TRM Genesis RealWorld 40K: Fact-Based Q&A with Citations

Within Us AI created a dataset of 40,000 fact-based question-answer pairs. Each answer includes citations and a verification tag, and prompts are optimized for tiny language model contexts. The dataset was last updated on January 4, 2026.

Use Cases

• Training factual question-answering models based on the described fact-based Q&A structure.
• Evaluating model citation and source attribution capabilities based on the included fact_sources field.
• Benchmarking language model performance in constrained-context environments based on the TRM-native prompt design.
• Fine-tuning models for verifiable information retrieval based on the verification tags mentioned in the description.

Strengths

• 40,000 examples provide a substantial corpus for training.
• Each answer includes fact_sources (citations) and a verification tag, adding a layer of verifiability.
• Prompts are described as optimized for tiny contexts, which is a specific design strength.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update date is in the future (2026).

Provenance

Source

Within Us AI

Freshness

Last updated 2026-01-04 00:10:29