onnxruntime

28e4c0ed - Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training (#4268)

Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training (#4268) * Keep loss subgraph as FP32 when mixed-p training. * Fix case where there is no white-list loss op. * Get nodes from loss_scale instead of whitelist. * rename const variables. Co-authored-by: Vincent Wang <weicwang@OrtDevTest2v100.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>