deep-prune

Reduce redundant tokens in parallel reasoning by 80% while maintaining accuracy via dynamic pruning of equivalent reasoning paths. Trigger: improve efficiency of consensus-based reasoning (multiple CoT generation).

Independent GitHub snapshot · Automated classification · Original documentation governs