how2everything/how2mine
Viewer
•
Updated
•
351k
•
6
•
1
Data release for "How2Everything: Mining the Web for How-To Procedures to Evaluate and Improve LLMs"
Note The full 351K step-by-step procedures mined from 980K web documents across 14 topics.
Note 7K evaluation examples sourced from the How2Mine pool, evenly balanced across 14 topics.
Note 102K training examples sourced from the How2Mine pool, fuzzy-deduplicated against How2Bench.
Note WildChat tagged by the OpenAI query type classifier
Note lmsys-chat tagged by the OpenAI query type classifier