CoolFace
Datasetpublic

Yujivus/nanochat-climbmix-arithmetic-base7

nanochat ClimbMix + Arithmetic: base-7 numeral world This is a deterministic base-7 rendering of Yujivus/nanochat-climbmix-arithmetic-base10. It preserves the exact shard names, row order, document order, arithmetic-document placement, and non-numeric text of the source dataset. Transformation rule Every maximal ASCII digit run matching [0-9]+ is interpreted as a base-10 integer and rendered in base 7. Leading zeros are preserved as a prefix; signs, punctuation… See the full description on the dataset page: https://huggingface.co/datasets/Yujivus/nanochat-climbmix-arithmetic-base7.

sourceHugging Facecc-by-nc-4.0updated 1mo agoView on Hugging Face
0likes1.3kdownloads
Dataset Card

nanochat ClimbMix + Arithmetic: base-7 numeral world

This is a deterministic base-7 rendering of `Yujivus/nanochat-climbmix-arithmetic-base10`. It preserves the exact shard names, row order, document order, arithmetic-document placement, and non-numeric text of the source dataset.

Transformation rule

Every maximal ASCII digit run matching [0-9]+ is interpreted as a base-10 integer and rendered in base 7. Leading zeros are preserved as a prefix; signs, punctuation, spacing, and all non-digit characters are unchanged.

Examples:

text
12 + 34 = 46  ->  15 + 46 = 64
19 - 7 = 12   ->  25 - 10 = 15
year 2026     ->  year 5623
0012          ->  0015

The rule is deliberately lexical and global. Consequently, digit runs inside dates, URLs, identifiers, version strings, decimal notation, and scientific notation are transformed independently. This is not a rational-number rewrite of fractional notation.

Contents

  • —Training: shard_00000.parquet through shard_00169.parquet
  • —Validation: shard_06542.parquet
  • —base7_manifest.json: pinned source revision, transformation definition, per-shard row counts, byte sizes, replacement counts, and SHA-256 hashes
  • —source_arithmetic_manifest.json: the source dataset's arithmetic manifest

Pinned source revision: 61e1951a802b0d5a17b69bf712407e7ec0acb411

Intended controlled comparison

The paired base-10/base-7 experiments use the same 32,768-token digit-level tokenizer, nanochat architecture, seed, optimizer, schedule, and token budget. Only the dataset's numeral rendering changes.

Provenance and license

  • —Source mixed dataset: https://huggingface.co/datasets/Yujivus/nanochat-climbmix-arithmetic-base10
  • —nanochat: https://github.com/karpathy/nanochat
  • —Upstream ClimbMix: https://huggingface.co/datasets/nvidia/Nemotron-ClimbMix

The source dataset is marked CC BY-NC 4.0, so this derived dataset is conservatively marked the same. Users are responsible for reviewing all upstream terms.