KnowRL-KP-Annotations: Knowledge Point Labels for Math Reasoning Benchmarks

KnowRL-KP-Annotations augments existing mathematical reasoning benchmarks like AIME and AMC with fine-grained Knowledge Point annotations. The dataset, created by HasuerYu, serves as a companion evaluation resource for the paper 'KnowRL: Boosting LLM Reasoning via Reinforcement Learning with Minimal-Sufficient Knowledge Guidance'. It was last updated on April 15, 2026.

Use Cases

• Evaluating reinforcement learning policies for knowledge point selection based on the annotated KP subsets.
• Training or fine-tuning LLMs on mathematical reasoning with structured knowledge guidance.
• Benchmarking the performance of reasoning models against the annotated AIME and AMC problems.
• Analyzing the relationship between specific knowledge points and problem-solving success in math benchmarks.

Strengths

• Provides annotations for established mathematical reasoning benchmarks like AIME and AMC.
• Includes KP subset selections generated under different reinforcement learning policies.
• Last updated on 2026-04-15, suggesting recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset does not introduce new problems, only enriches existing ones.

Provenance

Source

HasuerYu on Hugging Face

Collection Method

Enrichment of existing benchmark examples with Knowledge Point annotations.

Freshness

Last updated 2026-04-15 14:51:40