fairseq/optim/adafactor.py

Survived

Mutant 3235

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -10,7 +10,7 @@
 from . import FairseqOptimizer, register_optimizer
 
 
-@register_optimizer('adafactor')
+@register_optimizer('XXadafactorXX')
 class FairseqAdafactor(FairseqOptimizer):
     def __init__(self, args, params):
         super().__init__(args)

Mutant 3236

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -9,8 +9,6 @@
 
 from . import FairseqOptimizer, register_optimizer
 
-
-@register_optimizer('adafactor')
 class FairseqAdafactor(FairseqOptimizer):
     def __init__(self, args, params):
         super().__init__(args)

Mutant 3237

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -16,7 +16,6 @@
         super().__init__(args)
         self._optimizer = Adafactor(params, **self.optimizer_config)
 
-    @staticmethod
     def add_args(parser):
         """Add optimizer-specific arguments to the parser."""
         # fmt: off

Mutant 3238

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -37,9 +37,7 @@
                                  'otherwise use external learning rate')
         parser.add_argument('--warmup-init', action='store_true',
                             help='use relative step for warm-up learning rate schedule')
-        # fmt: on
-
-    @property
+
     def optimizer_config(self):
         """
         Return a kwarg dictionary that will be used to override optimizer

Mutant 3239

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -98,7 +98,7 @@
 
     def __init__(self, params, lr=None, eps=(1e-30, 1e-3), clip_threshold=1.0,
                  decay_rate=-0.8, beta1=None, weight_decay=0.0, scale_parameter=True,
-                 relative_step=True, warmup_init=False):
+                 relative_step=False, warmup_init=False):
         if lr is not None and relative_step:
             raise ValueError('Cannot combine manual lr and relative_step options')
         if warmup_init and not relative_step:

Mutant 3240

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -98,7 +98,7 @@
 
     def __init__(self, params, lr=None, eps=(1e-30, 1e-3), clip_threshold=1.0,
                  decay_rate=-0.8, beta1=None, weight_decay=0.0, scale_parameter=True,
-                 relative_step=True, warmup_init=False):
+                 relative_step=True, warmup_init=True):
         if lr is not None and relative_step:
             raise ValueError('Cannot combine manual lr and relative_step options')
         if warmup_init and not relative_step:

Mutant 3241

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -109,7 +109,6 @@
                         relative_step=relative_step, warmup_init=warmup_init)
         super(Adafactor, self).__init__(params, defaults)
 
-    @property
     def supports_memory_efficient_fp16(self):
         return True
 

Mutant 3242

--- fairseq/optim/adafactor.py
+++ fairseq/optim/adafactor.py
@@ -113,7 +113,6 @@
     def supports_memory_efficient_fp16(self):
         return True
 
-    @property
     def supports_flat_params(self):
         return False