Sign in to view source links and access this dataset
Description
9,437 (task, agent-trajectory) pairs form a cold-start dataset for training AI agents. The dataset is part of the OpenThoughts-Agent open-source effort, which includes datasets, models, and research code. It was last updated on June 9, 2026, by the author 'open-thoughts'.
Use Cases
Teaching a base model agentic behavior and format based on the described trajectory pairs.
Providing a cold-start for reinforcement learning training pipelines as mentioned in the dataset's purpose.
Benchmarking agent training methodologies within the OpenThoughts-Agent framework.
Strengths
Contains 9,437 specific data pairs, providing a defined scale for training.
Explicitly designed for a documented SFT→RL training recipe, indicating a clear purpose.
Part of a broader open-source effort that includes models and code, suggesting potential for reproducibility.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is known but other scale metrics like file size and format are unknown.
Description metadata is limited; actual data quality and structure require manual inspection.
Provenance
Source
open-thoughts (author) on Hugging Face
Collection Method
Curated as part of the OpenThoughts-Agent project, but specific collection methodology is not detailed.
Freshness
Last updated 2026-06-09 10:45:40
License information is unknown, which may restrict commercial or derivative use.