Datasets:
docs: align the README template with the published README wording
Browse files
components/google-math/tools/build_google_math.py
CHANGED
|
@@ -930,10 +930,16 @@ school-level difficulty. Those pairs are **not research mathematics**, and the s
|
|
| 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.**
|
| 934 |
-
|
| 935 |
-
|
| 936 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 937 |
|
| 938 |
## Included sources
|
| 939 |
|
|
@@ -996,6 +1002,7 @@ The short version:
|
|
| 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.
|
|
@@ -1069,10 +1076,11 @@ boundary masks, and re-encodes every record's token stream against `data/*.jsonl
|
|
| 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
|
| 1073 |
-
|
| 1074 |
-
the corresponding span of `[BOS] + SentencePiece(text) + [EOS]` for its
|
| 1075 |
-
(b) confirmed the decoded text equals the decoded source text modulo
|
|
|
|
| 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
|
|
@@ -1080,31 +1088,34 @@ match, and confirmed a `formal_lean_proof`, a `formal_lean_statement`, a `genera
|
|
| 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
|
| 1084 |
-
|
| 1085 |
-
|
| 1086 |
-
|
| 1087 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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
|
| 1091 |
-
|
| 1092 |
-
read the corpus whole.
|
| 1093 |
|
| 1094 |
## Publication
|
| 1095 |
|
| 1096 |
-
The component is
|
| 1097 |
-
[MMLA-ORG/Cayley-Math discussion #
|
| 1098 |
-
on ref `refs/pr/
|
| 1099 |
-
|
| 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
|
| 1105 |
pull request, or re-running `hf upload` without `--create-pr` once such a token is available, completes
|
| 1106 |
-
the submission.
|
| 1107 |
-
closed, because closing them was outside this task's scope.
|
| 1108 |
|
| 1109 |
## Limitations
|
| 1110 |
|
|
|
|
| 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.** Measured over the delivered
|
| 934 |
+
data, `generated_qa_pair` records are 477,276 of 480,584 records, which is 99.31% of records (0.993117)
|
| 935 |
+
and 35,710,731 of 78,364,888 text characters (45.57%). The formal subset is a small minority by record
|
| 936 |
+
count. Lean records total 3,027, namely 535 `formal_lean_proof` plus 2,492 `formal_lean_statement`, which
|
| 937 |
+
is 0.63% of records (0.006299). Geometry problems, prose proofs and notebooks add 281 more records, so
|
| 938 |
+
every non-generator record together is 3,308 records, 0.69% of the corpus (0.006883). Lean text does carry
|
| 939 |
+
54.22% of the characters (0.542177) because Lean files are long, so a character-weighted view inverts the
|
| 940 |
+
balance while the record-weighted view does not. Read the composition from records, not from bytes.
|
| 941 |
+
**No Lean compilation and no proof verification was performed on any subset**, so no record in this corpus
|
| 942 |
+
has a checked proof or a checked statement.
|
| 943 |
|
| 944 |
## Included sources
|
| 945 |
|
|
|
|
| 1002 |
measured yield.
|
| 1003 |
- `audit/excluded-duplicates.jsonl`, `audit/packing.json`, `audit/source-scan.json`.
|
| 1004 |
- `audit/validation-evidence.json`: the independent decode and round-trip evidence described below.
|
| 1005 |
+
- `audit/publish-check.json`: the second publish gate, measured from the delivered files after the build.
|
| 1006 |
- `statistics.json`: measured records, groups, splits, characters and packed counts, per source.
|
| 1007 |
- `manifest.json`: SHA256 and byte size for every delivered file except itself and validation.json.
|
| 1008 |
- `packed/`: mmla-type2-packed-v1 arrays, manifest, row attribution and per-record token counts.
|
|
|
|
| 1076 |
`all_packed_tokens_reencoded_source_shift_padding_masks_and_attribution`, `statistics_match_artifacts`).
