oumi-synthetic-claims is a text dataset designed to fine-tune language models for Claim Verification.
Prompts and responses were produced synthetically from Llama-3.1-405B-Instruct.
oumi-synthetic-claims was used to train HallOumi-8B, which achieves 77.2% Macro F1, outperforming SOTA models such as Claude Sonnet 3.5, OpenAI o1, etc.
Curated by: Oumi AI using Oumi inference
Language(s) (NLP): English
License: Llama 3.1 Community License… See the full description on the dataset page:
https://huggingface.co/datasets/oumi-ai/oumi-synthetic-claims.
