JunyiZou commited on
Commit
ae848f8
·
verified ·
1 Parent(s): 32f387d

Expose all frozen Anholon math windows in Data Studio

Browse files
README.md CHANGED
@@ -11,11 +11,17 @@ tags:
11
  - language-modeling
12
  - mathematics
13
  configs:
14
- - config_name: phase1-preview
15
  default: true
16
  data_files:
17
  - split: train
18
- path: phase1/preview/*.parquet
 
 
 
 
 
 
19
  - config_name: openai-math
20
  data_files:
21
  - split: train
@@ -40,15 +46,16 @@ Cayley-Math is a collaboratively maintained collection of mathematical text for
40
 
41
  | Use | Files | Coverage |
42
  | --- | --- | --- |
43
- | Train with the frozen release | [NumPy archives](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/tree/main/phase1/math) | 50,000,000 training labels |
44
- | Inspect examples | [Parquet preview](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/tree/main/phase1/preview) | 4 excerpts from the training split |
45
  | Trace sources and terms | [Source catalog](phase1/sources.json) · [Attribution guide](phase1/attribution/README.md) | All included components |
46
 
47
  ## Phase 1 Summary
48
 
49
  | Split | Loss-bearing labels | Packed rows |
50
  | --- | ---: | ---: |
51
- | Train | 50,000,000 | 43,833 |
 
52
  | Development | 1,024 | 1 |
53
  | Test | 1,024 | 1 |
54
 
@@ -60,7 +67,7 @@ Sources include mathematical arXiv passages cached by Common Pile, excerpts from
60
  phase1/
61
  ├── math/ # Mathematics archive and packed manifest
62
  ├── attribution/ # Document-level sources, licenses, and notices
63
- ├── preview/ # Parquet excerpts
64
  ├── sources.json # Component counts and provenance references
65
  └── upload-manifest.json # File sizes and hashes by repository path
66
  components/ # Independently contributed collections
@@ -68,28 +75,13 @@ tokenizer/ # SentencePiece model and vocabulary for Phase 1
68
  examples/ # Synthetic illustration of a source record
69
  ```
70
 
71
- ## Data Preview
72
-
73
- The Parquet preview contains **4 excerpts** decoded from frozen training windows. Each excerpt contains up to 192 tokens and includes its domain, source, license, and packed row number. The files are under [`phase1/preview/`](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/tree/main/phase1/preview).
74
-
75
- | Domain | Example excerpt | Source |
76
- | --- | --- | --- |
77
- | Mathematics | , we start with the notion of joint transitivity. ###### Definition 1.4. Let $(X,S_{1},\ldots,S_{k})$ be a $\mathbb{Z}^{k}$-system, $T_{1},\ldots,T_{d}\in\langl… | [Upstream](https://arxiv.org/abs/2404.07876) |
78
 
79
- The preview uses four fixed row offsets per domain at 20%, 40%, 60%, and 80% of its packed training rows. Each excerpt includes up to 192 tokens. The `packed_row` field is a zero-based index into Anholon's full Phase 1 train split, including in the mathematics mirror.
80
 
81
- ```python
82
- import pyarrow.parquet as pq
83
- from huggingface_hub import hf_hub_download
84
 
85
- preview = hf_hub_download(
86
- repo_id="MMLA-ORG/Cayley-Math",
87
- repo_type="dataset",
88
- filename="phase1/preview/math.parquet",
89
- endpoint="https://huggingface.co",
90
- )
91
- table = pq.read_table(preview)
92
- ```
93
 
94
  ## Data Fields
95
 
@@ -118,7 +110,7 @@ math = root / "phase1" / "math"
118
  subprocess.run(["tar", "--zstd", "-xf", str(math / "math.tar.zst"), "-C", str(math)], check=True)
119
  ```
120
 
121
- Use `math/manifest.json` to locate and verify the extracted arrays. Development and test are small monitoring splits.
122
 
123
  ## Processing and Source Terms
124
 
 
11
  - language-modeling
12
  - mathematics
13
  configs:
14
+ - config_name: anholon-phase1
15
  default: true
16
  data_files:
17
  - split: train
18
+ path:
19
+ - phase1/preview/full/train-initial.parquet
20
+ - phase1/preview/full/train-continuation.parquet
21
+ - split: validation
22
+ path: phase1/preview/full/validation.parquet
23
+ - split: test
24
+ path: phase1/preview/full/test.parquet
25
  - config_name: openai-math
26
  data_files:
27
  - split: train
 
46
 
47
  | Use | Files | Coverage |
48
  | --- | --- | --- |
49
+ | Train with the frozen release | [NumPy archives](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/tree/main/phase1/math) | 250,000,000 training labels across both batches |
50
+ | Browse every packed mathematics window | [Data Studio](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/viewer/anholon-phase1/train) | 146,965 training rows plus validation and test |
51
  | Trace sources and terms | [Source catalog](phase1/sources.json) · [Attribution guide](phase1/attribution/README.md) | All included components |
52
 
53
  ## Phase 1 Summary
54
 
55
  | Split | Loss-bearing labels | Packed rows |
56
  | --- | ---: | ---: |
57
+ | Train, initial batch | 50,000,000 | 43,833 |
58
+ | Train, continuation | 200,000,000 | 103,132 |
59
  | Development | 1,024 | 1 |
60
  | Test | 1,024 | 1 |
61
 
 
67
  phase1/
68
  ├── math/ # Mathematics archive and packed manifest
69
  ├── attribution/ # Document-level sources, licenses, and notices
70
+ ├── preview/ # Full text viewer under full/
71
  ├── sources.json # Component counts and provenance references
72
  └── upload-manifest.json # File sizes and hashes by repository path
73
  components/ # Independently contributed collections
 
