HF RL Explorer

I want a ReporterList(reporters: list) class for training loops that keeps metric state across a sequence…

I want a ReporterList(reporters: list) class for training loops that keeps metric state across a sequence…: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). For scalar or tensor batch averages, is averaged=True should weight each value by the current batch size…

The task

For scalar or tensor batch averages, `is_averaged=True` should weight each value by the current batch size before reducing, so adding loss values `[5, 20, 30, 1, 4, 8]` with batch sizes `[32, 32, 32, 32, 32, 10]` and then calling `report()` should put `(values * batches).sum() / batches.sum()` at…

Part of FineEnvs/MiMo-V2.6-RL-harbor-code.