summaryrefslogtreecommitdiff
path: root/bitsandbytes/optim/sgd.py
diff options
context:
space:
mode:
authorTitus von Koeller <titus@vonkoeller.com>2022-08-01 03:31:48 -0700
committerTitus von Koeller <titus@vonkoeller.com>2022-08-01 03:31:48 -0700
commitbfa0e33294f2b1dc25e65a33be2397f989824298 (patch)
tree396b5d722fdd79da068882ca7376e3636fcb3bb8 /bitsandbytes/optim/sgd.py
parent597a8521b29e90958c31e47421016494da998648 (diff)
ran black and isort for coherent code formatting
Diffstat (limited to 'bitsandbytes/optim/sgd.py')
-rw-r--r--bitsandbytes/optim/sgd.py109
1 files changed, 88 insertions, 21 deletions
diff --git a/bitsandbytes/optim/sgd.py b/bitsandbytes/optim/sgd.py
index 0529879..f7b8934 100644
--- a/bitsandbytes/optim/sgd.py
+++ b/bitsandbytes/optim/sgd.py
@@ -1,32 +1,99 @@
-# Copyright (c) Facebook, Inc. and its affiliates.
-#
-# This source code is licensed under the MIT license found in the
+# Copyright (c) Facebook, Inc. and its affiliates.
+#
+# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
from bitsandbytes.optim.optimizer import Optimizer1State
+
class SGD(Optimizer1State):
- def __init__(self, params, lr, momentum=0, dampening=0,
- weight_decay=0, nesterov=False, optim_bits=32, args=None,
- min_8bit_size=4096, percentile_clipping=100, block_wise=True):
+ def __init__(
+ self,
+ params,
+ lr,
+ momentum=0,
+ dampening=0,
+ weight_decay=0,
+ nesterov=False,
+ optim_bits=32,
+ args=None,
+ min_8bit_size=4096,
+ percentile_clipping=100,
+ block_wise=True,
+ ):
if momentum == 0:
- raise NotImplementedError(f'SGD without momentum is not supported!')
- super(SGD, self).__init__('momentum', params, lr, (momentum, dampening), 0.0,
- weight_decay, optim_bits, args, min_8bit_size, percentile_clipping, block_wise)
+ raise NotImplementedError(f"SGD without momentum is not supported!")
+ super(SGD, self).__init__(
+ "momentum",
+ params,
+ lr,
+ (momentum, dampening),
+ 0.0,
+ weight_decay,
+ optim_bits,
+ args,
+ min_8bit_size,
+ percentile_clipping,
+ block_wise,
+ )
+
class SGD8bit(Optimizer1State):
- def __init__(self, params, lr, momentum=0, dampening=0,
- weight_decay=0, nesterov=False, args=None,
- min_8bit_size=4096, percentile_clipping=100, block_wise=True):
+ def __init__(
+ self,
+ params,
+ lr,
+ momentum=0,
+ dampening=0,
+ weight_decay=0,
+ nesterov=False,
+ args=None,
+ min_8bit_size=4096,
+ percentile_clipping=100,
+ block_wise=True,
+ ):
if momentum == 0:
- raise NotImplementedError(f'SGD without momentum is not supported!')
- super(SGD8bit, self).__init__('momentum', params, lr, (momentum, dampening), 0.0,
- weight_decay, 8, args, min_8bit_size, percentile_clipping, block_wise)
+ raise NotImplementedError(f"SGD without momentum is not supported!")
+ super(SGD8bit, self).__init__(
+ "momentum",
+ params,
+ lr,
+ (momentum, dampening),
+ 0.0,
+ weight_decay,
+ 8,
+ args,
+ min_8bit_size,
+ percentile_clipping,
+ block_wise,
+ )
+
class SGD32bit(Optimizer1State):
- def __init__(self, params, lr, momentum=0, dampening=0,
- weight_decay=0, nesterov=False, args=None,
- min_8bit_size=4096, percentile_clipping=100, block_wise=True):
+ def __init__(
+ self,
+ params,
+ lr,
+ momentum=0,
+ dampening=0,
+ weight_decay=0,
+ nesterov=False,
+ args=None,
+ min_8bit_size=4096,
+ percentile_clipping=100,
+ block_wise=True,
+ ):
if momentum == 0:
- raise NotImplementedError(f'SGD without momentum is not supported!')
- super(SGD32bit, self).__init__('momentum', params, lr, (momentum, dampening), 0.0,
- weight_decay, 32, args, min_8bit_size, percentile_clipping, block_wise)
+ raise NotImplementedError(f"SGD without momentum is not supported!")
+ super(SGD32bit, self).__init__(
+ "momentum",
+ params,
+ lr,
+ (momentum, dampening),
+ 0.0,
+ weight_decay,
+ 32,
+ args,
+ min_8bit_size,
+ percentile_clipping,
+ block_wise,
+ )