hanzceo commited on
Commit
84ea80a
·
verified ·
1 Parent(s): e86f79b

Penny-1B-Base-L1: 4,076,511 documents (text)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. LICENSE +4 -0
  2. README.md +107 -0
  3. data/train/code-00000.parquet +3 -0
  4. data/train/code-00001.parquet +3 -0
  5. data/train/code-00002.parquet +3 -0
  6. data/train/kg-00000.parquet +3 -0
  7. data/train/kg-00001.parquet +3 -0
  8. data/train/kg-00002.parquet +3 -0
  9. data/train/math_solutions-00000.parquet +3 -0
  10. data/train/warmup-00000.parquet +3 -0
  11. data/train/web-00000.parquet +3 -0
  12. data/train/web-00001.parquet +3 -0
  13. data/train/web-00002.parquet +3 -0
  14. data/train/web-00003.parquet +3 -0
  15. data/train/web-00004.parquet +3 -0
  16. data/train/web-00005.parquet +3 -0
  17. data/train/web-00006.parquet +3 -0
  18. data/train/web-00007.parquet +3 -0
  19. data/train/web-00008.parquet +3 -0
  20. data/train/web-00009.parquet +3 -0
  21. data/train/web-00010.parquet +3 -0
  22. data/train/web-00011.parquet +3 -0
  23. data/train/web-00012.parquet +3 -0
  24. data/train/web-00013.parquet +3 -0
  25. data/train/web-00014.parquet +3 -0
  26. data/train/web-00015.parquet +3 -0
  27. data/train/web-00016.parquet +3 -0
  28. data/train/web-00017.parquet +3 -0
  29. data/train/web-00018.parquet +3 -0
  30. data/train/web-00019.parquet +3 -0
  31. data/train/web-00020.parquet +3 -0
  32. data/train/web-00021.parquet +3 -0
  33. data/train/web-00022.parquet +3 -0
  34. data/train/web-00023.parquet +3 -0
  35. data/train/web-00024.parquet +3 -0
  36. data/train/web-00025.parquet +3 -0
  37. data/train/web-00026.parquet +3 -0
  38. data/train/web-00027.parquet +3 -0
  39. data/train/web-00028.parquet +3 -0
  40. data/train/web-00029.parquet +3 -0
  41. data/train/web-00030.parquet +3 -0
  42. data/train/web-00031.parquet +3 -0
  43. data/train/worlds-00000.parquet +3 -0
  44. data/train/worlds-00001.parquet +3 -0
  45. data/train/worlds-00002.parquet +3 -0
  46. data/train/worlds-00003.parquet +3 -0
  47. data/validation/code-00000.parquet +3 -0
  48. data/validation/kg-00000.parquet +3 -0
  49. data/validation/math_solutions-00000.parquet +3 -0
  50. data/validation/warmup-00000.parquet +3 -0
LICENSE ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ This dataset mixes sources under different licenses (see README.md, 'Sources'):
2
+ - warmup, worlds, kg, code: generated for Penny, Apache License 2.0
3
+ - web: from openbmb/Ultra-FineWeb-L3, Apache License 2.0
4
+ - math_solutions: from nvidia/OpenMathInstruct-2, CC BY 4.0 (attribution: NVIDIA)
README.md ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: mixed
4
+ license_link: LICENSE
5
+ language: [en]
6
+ task_categories: [text-generation]
7
+ size_categories: [1M<n<10M]
8
+ pretty_name: Penny-1B-Base-L1 pretraining data
9
+ tags: [penny, pretraining, synthetic, reading-comprehension, knowledge-light]
10
+ configs:
11
+ - config_name: default
12
+ default: true
13
+ data_files:
14
+ - split: train
15
+ path: data/train/*.parquet
16
+ - split: validation
17
+ path: data/validation/*.parquet
18
+ - config_name: warmup
19
+ data_files:
20
+ - split: train
21
+ path: data/train/warmup-*.parquet
22
+ - split: validation
23
+ path: data/validation/warmup-*.parquet
24
+ - config_name: web
25
+ data_files:
26
+ - split: train
27
+ path: data/train/web-*.parquet
28
+ - split: validation
29
+ path: data/validation/web-*.parquet
30
+ - config_name: worlds
31
+ data_files:
32
+ - split: train
33
+ path: data/train/worlds-*.parquet
34
+ - split: validation
35
+ path: data/validation/worlds-*.parquet
36
+ - config_name: kg
37
+ data_files:
38
+ - split: train
39
+ path: data/train/kg-*.parquet
40
+ - split: validation
41
+ path: data/validation/kg-*.parquet
42
+ - config_name: code
43
+ data_files:
44
+ - split: train
45
+ path: data/train/code-*.parquet
46
+ - split: validation
47
+ path: data/validation/code-*.parquet
48
+ - config_name: math_solutions
49
+ data_files:
50
+ - split: train
51
+ path: data/train/math_solutions-*.parquet
52
+ - split: validation
53
+ path: data/validation/math_solutions-*.parquet
54
+ ---
55
+
56
+ # Penny-1B-Base-L1 pretraining data
57
+
58
+ The text of every document in the pretraining stage `Penny-1B-Base-L1` of
59
+ [Penny](https://huggingface.co/LocalWisdom), 2.99B tokens of training in all.
