IFDecorator: Synthetic Datasets for Instruction Following RL with Verifiable Rewards

3,625 training and 200 validation examples engineered for Reinforcement Learning with Verifiable Rewards (RLVR). The dataset, created by guox18, contains two complementary synthetic datasets with different synthesis approaches and difficulty distributions. It was last updated on August 8, 2025.

Use Cases

• Training instruction-following agents based on the described high-quality synthetic data.
• Benchmarking RLVR algorithms based on the controlled difficulty distributions mentioned.
• Studying the impact of different synthetic data generation approaches on RL performance.

Strengths

• Contains 3,825 total examples (3,625 training + 200 validation).
• Features controlled difficulty distributions as described.
• Comprises two complementary datasets with different synthesis approaches.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface user guox18

Collection Method

Synthetic data engineering for RLVR, as described.

Freshness

Last updated 2025-08-08 08:46:53.