Datasets:

CrowdMind
/

pretrain-tasks

pretty_name: Merged Text Splits Clean
tags:
  - text
  - merged-splits
configs:
  - config_name: default
    default: true
    data_files:
      - split: n1infinibyte
        path: data/n1infinibyte/*.parquet
      - split: n1math
        path: data/n1math/*.parquet
      - split: n1rqa
        path: data/n1rqa/*.parquet
      - split: n1stemsft
        path: data/n1stemsft/*.parquet
      - split: n1scicoding
        path: data/n1scicoding/*.parquet
      - split: n1wiki
        path: data/n1wiki/*.parquet
      - split: n11code
        path: data/n11code/*.parquet
      - split: n11econ
        path: data/n11econ/*.parquet
      - split: n11logic
        path: data/n11logic/*.parquet
      - split: n11mc
        path: data/n11mc/*.parquet
      - split: n11algo
        path: data/n11algo/*.parquet
      - split: n12facts
        path: data/n12facts/*.parquet
      - split: n12gen
        path: data/n12gen/*.parquet
      - split: n12moral
        path: data/n12moral/*.parquet
      - split: n12mc
        path: data/n12mc/*.parquet
      - split: behavscience
        path: data/behavscience/*.parquet
      - split: behavformat
        path: data/behavformat/*.parquet
      - split: behavgames
        path: data/behavgames/*.parquet
      - split: behavgeneral
        path: data/behavgeneral/*.parquet
      - split: behavother
        path: data/behavother/*.parquet
      - split: behavplanning
        path: data/behavplanning/*.parquet
      - split: behavreasoning
        path: data/behavreasoning/*.parquet
      - split: bashcode
        path: data/bashcode/*.parquet
dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: n1infinibyte
      num_examples: 10300
    - name: n1math
      num_examples: 24
    - name: n1rqa
      num_examples: 13
    - name: n1stemsft
      num_examples: 27
    - name: n1scicoding
      num_examples: 39
    - name: n1wiki
      num_examples: 63
    - name: n11code
      num_examples: 15225409
    - name: n11econ
      num_examples: 37300
    - name: n11logic
      num_examples: 490648
    - name: n11mc
      num_examples: 60
    - name: n11algo
      num_examples: 181235
    - name: n12facts
      num_examples: 27100
    - name: n12gen
      num_examples: 60
    - name: n12moral
      num_examples: 39017
    - name: n12mc
      num_examples: 100
    - name: behavscience
      num_examples: 13
    - name: behavformat
      num_examples: 40
    - name: behavgames
      num_examples: 13
    - name: behavgeneral
      num_examples: 14
    - name: behavother
      num_examples: 18100
    - name: behavplanning
      num_examples: 14
    - name: behavreasoning
      num_examples: 13
    - name: bashcode
      num_examples: 342717

Merged Text Splits Clean

A single text dataset containing 23 source subsets. Each subset has a short, lowercase alphanumeric split name and a consistent text column.

Load a subset

from datasets import load_dataset

ds = load_dataset("CrowdMind/merged-text-splits-v2", split="n11code")
print(ds[0]["text"])

Use any split name from the table below. For example, n1math loads the v1 math textbooks subset, n11code loads the v1.1 code concepts subset, and bashcode loads the Bash code subset.

Subsets

Split	Source dataset	Source config	Rows
`n1infinibyte`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-InfiniByte-Reasoning`	10,300
`n1math`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-Math-Textbooks`	24
`n1rqa`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-RQA`	13
`n1stemsft`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-STEM-SFT`	27
`n1scicoding`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-Scientific-Coding`	39
`n1wiki`	`nvidia/Nemotron-Pretraining-Specialized-v1`	`Nemotron-Pretraining-Wiki-Rewrite`	63
`n11code`	`nvidia/Nemotron-Pretraining-Specialized-v1.1`	`Nemotron-Pretraining-Code-Concepts`	15,225,409
`n11econ`	`nvidia/Nemotron-Pretraining-Specialized-v1.1`	`Nemotron-Pretraining-Economics`	37,300
`n11logic`	`nvidia/Nemotron-Pretraining-Specialized-v1.1`	`Nemotron-Pretraining-Formal-Logic`	490,648
`n11mc`	`nvidia/Nemotron-Pretraining-Specialized-v1.1`	`Nemotron-Pretraining-Multiple-Choice`	60
`n11algo`	`nvidia/Nemotron-Pretraining-Specialized-v1.1`	`Nemotron-Pretraining-Unconditional-Algorithmic`	181,235
`n12facts`	`nvidia/Nemotron-Pretraining-Specialized-v1.2`	`Nemotron-Pretraining-Fact-Seeking`	27,100
`n12gen`	`nvidia/Nemotron-Pretraining-Specialized-v1.2`	`Nemotron-Pretraining-Generative`	60
`n12moral`	`nvidia/Nemotron-Pretraining-Specialized-v1.2`	`Nemotron-Pretraining-Moral-Scenarios`	39,017
`n12mc`	`nvidia/Nemotron-Pretraining-Specialized-v1.2`	`Nemotron-Pretraining-Multiple-Choice`	100
`behavscience`	`IFM/Pretrain-Behaviors`	`data-science`	13
`behavformat`	`IFM/Pretrain-Behaviors`	`format-rewrites`	40
`behavgames`	`IFM/Pretrain-Behaviors`	`games`	13
`behavgeneral`	`IFM/Pretrain-Behaviors`	`general`	14
`behavother`	`IFM/Pretrain-Behaviors`	`other`	18,100
`behavplanning`	`IFM/Pretrain-Behaviors`	`planning`	14
`behavreasoning`	`IFM/Pretrain-Behaviors`	`reasoning`	13
`bashcode`	`GunA-SD/bash_code`	`default`	342,717

Total: 16,372,319 rows across 23 subsets, stored in 345 Parquet shards.

Data format

Every subset has one text column of type string.
Split names use lowercase letters and digits only.
Large subsets are divided into numbered Parquet shards.
manifest.json records source datasets, source configs, row counts, shard paths, and import warnings.

Source coverage

This repository republishes the files present in CrowdMind/merged-text-splits with shorter split names. It does not reconstruct missing rows from the upstream datasets. The source manifest reports viewer rate limits, preview-only imports, and other coverage limits for some upstream subsets; consult manifest.json before treating a subset as complete relative to its upstream source.

Sources

The source datasets retain their own licenses and terms. Review them before use or redistribution.