|
| 1077 |
|
| 1078 |
`tools/verify_google_math.py` is the independent check against the delivered files, and its output is
|
| 1079 |
+
[`audit/validation-evidence.json`](audit/validation-evidence.json). It decoded 96 packed rows, covering
|
| 1080 |
+
68 of the 70 train shards plus the single dev and test shards, and for each row (a) confirmed every row
|
| 1081 |
+
slice equals the re-encoding of the corresponding span of `[BOS] + SentencePiece(text) + [EOS]` for its
|
| 1082 |
+
attributed source record and (b) confirmed the decoded text equals the decoded source text modulo
|
| 1083 |
+
whitespace. It recorded 8 literal
|
| 1084 |
unified diffs of a decoded row against its source record; every diff had 0 lines, whitespace identical.
|
| 1085 |
It round-tripped 3 Parquet rows (indices 0, 239,658 and 479,315) back to their JSONL records with exact
|
| 1086 |
match, and confirmed a `formal_lean_proof`, a `formal_lean_statement`, a `generated_qa_pair`, a
|
|
|
|
| 1088 |
matching `text_sha256`. Split separation was recomputed from the data rather than trusted: 3,403 groups
|
| 1089 |
in the audit and 3,403 in the data, no group spanning two splits, train/dev/test at 3,333/35/35 groups.
|
| 1090 |
|
| 1091 |
+
[`audit/publish-check.json`](audit/publish-check.json) is a second publish gate with its own code rather
|
| 1092 |
+
than this component's tools. It agrees with every total above (480,584 records, 3,403 groups, 78,364,888
|
| 1093 |
+
characters, and the packed counts for all three splits), decodes 12 packed rows across all three splits
|
| 1094 |
+
and compares 263 attributed segments. All 263 row slices matched their canonical token slices by exact
|
| 1095 |
+
identity, and all 248 segments that cover a whole record decoded to text identical to the source record
|
| 1096 |
+
under whitespace normalisation and re-encoded to the identical token sequence. The other 15 segments are
|
| 1097 |
+
row-boundary fragments that begin or end inside a SentencePiece piece, so they cannot re-encode
|
| 1098 |
+
fragment-by-fragment; each one's decoded text was still located verbatim inside its attributed record.
|
| 1099 |
+
It matched 5 of 5 Parquet round-trips exactly and recomputed group separation from `data/*.jsonl.gz`
|
| 1100 |
+
directly, finding 0 groups spanning two splits.
|
| 1101 |
|
| 1102 |
The record mix is deliberately stated because it is skewed: `generated_qa_pair` records are 99.31% of
|
| 1103 |
+
records but 45.57% of characters, while formal Lean records are 0.63% of records and 54.22% of characters,
|
| 1104 |
+
and all non-generator records together are 0.69% of records. A consumer wanting formal or prose
|
| 1105 |
+
mathematics should filter on `record_type` rather than read the corpus whole.
|
| 1106 |
|
| 1107 |
## Publication
|
| 1108 |
|
| 1109 |
+
The component is delivered as Hugging Face dataset pull request
|
| 1110 |
+
[MMLA-ORG/Cayley-Math discussion #3](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/discussions/3)
|
| 1111 |
+
on ref `refs/pr/3`. It carries the whole component plus the top-level `README.md` config and Collections
|
| 1112 |
+
update, and it has not been merged into `main`. Consult that discussion for the current merge state.
|
|
|
|
| 1113 |
|
| 1114 |
A token with `write` scope or org-admin rights is required to merge it. The token used for this build can
|
| 1115 |
read the repository and open pull requests, but a direct commit to `main` is rejected with
|
| 1116 |
+
`403 Forbidden: You have read access but not the required permissions for this operation`. Merging the
|
| 1117 |
pull request, or re-running `hf upload` without `--create-pr` once such a token is available, completes
|
| 1118 |
+
the submission.
|
|
|
|
| 1119 |
|
| 1120 |
## Limitations
|
| 1121 |
|