The learning-rate scheduler utility needs to remain usable with the PyTorch versions supported by this…
The learning-rate scheduler utility needs to remain usable with the PyTorch versions supported by this…: a task in MiMo-V2.6-RL-oss: Agentic RL Environments (MiMo RL release). At present, constructing CosineWarmupScheduler can fail because the underlying LambdaLR API no longer accepts the legacy…
The task
At present, constructing CosineWarmupScheduler can fail because the underlying LambdaLR API no longer accepts the legacy verbose keyword, and the warmup/decay arithmetic has undefined cases when warmup is disabled or the configured horizon is inconsistent. Keep the existing…
Part of XiaomiMiMo/MiMo-V2.6-RL-oss.