FORMAT: Reasoning Datasets - DeepSeek Format
Collection
Aggregating reasoning datasets (5M) • 20 items • Updated • 1
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 2346359491 bytes, limit is 300000000 bytes
Make sure that
1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code: TooBigContentError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
reasoning-sft-PleIAs-SYNTH-1M
1M English samples subsampled from PleIAs/SYNTH and converted into a clean messages format for SFT/reasoning model training.
Each row has two relevant columns:
input— list of dicts (conversation turns withroleandcontent), ending on the last user turnresponse— string formatted as<think>\n{chain-of-thought}\n</think>\n{final answer}
Sampled to maximise distribution across query_seed_url, model, and exercise axes via round-robin across ~2M candidate rows pulled from all 500 source files.
License
CDLA-Permissive-2.0 — free to use, modify, and republish commercially.
Seed data sourced from Wikipedia under CC-BY-SA 4.0.
Credits
Original dataset: PleIAs/SYNTH by Pleias
Seed data: Wikimedia / Structured Wikipedia
- Downloads last month
- 16