pytorch

e2f75d63 - Decomposition - batch_norm, save_mean and save_variance always float32 (#84013)

Decomposition - batch_norm, save_mean and save_variance always float32 (#84013) AMP error shown here - https://github.com/pytorch/torchdynamo/issues/835 Test missing Pull Request resolved: https://github.com/pytorch/pytorch/pull/84013 Approved by: https://github.com/ezyang