60
+ Each row is one whole document, exactly as the stage's build kept it, before tokenization.
61
+
62
+ Penny is meant to be **knowledge-light**: it should learn from the material it is given (a knowledge
63
+ graph, documentation) and reason over it, rather than recall facts. So about half of this stage is
64
+ generated documents that define something new (rules, a knowledge graph, a programming language,
65
+ every name invented for that document) and then use it. Nothing in them can be memorized, so a
66
+ model has to read them. The rest is English web text, for language, and worked math solutions.
67
+
68
+ ## Sources
69
+
70
+ | source | what it is | license | unique train tokens | tokens in the stage | train docs | validation docs | files |
71
+ |---|---|---|---:|---:|---:|---:|---:|
72
+ | `warmup` | procedural warm-up: short algorithmic tasks (bracket matching, sorting, set operations, stack machines, string transforms), trained first | apache-2.0 (generated) | 0.010B | 0.010B (0.3%) | 6,759 | 45 | 2 |
73
+ | `web` | English web text, [openbmb/Ultra-FineWeb-L3](https://huggingface.co/datasets/openbmb/Ultra-FineWeb-L3) (`multi_style`) | apache-2.0 (upstream) | 1.390B | 1.390B (46.5%) | 2,983,202 | 1,830 | 33 |
74
+ | `worlds` | generated: rule systems, counterfactual worlds, invented units, operators and languages, each defined in the document, then statements of what follows (easy ones first) | apache-2.0 (generated) | 0.600B | 0.600B (20.1%) | 383,582 | 338 | 5 |
75
+ | `kg` | generated: knowledge graphs of invented people, organizations and places, then plain statements of what follows from them (one-hop lookups first, then chains, counts and comparisons) | apache-2.0 (generated) | 0.450B | 0.450B (15.1%) | 241,510 | 286 | 4 |
76
+ | `code` | generated: invented programming languages and libraries (reference first, then programs with their outputs and programs written from a description) | apache-2.0 (generated) | 0.450B | 0.450B (15.1%) | 205,934 | 227 | 4 |
77
+ | `math_solutions` | worked math solutions, [nvidia/OpenMathInstruct-2](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2) | cc-by-4.0 (upstream) | 0.090B | 0.090B (3.0%) | 252,058 | 740 | 2 |
78
+
79
+ "Unique train tokens" are what the build tokenized (generated sources: including the end-of-window
80
+ padding of packed 32K windows); "tokens in the stage" is what training reads from that source.
81
+ Tokenizer: `tiny-random/deepseek-v4.1-flash-1017` (the DeepSeek-V4.1 tokenizer); documents are
82
+ separated by token id 1 in training.
83
+
84
+ ## Use
85
+
86
+ ```python
87
+ from datasets import load_dataset
88
+
89
+ ds = load_dataset("LocalWisdom/Penny-1B-Base-L1-data", split="train") # every source
90
+ kg = load_dataset("LocalWisdom/Penny-1B-Base-L1-data", "kg", split="validation") # one source
91
+ ```
92
+
93
+ Columns: `text` (the document), `source` (one of the names above). Files are zstd parquet of at
94
+ most ~95 MB.
