backbone as FP16, save default to FP32

train.py

                 ckpt = {'epoch': epoch,
                         'best_fitness': best_fitness,
                         'training_results': f.read(),
-                        'model': ema.ema.module.half() if hasattr(model, 'module') else ema.ema.half(),
+                        'model': ema.ema.module if hasattr(model, 'module') else ema.ema,
                         'optimizer': None if final_epoch else optimizer.state_dict()}
 
             # Save last, best and delete

utils/utils.py

 def create_backbone(f='weights/best.pt', s='weights/backbone.pt'):  # from utils.utils import *; create_backbone()
     # create backbone 's' from 'f'
     device = torch.device('cpu')
-    x = torch.load(f, map_location=device)
-    torch.save(x, s)  # update model if SourceChangeWarning
     x = torch.load(s, map_location=device)
 
     x['optimizer'] = None
     x['training_results'] = None
     x['epoch'] = -1
+    x['model'].half()  # to FP16
     for p in x['model'].parameters():
         p.requires_grad = True
     torch.save(x, s)