53,000 training examples filtered from DeepScaleR-Preview-Dataset and AReal-boba-Data for Polaris Preview models. Each row contains a problem, its answer, and a difficulty score estimated by a language model. The dataset was created by POLARIS-Project and last updated on June 18, 2025.
Use Cases
- Training or fine-tuning reasoning models based on the provided problem-answer pairs.
- Benchmarking model performance across difficulty levels based on the estimated pass rate.
- Analyzing the relationship between problem difficulty and model success based on the difficulty score feature.
Strengths
- 53,000 problem-answer pairs for training.
- Includes a difficulty score (pass rate) for each problem, estimated by a specific model (Deepseek-R1-distill-Qwen-7B).
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- POLARIS-Project
- Collection Method
- Filtered from DeepScaleR-Preview-Dataset and AReal-boba-Data.
- Time Range
- null
- Freshness
- Last updated 2025-06-18 08:24:32; freshness should be verified.
- Geography
- null