I'm seeing a weird imbalance with Sarama's round-robin consumer group balancing: if I have several topics…

I'm seeing a weird imbalance with Sarama's round-robin consumer group balancing: if I have several topics…: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). What the agent is asked, how it is graded, what it runs in, and its files.