Hi guys, I’m creating a Crew (NLP Analyst) based on the deeplearning.ai course, creating a Score for “intents” that are created in my platform. The pain point here is how to keep the consistency of the Crew because running the same input each time results in a different score. Do you have any tips about that?
Any luck on this. Having same issue here. I’m also working on creating an agent which can generate lead score. Interestingly, when I directly provide the prompt (emitted within crewai verbose logs), to LLM (via ollama), I get more consistent results. However, when using same llm, same prompt data to Crewai, it generate different responses across multiple runs of crewai agents.
Additional info:
LLM : Phi4
Temperature: 0.1
Agent count: 1
Tools: None