StructEval cleaned thinkless dataset (v08)Merged from v4 and hard-4k; assistant outputs are normalized to + raw structured output.
u-10bei/structured_data_with_cot_dataset_512_v4
daichira/structured-hard-sft-4k
Merge train splits from two source datasets
Remove CoT/preamble/code fences in assistant outputs
Normalize assistant output format to:
{raw structured output}
train
