Sign in to view source links and access this dataset
Description
PatronusAI's World Model Corpus contains generated trajectories for text-based world modeling tasks. The dataset includes trajectories from nine distinct environments: Tau2Bench, SWE-Smith, DeepresearchQA, Openresearcher, Gorilla/BFCLv4, Webshop, Toolathlon, Pandora, and Coderforge. It was created to support the paper "Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL".
Use Cases
Training text-based world models based on the generated trajectory data.
Benchmarking agentic RL algorithms across the nine distinct environments mentioned.
Analyzing synthetic agent behavior in text-based environments like Webshop or Coderforge.
Fine-tuning language models for task-specific reasoning using the structured trajectories.
Strengths
Dataset is explicitly linked to a specific research paper, providing clear context.
Covers trajectories from nine distinct text-based environments, suggesting diversity in task types.
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
PatronusAI
Collection Method
Generated trajectories shaped for text-based world modeling tasks.
Freshness
Last updated 2026-05-18 23:02:21; freshness should be verified.
License is unknown; users should verify terms before use.