Instruction following data formatted for reinforcement learning with verifiable rewards. Prompts were generated by sampling from the Tulu 2 SFT mixture and adding constraints from IFEval. The dataset was created by AllenAI and last updated on November 21, 2024.
Use Cases
- Training instruction-following models based on prompts with verifiable constraints.
- Benchmarking model performance on tasks with specific, measurable outcomes.
- Research into reinforcement learning techniques for language model alignment.
- Fine-tuning models as part of the Tulu 3 release pipeline.
Strengths
- Data is specifically formatted for use with open-instruct and verifiable reward training.
- Prompts are derived from established sources: the Tulu 2 SFT mixture and IFEval constraints.
- Part of the Tulu 3 release, indicating integration with a known model family.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- AllenAI
- Collection Method
- Prompts generated by sampling from the Tulu 2 SFT mixture and randomly adding constraints from IFEval.
- Time Range
- null
- Freshness
- Last updated 2024-11-21 07:17:40; freshness should be verified.
- Geography
- null