75
  examples/ # Synthetic illustration of a source record
76
  ```
77
 
78
+ ## Browse the dataset
 
 
 
 
 
 
79
 
80
+ Select [`anholon-phase1`](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/viewer/anholon-phase1/train) in Data Studio. The `train` split contains all 43,833 initial mathematics windows and all 103,132 continuation mathematics windows. The original validation and test splits each contain one row. Select `openai-math` for the independently contributed collection.
81
 
82
+ Each Parquet row decodes one frozen window and records its batch, row number within that batch and split, supervised label count, source, license and attribution. Padding and control tokens are omitted in the displayed text. Consecutive windows can share a boundary token. The NumPy archives contain the original token IDs and label masks.
 
 
83
 
84
+ The Parquet files and their hashes are in [phase1/preview/full](https://huggingface.co/datasets/MMLA-ORG/Cayley-Math/tree/main/phase1/preview/full). The Stacks Project rows have a [compressed editable UTF-8 copy](phase1/preview/full/stacks-transparent.txt.gz) with the source notices and GFDL text.
 
 
 
 
 
 
 
85
 
86
  ## Data Fields
87
 
 
110
  subprocess.run(["tar", "--zstd", "-xf", str(math / "math.tar.zst"), "-C", str(math)], check=True)
111
  ```
112
 
113
+ Use `math/manifest.json` to locate and verify the extracted initial arrays. The [continuation bundle](phase1/math/continuation-4b/README.md) has its own archive and manifest. Development and test are small monitoring splits.
114
 
115
  ## Processing and Source Terms
116
 
phase1/preview/full/manifest.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "source_manifests_sha256": {
3
+ "initial": "b09db946a42ccf762d92fc4f482d6c2bfc9894a1efdb2eed57fae6a4f3885b00",
4
+ "continuation": "2b44a794be940fbef2fc76f9ba27c633e5e08970cd9ffda2db5992e079f93ab2"
5
+ },
6
+ "coverage": {
7
+ "train-initial.parquet": {
8
+ "rows": 43833,
9
+ "labels": 50000000,
10
+ "stacks_rows": 18084
11
+ },
12
+ "train-continuation.parquet": {
13
+ "rows": 103132,
14
+ "labels": 200000000,
15
+ "stacks_rows": 0
16
+ },
17
+ "validation.parquet": {
18
+ "rows": 1,
19
+ "labels": 1024,
20
+ "stacks_rows": 0
21
+ },
22
+ "test.parquet": {
23
+ "rows": 1,
24
+ "labels": 1024,
25
+ "stacks_rows": 0
26
+ }
27
+ },
28
+ "files": {
29
+ "train-initial.parquet": {
30
+ "sha256": "953bef3df7da17d65956b1530f62130e3335731b467ca267b4878b9e13f924f2",
31
+ "bytes": 38221347
32
+ },
33
+ "train-continuation.parquet": {
34
+ "sha256": "3c2a720a9473c98a2a4d3393c5a099f0e777ef6ae70142cf8c004379cc6a7a2d",
35
+ "bytes": 156753022
36
+ },
37
+ "validation.parquet": {
38
+ "sha256": "a5aeec8e4ecb5e71f2c07657666cb220ade5879000701ef5035f97b515a6bbd6",
39
+ "bytes": 6162
40
+ },
41
+ "test.parquet": {
42
+ "sha256": "1d53652017c3567163c34fcf845fffaab8e730b37568bfd42c3dfe7a5679adfc",
43
+ "bytes": 5578
44
+ },
45
+ "stacks-transparent.txt.gz": {
46
+ "sha256": "878fc5d5b5bd99d84861c7aba1ff67919d8ebf2002a4d23941e588dc6ca32fc6",
47
+ "bytes": 1469199
48
+ }
49
+ }
50
+ }
phase1/preview/full/stacks-transparent.txt.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:878fc5d5b5bd99d84861c7aba1ff67919d8ebf2002a4d23941e588dc6ca32fc6
3
+ size 1469199
phase1/preview/full/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d53652017c3567163c34fcf845fffaab8e730b37568bfd42c3dfe7a5679adfc
3
+ size 5578
phase1/preview/full/train-continuation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c2a720a9473c98a2a4d3393c5a099f0e777ef6ae70142cf8c004379cc6a7a2d
3
+ size 156753022
phase1/preview/full/train-initial.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:953bef3df7da17d65956b1530f62130e3335731b467ca267b4878b9e13f924f2
3
+ size 38221347
phase1/preview/full/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5aeec8e4ecb5e71f2c07657666cb220ade5879000701ef5035f97b515a6bbd6
3
+ size 6162
phase1/upload-manifest.json CHANGED
@@ -39,8 +39,8 @@
39
  "sha256": "5ed89ba3ac20243fecf1a8945a7d21e41a9a39e7ff3dfc1a584889360a92a646"
40
  },
41
  "README.md": {
42
- "bytes": 5590,
43
- "sha256": "5b1bfda83e351adaff742c45d71255b410d594daf1b7077968147db6642a8e9f"
44
  },
45
  "phase1/preview/math.parquet": {
46
  "bytes": 5952,
 
39
  "sha256": "5ed89ba3ac20243fecf1a8945a7d21e41a9a39e7ff3dfc1a584889360a92a646"
40
  },
41
  "README.md": {
42
+ "bytes": 6261,
43
+ "sha256": "ef32c347dd68b40ed2c4dd04df1866bb1a13219b4ce2c6e4faf6a20d3d7863dd"
44
  },
45
  "phase1/preview/math.parquet": {
46
  "bytes": 5952,