Dolci Think SFT: 1,000-10,000 Hybrid Tool Use Records for OLMo

This text dataset contains between 1,000 and 10,000 records designed for Supervised Fine-Tuning (SFT) of the OLMo model family, specifically targeting hybrid tool-use capabilities. Developed by the Allen Institute for AI (allenai) and updated in March 2026, it provides instruction-response pairs formatted for tool-augmented reasoning.

Use Cases

• Fine-tuning OLMo models for tool-calling accuracy using instruction-response pairs
• Evaluating hybrid reasoning capabilities in text-based agents
• Benchmarking SFT performance for tool-augmented language models

Strengths

• 1,000 to 10,000 record scale for targeted fine-tuning
• Parquet format for efficient data loading
• Produced by the Allen Institute for AI (allenai)

Limitations

• Small sample size restricted to under 10,000 records
• Specific to the OLMo/Dolci model architecture and reasoning style
• Potential license ambiguity between ODC-By in description and CC-BY-SA-4.0 in tags

Provenance

Source

Allen Institute for AI (allenai)

Freshness

Last updated March 2026.

Geography

United States