redid conversion with --half flag and --to_safetensors flag to export weights in safetensors format in fp16 format to save storage space and vram for training (it exportrd weights in FP32 by defaut so the resulting files were ~13.9 GB in combined size)
