob, license headers included.
1051
+ Notebook records join the cell sources of each notebook with a newline; no cell text is altered.
1052
+ `natural_language_proof` records come from pypdf page-text extraction and are **lossy**: math symbols
1053
+ arrive as Unicode substitutions, ligature artifacts appear, and inline formulas lose their two-dimensional
1054
+ layout. C0/C1 control characters were replaced and blank-line runs collapsed. Each PDF record carries
1055
+ `extraction_quality = lossy_pdf_text_layer` with page and character counts, and `audit/extraction.jsonl`
1056
+ holds the same per-file audit.
1057
+
1058
+ ## Verification evidence
1059
+
1060
+ `tools/validate_google_math.py` checks file hashes, source coverage, unique record IDs and unique text,
1061
+ group split separation, lossless Parquet reconstruction, packed array shapes, causal shift, padding and
1062
+ boundary masks, and re-encodes every record's token stream against `data/*.jsonl.gz`. It writes
1063
+ `validation.json`; the last run passed all 13 checks
1064
+ (`all_delivered_file_hashes_and_sizes`, `pinned_sources_metadata_and_excluded_candidates`,
1065
+ `source_file_inventory_local_bytes`, `group_split_policy_recomputed_from_scratch`,
1066
+ `lossless_jsonl_parquet_exact_text_uniqueness_and_record_provenance`, `group_split_separation`,
1067
+ `verbatim_lean_source_equality_local_bytes`, `lean_and_generated_records_survive`,
1068
+ `per_file_extraction_audit`, `dedup_audit_present`, `generator_audit_seed_and_quota`,
1069
+ `all_packed_tokens_reencoded_source_shift_padding_masks_and_attribution`, `statistics_match_artifacts`).
1070
+
1071
+ `tools/verify_google_math.py` is the independent check against the delivered files, and its output is
1072
+ [`audit/validation-evidence.json`](audit/validation-evidence.json). It decoded 96 packed rows spread over
1073
+ every shard of all three splits, and for each row (a) confirmed every row slice equals the re-encoding of
1074
+ the corresponding span of `[BOS] + SentencePiece(text) + [EOS]` for its attributed source record and
1075
+ (b) confirmed the decoded text equals the decoded source text modulo whitespace. It recorded 8 literal
1076
+ unified diffs of a decoded row against its source record; every diff had 0 lines, whitespace identical.
1077
+ It round-tripped 3 Parquet rows (indices 0, 239,658 and 479,315) back to their JSONL records with exact
1078
+ match, and confirmed a `formal_lean_proof`, a `formal_lean_statement`, a `generated_qa_pair`, a
1079
+ `formal_geometry_problem`, a `natural_language_proof` and a `research_notebook` record all survive with
1080
+ matching `text_sha256`. Split separation was recomputed from the data rather than trusted: 3,403 groups
1081
+ in the audit and 3,403 in the data, no group spanning two splits, train/dev/test at 3,333/35/35 groups.
1082
+
1083
+ [`audit/publish-check.json`](audit/publish-check.json) is a third check written by an independent publish
1084
+ gate with its own code rather than this component's tools. It agrees with every total above (480,584
1085
+ records, 3,403 groups, 78,364,888 characters), decodes 12 packed rows covering 248 whole-record segments
1086
+ with 248/248 token-identical and whitespace-identical, matches 5/5 Parquet round-trips, and reports zero
1087
+ groups spanning two splits.
1088
+
1089
+ The record mix is deliberately stated because it is skewed: `generated_qa_pair` records are 99.31% of
1090
+ records but 45.57% of characters, while formal Lean and prose records are 0.69% of records and 54.43% of
1091
+ characters. A consumer wanting formal or prose mathematics should filter on `record_type` rather than
1092
+ read the corpus whole.
1093
+
1094
+ ## Publication
1095
+
1096
+ The component is published as Hugging Face dataset pull request
1097
+ [MMLA-ORG/Cayley-Math discussion #5](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/discussions/5),
1098
+ on ref `refs/pr/5`, head commit `1894ffe2dc7f7f59abfb830f91e79738b1eb11d3`. It carries the whole
1099
+ component plus the top-level `README.md` config and Collections update, and it has not been merged into
1100
+ `main`.
1101
+
1102
+ A token with `write` scope or org-admin rights is required to merge it. The token used for this build can
1103
+ read the repository and open pull requests, but a direct commit to `main` is rejected with
1104
+ `403 Forbidden: You have read access but not the required permissions for this operation`. Merging that
1105
+ pull request, or re-running `hf upload` without `--create-pr` once such a token is available, completes
1106
+ the submission. Earlier pull requests #3 and #4 are superseded by #5 and were left open rather than
1107
+ closed, because closing them was outside this task's scope.
1108
+
1109
+ ## Limitations
1110
+
1111
+ - No Lean compilation, proof checking, Comparator run or mathematical verification of any kind.
1112
+ - The build ran in a sandbox-free subprocess only for the generator; `cryptids`, `formal-conjectures`,
1113
+ `formal-imo`, `formal-putnam-like`, `alphageometry`, `alphaproof-nexus` and the notebooks were never
1114
+ built or executed, only read as text.
1115
+ - PDF prose proofs are noisy pypdf text; symbol and ligature fidelity is not guaranteed.
1116
+ - The generator subset is a seeded sample of one curriculum, not the paper's full dataset, and its
1117
+ questions are school-level rather than research mathematics. One module, `algebra__linear_1d`, stopped
1118
+ at 3,055 of its 8,000 quota when upstream's own `assert entropy == 0` fired; nothing was repaired, and
1119
+ the shortfall is recorded in `audit/mathematics-dataset-generation.json`.
1120
+ - `formal-conjectures` upstream states that some content derives from third-party sources (Wikipedia,
1121
+ MathOverflow and OEIS under CC-BY-SA 4.0). Per-file reference URLs are retained inside the Lean text; no
1122
+ separate per-file license audit was performed.
1123
+ - Near-duplicate detection, Phase 1 overlap analysis and editorial review were not performed. 1,629
1124
+ exact duplicate texts were dropped and are listed in `audit/excluded-duplicates.jsonl`.
1125
+ - The SentencePiece model normalises runs of spaces, so packed text decodes whitespace-normalised rather
1126
+ than byte-identically to the sources.
1127
+ - Only problems that upstream published are present, so several subsets are selection-biased. `cryptids`
1128
+ contributes 31.6M of the 78.4M characters because a handful of Busy Beaver traces are enormous, so the
1129
+ corpus is dominated by machine-generated proof script rather than mathematics prose.
1130
+
1131
+ ## Statistics
1132
+
1133
+ ~~~json
1134
+ {stats_json}
1135
+ ~~~
1136
+ """
1137
+ (root / 'README.md').write_text(readme, encoding='utf-8')
1138
+
1139
+
1140
+ if __name__ == '__main__':
1141
+ main()