😎Nvidia have published a new dataset for training faintune models
HelpSteer2 is an English-language dataset developed by NVIDIA and hosted on the Hugging Face platform. It includes 21,362 rows and is designed to train reward models that help improve the utility, factual accuracy, and coherence of answers generated by large language models (LLMs).
Each row in the dataset contains a query, a response, and five human annotated response attributes:
✅Utility (usefulness)
✅ Correctness
✅ Coherence
✅ Complexity
✅ Verbosity
The dataset can be used to fine-tune LLMs to generate more relevant and better responses to user queries.
Post #745
1.06K