Orchard: Agentic Modeling Trajectories from Microsoft

A 2026 release of agentic modeling trajectories accompanying the paper 'Orchard: An Open-Source Agentic Modeling Framework' by Peng et al. The dataset bundles two parallel datasets distilled from strong teacher models, as described by Microsoft. It was last updated on Hugging Face in May 2026, though the release is currently on hold pending a re-upload.

Use Cases

• Training agentic models based on the distilled trajectory data.
• Benchmarking agentic AI frameworks against the provided parallel datasets.
• Analyzing the behavior and performance of teacher-student distillation in agentic systems.
• Researching trajectory-based learning and planning in AI agents.

Strengths

• Dataset is associated with a 2026 academic paper from Microsoft, suggesting research relevance.
• Contains two parallel datasets distilled from strong teacher models, as mentioned in the description.

Limitations

• The release is on hold; the current schema and contents may not be final, requiring caution for integration.
• Row count, column definitions, and file formats are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Microsoft

Collection Method

Distilled from strong teacher models, as described in the accompanying paper.

Freshness

Last updated 2026-05-14 20:55:42