Change --amp flags, no more --apex-amp and --native-amp, add --amp-impl to select apex, and --amp-dtype to allow bfloat16 AMP dtype

train.py

@@ -315,10 +315,10 @@ group.add_argument('--save-images', action='store_true', default=False,
                     help='save images of input bathes every log interval for debugging')
 group.add_argument('--amp', action='store_true', default=False,
                     help='use NVIDIA Apex AMP or Native AMP for mixed precision training')
-group.add_argument('--apex-amp', action='store_true', default=False,
+group.add_argument('--amp-dtype', default='float16', type=str,
-                    help='Use NVIDIA Apex AMP mixed precision')
+                    help='lower precision AMP dtype (default: float16)')
-group.add_argument('--native-amp', action='store_true', default=False,
+group.add_argument('--amp-impl', default='native', type=str,
-                    help='Use Native Torch AMP mixed precision')
+                    help='AMP impl to use, "native" or "apex" (default: native)')
 group.add_argument('--no-ddp-bb', action='store_true', default=False,
                     help='Force broadcast buffers for native DDP to off.')
 group.add_argument('--pin-mem', action='store_true', default=False,
@@ -385,19 +385,18 @@ def main():
 
     # resolve AMP arguments based on PyTorch / Apex availability
     use_amp = None
+    amp_dtype = torch.float16
     if args.amp:
-        # `--amp` chooses native amp before apex (APEX ver not actively maintained)
+        if args.amp_impl == 'apex':
-        if has_native_amp:
+            assert has_apex, 'AMP impl specified as APEX but APEX is not installed.'
-            args.native_amp = True
+            use_amp = 'apex'
-        elif has_apex:
+            assert args.amp_dtype == 'float16'
-            args.apex_amp = True
+        else:
-    if args.apex_amp and has_apex:
+            assert has_native_amp, 'Please update PyTorch to a version with native AMP (or use APEX).'
-        use_amp = 'apex'
+            use_amp = 'native'
-    elif args.native_amp and has_native_amp:
+            assert args.amp_dtype in ('float16', 'bfloat16')
-        use_amp = 'native'
+        if args.amp_dtype == 'bfloat16':
-    elif args.apex_amp or args.native_amp:
+            amp_dtype = torch.bfloat16
-        _logger.warning("Neither APEX or native Torch AMP is available, using float32. "
-                        "Install NVIDA apex or upgrade to PyTorch 1.6")
 
     utils.random_seed(args.seed, args.rank)
 
@@ -484,7 +483,7 @@ def main():
         batch_ratio = global_batch_size / args.lr_base_size
         if not args.lr_base_scale:
             on = args.opt.lower()
-            args.base_scale = 'sqrt' if any([o in on for o in ('ada', 'lamb')]) else 'linear'
+            args.lr_base_scale = 'sqrt' if any([o in on for o in ('ada', 'lamb')]) else 'linear'
         if args.lr_base_scale == 'sqrt':
             batch_ratio = batch_ratio ** 0.5
         args.lr = args.lr_base * batch_ratio
@@ -505,7 +504,7 @@ def main():
         if utils.is_primary(args):
             _logger.info('Using NVIDIA APEX AMP. Training in mixed precision.')
     elif use_amp == 'native':
-        amp_autocast = partial(torch.autocast, device_type=device.type)
+        amp_autocast = partial(torch.autocast, device_type=device.type, dtype=amp_dtype)
         if device.type == 'cuda':
             loss_scaler = NativeScaler()
         if utils.is_primary(args):