Prm800K: Preference Data for Mathematical Problem Solving

Phase 1 of the PRM800K dataset, processed for preference learning. The dataset includes data from the MATH test set, using a different test split than typical. It was authored by euclaise and last updated on HuggingFace on 2024-01-13.

Use Cases

• Train reward models for mathematical problem-solving based on preference data.
• Fine-tune language models for improved mathematical reasoning based on the MATH test set data.
• Benchmark preference learning algorithms on a domain-specific (mathematics) task.

Strengths

• Focuses on a specific, challenging domain (mathematical problem solving).
• Explicitly includes data from the MATH test set, which is a standard benchmark.
• Processed specifically for preference learning tasks.

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

HuggingFace user euclaise.

Collection Method

Processed from Phase 1 of the PRM800K project.

Freshness

Last updated 2024-01-13 20:45:34; freshness should be verified.