k-10-12-mixed

Mixed K-10/K-12 pretraining corpus in a single parquet file, assembled by round-robin interleaving of the 14 source datasets in two phases.

Rows: 489,956 (rows 0–179,997 = Phase-1, rows 179,998–489,955 = Phase-2) Fields: text (string), subject (string), grade (int), word_count (int)

Mixing procedure

Rows are taken 10 at a time from each dataset in round-robin order. When a dataset is exhausted it is skipped, and the pass continues with the remaining datasets until every dataset in the phase is fully drained. Phase-1 is completed before Phase-2 begins.

Phase-1 (rows 0–179,997), in round-robin order

Dataset	Rows
k-10-economics-30k	30,000
k-10-english-10k	10,000
k-10-english-short-stories-40k	39,998
k-10-general-knowledge-50k	50,000
k-10-geography-20k	20,000
k-10-history-civics-10k	10,000
k-10-maths-10k	10,000
k-10-science-10k	10,000

Phase-2 (rows 179,998–489,955), in round-robin order

Dataset	Rows
k-10-aptitude-reasoning-50k	49,999
k-10-business-commerce-40k	40,000
k-10-computer-python-40k	39,996
k-10-maths-arithmetic-60k	59,986
k-10-science-2-60k	60,000
k-12-biology-60k	59,977

All source content is conserved exactly: every row of every source dataset appears exactly once.

Usage

from datasets import load_dataset
ds = load_dataset("sujitpandey/k-10-12-mixed", split="train")