95
+
96
+ ## How it was built
97
+
98
+ `scripts/prepare_mix.py --stage Penny-1B-Base-L1 --text ...` from the Penny code: web and math rows in the
99
+ spec's file order up to each source's token budget (validation first, then train), generated
100
+ documents from fixed seeds (`penny.synth`, no language model involved). The build is deterministic,
101
+ so the same command rebuilds this dataset.
102
+
103
+ ## License
104
+
105
+ Mixed, per source (table above): the generated sources (`warmup`, `worlds`, `kg`, `code`) are
106
+ Apache-2.0; `web` keeps Ultra-FineWeb's Apache-2.0; `math_solutions` keeps OpenMathInstruct-2's
107
+ CC-BY-4.0 (attribution: NVIDIA, OpenMathInstruct-2). Check the upstream cards for their terms.
data/train/code-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c0be34bd70b4364c7f0e29bb590fb3a48cfd7bd479464c6b22f445b97146b24
3
+ size 96678448
data/train/code-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48135587802e28940e95b6ee952f61e77ce818851b76319cdab2e1c918878d2e
3
+ size 96701046
data/train/code-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:828aaac2cf0145af84c78ce7937b3ce6120080126e53538aea399f553242ee30
3
+ size 33745133
data/train/kg-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bf6464ba7a506d951b678f092c2c5b6015a0cdaa87a38806b019016968ad788
3
+ size 96366819
data/train/kg-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f62a7e38bca3a4502384e7c512608bc9b809b62de651196d1a23e7ceb63af53c
3
+ size 96312110
data/train/kg-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f8fe3fac20081846b5eda1c9e1dbd16d577053fe42d22e9a959992338478c9c
3
+ size 45659143
data/train/math_solutions-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05935ceb5712f79cc8050bb6cb517bb1cca6da96ec7e3ac9dbf99a11ad71a23d
3
+ size 88504210
data/train/warmup-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e053a4e0ce78007dd7ee57dbe5bfb02ab2549520374bbf7f4d433c3e43b11966
3
+ size 6091247
data/train/web-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6aff029fbed00b4a2c946e2baa223acfad5e35d3e0836713824c08414d3304fd
3
+ size 98571747
data/train/web-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d0824f9a01c8fbccc243dc88640f0101facd4ea4d7183d894625cc2abf61238
3
+ size 98489580
data/train/web-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1ec4c82330f501b3aa4a8b43104ee1a2497742783d892820894622b00f0718f
3
+ size 98161339
data/train/web-00003.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:97be8c2c3907e03bb1149a0776f7328a639bd410f46e205692d7eb86e2a7e405
3
+ size 98112324
data/train/web-00004.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2cfe842f6e41faa42f324d157284836c2a387b8f92a9558da0e675c47f131097
3
+ size 98200037
data/train/web-00005.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b545cbd050744457e728e1d5e372123495e39eb0fcd1e5c76b24817f2be66c4
3
+ size 98300125
data/train/web-00006.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00819845d8d169b99ebb0fca0f2622e5e771d52fc628f85070c7ae9758a80211
3
+ size 98462533
data/train/web-00007.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccd04888e0aaf5758d0d3c767d713cad9fb0d32cb58c351d539a35218d70120f
3
+ size 98506658
data/train/web-00008.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48b037ed609ee325f9edb93e3dec77aa39310017b4c39228797176fd34015582
3
+ size 98193413
data/train/web-00009.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:973dc110c5e115f29c1b7d739bc37a11472b9012698c00e7bc4343cc178158bc
3
+ size 98446693
data/train/web-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ce97f03b021cf23a476e775628c38d14f06217c11c1e9a48242d6f8124ca757
3
+ size 98064722
data/train/web-00011.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d11fe9518f764336856e787060b778601579f4900b826671d775f5855a09afee
3
+ size 98251366
data/train/web-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f21d2ddc1123b57eb9040f864a7d30e0313b9266ad9c9a1f2e920b1d7a8c2f7
3
+ size 98363638
data/train/web-00013.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bffda89fbdc44b61ed5a8718a2bf9a458772b54dd2654e622689c8cbe64c194e
