Sign in to view source links and access this dataset
Description
DAPO-MATH-17k-oss-reasoning contains reasoning trajectories produced by the gpt-oss-120b model on the DAPO-Math-17k dataset. The dataset includes trajectories generated under three distinct effort levels—Low, Medium, and High—with corresponding average token counts of 1,300, 2,936, and 8,419. It was created by user thuzhizhi and last updated on June 2, 2026.
Use Cases
Analyzing the relationship between reasoning effort and token usage based on the provided Low, Medium, and High effort levels.
Studying keyword frequency as an indicator of model reasoning behavior, based on metrics for terms like 'wait', 'double check', and 'check'.
Benchmarking the efficiency and depth of reasoning steps in language models on mathematical problem-solving tasks.
Strengths
Provides explicit metrics for three distinct reasoning effort levels (Low, Medium, High).
Includes quantitative data on average token usage per effort level (1,300, 2,936, 8,419).
Contains keyword appearance frequency statistics that may indicate model reasoning behavior.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Provenance
Source
thuzhizhi on Hugging Face, derived from BytedTsinghua-SIA/DAPO-Math-17k.
Collection Method
Reasoning trajectories generated by the gpt-oss-120b language model.
Freshness
Last updated 2026-06-02 10:01:40; freshness should be verified.
License is unknown; terms of use must be verified before application.