Hi authors. Thanks for sharing this work.
I noticed the default temperature for both the agent and user simulator is 0 in the codebase. But the paper reports Pass^k curves. Could you provide what temperature was used for the agent model and user simulator respectively in your τ-bench experiments?
Hi authors. Thanks for sharing this work.
I noticed the default temperature for both the agent and user simulator is 0 in the codebase. But the paper reports Pass^k curves. Could you provide what temperature was used for the agent model and user simulator respectively in your τ-bench experiments?