TARA: Tool-Augmented Reward Modeling Dataset for AI Preference Learning

TARA is a novel Tool-Augmented Reward modeling datAset introduced in a research paper and used to train the Themis-7b model. It includes comprehensive comparison data of human preferences and detailed tool invocation processes. The dataset was created by ernie-research and was last updated on the Hugging Face platform in June 2025.

Use Cases

• Training reward models for reinforcement learning based on human preference comparisons.
• Developing or evaluating tool-augmented language models based on detailed tool invocation processes.
• Benchmarking model performance on tasks requiring integration with external tools like calculators and search engines.

Strengths

• Supports multiple tools including Calculator, Code, Translator, Google Search, Calendar, Weather, and WikiSearch.
• Includes both human preference comparisons and detailed tool invocation processes.
• Was used to train a specific model (Themis-7b), indicating practical utility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ernie-research on Hugging Face.

Collection Method

Introduced in a research paper; likely contains human-annotated preference comparisons and tool-use demonstrations.

Freshness

Last updated 2025-06-06 14:37:35; freshness should be verified.