Dolci-Instruct-SFT: 1K-10K Tool-Use Records for Olmo 3 7B

Between 1,000 and 10,000 text records comprise this tool-use instruction dataset developed by the Allen Institute for AI (AI2) for the Olmo 3 7B Instruct model. Released in late 2025, it functions as a specific subset of the broader Dolci supervised fine-tuning (SFT) collection.

Use Cases

• Fine-tuning LLMs for tool-calling capabilities using the instruction-response sequences
• Benchmarking model adherence to specific API-calling formats
• Researching supervised fine-tuning (SFT) strategies for the Olmo 3 model family

Strengths

• 1,000-10,000 curated records
• Official training data for Olmo 3 7B
• CC BY-SA 4.0 licensed

Limitations

• Small sample size compared to full SFT datasets
• Restricted to tool-use modality only
• Potential bias toward the specific tool-calling format of the Olmo 3 architecture

Provenance

Source

Allen Institute for AI (AI2)

Collection Method

annotated

Freshness

Last updated March 2026.