I want airdialogue prepro to support an alternate inference input mode for dialogue data.

I want airdialogue prepro to support an alternate inference input mode for dialogue data.: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). It should keep the existing --keep incorrect behavior: by default skip records whose JSON has correct sample: false , and…