Datasets:

JWei05
/

DAPO-17.4k

Modalities:

Text

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (2)

train · 16.4k rows

The dataset viewer is not available for this split.

Couldn't cast array of type list<element: struct<content: string, role: string>> to {'content': List(Value('string')), 'role': List(Value('string'))}

Error code:   UnexpectedError

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.

DAPO-17.4k

This dataset contains the 17,398-row DAPO math set in the RL parquet format used by verl.

Each prompt asks the model to return the final answer in \boxed{}:

Please output the final answer within \boxed{}.

Splits

train: 16,398 examples
validation: 1,000 examples

The validation set was sampled uniformly at random from dapo-math-17k.parquet with seed 42; the train split contains the remaining examples. The splits have no overlap by extra_info.index.

Schema

data_source: source label, currently math
prompt: single-turn user prompt with the boxed-answer instruction
reward_model: rule reward metadata with ground_truth
extra_info: original question, stable index, and split label

Downloads last month: 26