Weifan Yu comments

Repositories
Issues
Comments

Results 1 comments of


                                            Weifan Yu

Fix: try aligning dtype of matrixes when training with deepspeed and mixed-precision is set to bf16 or fp16

Hi, but when I turn to mixed_precision=bf16, it still arises the [mat1 and mat2 must have the same dtype, but got Float and BFloat16] error. I am running script "flux_train_control_net.py"...