3
+ size 98305814
data/train/web-00014.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4069da4c2df239ae4a9c804d9d7601bdd9ec095f1ca56ca1123e81cbbd7f9b92
3
+ size 98377738
data/train/web-00015.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2dd287e21442515889a267467578c3928c67cb1e88f11dd498b03d377e186acd
3
+ size 98603166
data/train/web-00016.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa2ac678cc33e476f6ed589c71aaefc156a896ec8348c7acfe0c68a5890282af
3
+ size 98238235
data/train/web-00017.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f36a1da88d869b346fe6d752326f88e0561699cdd7e88c03bd2a7917b03d527
3
+ size 98103508
data/train/web-00018.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb0ced67f2b0ed584572fbc5bbc39efde1867ed534bf000652ae22f51f0bf73e
3
+ size 98389226
data/train/web-00019.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8545d239352a700666d7c6a31d28f73a7e3ef692b67f3ea5c135782d754f917d
3
+ size 98082461
data/train/web-00020.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4bed4490eeb0cb25f0e2aa8309b57007543e4ac90c5828e1ca8ad45d862a09e9
3
+ size 98321696
data/train/web-00021.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd4e5e40c8cb1973a32e2b85cce9c6cdc1e5dc8e8a061a5df57e9ed602afb964
3
+ size 98321174
data/train/web-00022.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6160fccd67cfeaf3bad104187793b306393b332ea078db2eb07d849b77fea982
3
+ size 98401971
data/train/web-00023.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7268dbd18bd6a20658232a4c4742fe54f338c5d852b675640c60752ed90ec8b
3
+ size 98348830
data/train/web-00024.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17ccb8d6e001a611c9262332c57f1d42d976458a1ee943632ba403124366f758
3
+ size 98513829
data/train/web-00025.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c381522786246b71282cac14773d146de55777519c034c19808854c0c4bcec9
3
+ size 98021264
data/train/web-00026.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:345f4073377316f45740c227c0a67e6888b0abffddd7d0425058859d941bf018
3
+ size 98416079
data/train/web-00027.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8514b1f673e5e44a6f1454478b4165a4f67ed5457e68b52125103c561596d653
3
+ size 98270113
data/train/web-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef9d413ab7c3143d8bd5d8feebc49e7fdd2239bfd875fe341586834a399cd40b
3
+ size 98118332
data/train/web-00029.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b18e6df69100f3e1cd5674ab149b65e6c7791b15e3217f1abb4c0da6a0d3f546
3
+ size 98761738
data/train/web-00030.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b0c06c0b5c9549715e87fa75d56478971b1212dd2ed7d3594a64831ca571ddb
3
+ size 98280926
data/train/web-00031.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25395408cb30c72001744639cf7f63f18df53672102585cbea7f3ab8eafb9b72
3
+ size 40549690
data/train/worlds-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e65917b61fd766f4dded2f8991da6b445054e96fe9eaacbd78e2ee39f755cee
3
+ size 97094024
data/train/worlds-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca8e6ac07272d527549fa4a92c68362b141bda2eccfb61cfae7282d44acf3441
3
+ size 97146728
data/train/worlds-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a55a7328e70bca8e69c53ab1954b60e5cf9acbf9e3d1efbf1437023fa3a10dea
3
+ size 97089917
data/train/worlds-00003.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec6f93217a948a20053a98c1b366039de72957a058cae16a3aeac32ee4de8d33
3
+ size 86163631
data/validation/code-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a9d13ae49155c6b24a51af54c290180de8358913092422cfb584d632074ebc6
3
+ size 264247
data/validation/kg-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf0b0f72e7679865de920908b105ec2c1ee5ee578e3204680b8ae736582daeb7
3
+ size 291328
data/validation/math_solutions-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:685fee9aa9bc86540c45087263205ecb490b763bc0c0911c632ef65839df1931
3
+ size 260260
data/validation/warmup-00000.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eca7ad3955663a6438a7335383bbe0f06cd550defc3eb379996720c08609c2c4
3
+ size 56586