Tracing LLM calls wastes compute by re-running entire workflows

Original: @yoheinakajima super helpful lol. So many times i have used one of those tracing / eval setups and every single llm call is basically a rollout upto that point , and it just drives me crazy

Source: x.com ↗