Skywork-OR1-RL-Data: 100K-1M RL Problems with 0-16 Difficulty Levels

Skywork-OR1-RL-Data is a reinforcement learning training dataset containing between 100,000 and 1,000,000 text records released by Skywork in April 2025. The collection features problems categorized by difficulty levels ranging from 0 to 16, calibrated against specific DeepSeek-R1-Distill-Qwen model variants.

Use Cases

• RL training of LLMs using the 0-16 difficulty level labels to curate curriculum learning
• Benchmarking model distillation by comparing performance against the DeepSeek-R1-Distill-Qwen difficulty baselines
• Analyzing model failure modes on specific difficulty tiers identified in the metadata

Strengths

• 100,000 to 1,000,000 records
• Granular difficulty scoring on a 0-16 scale
• Calibrated against specific DeepSeek-R1-Distill-Qwen model variants

Limitations

• Difficulty scores are model-dependent and may not reflect absolute difficulty for non-Qwen architectures
• Intentional exclusion of extreme difficulty levels (0 and 16) creates a truncated dataset distribution

Provenance

Source

Skywork

Collection Method

Curated problems filtered by model-based difficulty assessment relative to DeepSeek-R1-Distill-Qwen

Freshness

Last updated May 29, 2025.