mindbomber/

aana-external-validity-hermes-head-to-head

This artifact reruns the AANA head-to-head agent-action comparisons on a second independent public tool-call source: NousResearch/hermes-function-calling-v1 The goal is external validity, not an official leaderboard score. The source dataset provides function-calling conversations and tool schemas, but not human-reviewed safety labels. Safety labels here are policy-derived by the included transform script:

Original Hermes… See the full description on the dataset page: https://huggingface.co/datasets/mindbomber/aana-external-validity-hermes-head-to-head.