YYYYYYibo/DAPO-Math-17k-Qwen3-235B-A22B-Thinking-2507-non-thinking-dedup
DAPO Math Qwen3-235B non-thinking, deduplicated This dataset is derived from Yang-Zhou/DAPO-Math-17k-Qwen3-235B-A22B-Thinking-2507-rejection-distill, specifically dapo_distill_boxed_non_thinking.json. It retains the original LLaMA-Factory-compatible instruction, input, and output columns. Duplicate rows are identified by collapsing consecutive whitespace in instruction, trimming leading/trailing whitespace, and hashing the normalized instruction. The first row in each duplicate… See the full description on the dataset page: https://huggingface.co/datasets/YYYYYYibo/DAPO-Math-17k-Qwen3-235B-A22B-Thinking-2507-non-thinking-dedup.
DAPO Math Qwen3-235B non-thinking, deduplicated
This dataset is derived from `Yang-Zhou/DAPO-Math-17k-Qwen3-235B-A22B-Thinking-2507-rejection-distill`, specifically dapo_distill_boxed_non_thinking.json.
It retains the original LLaMA-Factory-compatible instruction, input, and output columns. Duplicate rows are identified by collapsing consecutive whitespace in instruction, trimming leading/trailing whitespace, and hashing the normalized instruction. The first row in each duplicate group is retained.
Counts
- Source rows: 15,129
- Retained rows: 15,069
- Removed duplicate rows: 60
The Qwen thinking trace has already been removed by the upstream authors. See dedup_stats.json for machine-readable provenance and processing details.