CAEL Verified Reasoning dataset released with two million examples

discuss.huggingface.co1 min read

An AI developer has released CAEL Verified Reasoning, a dataset containing two million synthetic examples designed for fine-tuning small language models across eight different areas. The collection includes worked solutions, a Python answer checker, and roughly 37,000 examples specifically addressing missing information, contradictory premises, or unsupplied facts. The author hopes to test whether model fine-tuning remains effective when questions are generated by an external source. Open article →