summaryrefslogtreecommitdiff
path: root/bitsandbytes
diff options
context:
space:
mode:
Diffstat (limited to 'bitsandbytes')
-rw-r--r--bitsandbytes/functional.py6
-rw-r--r--bitsandbytes/nn/modules.py1
-rw-r--r--bitsandbytes/optim/adagrad.py3
-rw-r--r--bitsandbytes/optim/adamw.py2
-rw-r--r--bitsandbytes/optim/lars.py6
-rw-r--r--bitsandbytes/optim/rmsprop.py11
-rw-r--r--bitsandbytes/optim/sgd.py6
7 files changed, 14 insertions, 21 deletions
diff --git a/bitsandbytes/functional.py b/bitsandbytes/functional.py
index 44116cc..fbd7564 100644
--- a/bitsandbytes/functional.py
+++ b/bitsandbytes/functional.py
@@ -2,13 +2,13 @@
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
+import ctypes as ct
import os
import random
-import math
-import ctypes as ct
+from typing import Tuple
+
import torch
from torch import Tensor
-from typing import Tuple
lib = ct.cdll.LoadLibrary(os.path.dirname(__file__) + '/libbitsandbytes.so')
name2qmap = {}
diff --git a/bitsandbytes/nn/modules.py b/bitsandbytes/nn/modules.py
index dc0a171..c5460fb 100644
--- a/bitsandbytes/nn/modules.py
+++ b/bitsandbytes/nn/modules.py
@@ -7,7 +7,6 @@ import torch
from typing import Optional
from torch import Tensor
-from torch.nn.parameter import Parameter
import torch.nn.functional as F
from bitsandbytes.optim import GlobalOptimManager
diff --git a/bitsandbytes/optim/adagrad.py b/bitsandbytes/optim/adagrad.py
index 84ade3c..4f51250 100644
--- a/bitsandbytes/optim/adagrad.py
+++ b/bitsandbytes/optim/adagrad.py
@@ -2,11 +2,8 @@
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
-import torch
from bitsandbytes.optim.optimizer import Optimizer1State
-torch.optim.Adagrad
-
class Adagrad(Optimizer1State):
def __init__(self, params, lr=1e-2, lr_decay=0, weight_decay=0, initial_accumulator_value=0, eps=1e-10,
optim_bits=32, args=None, min_8bit_size=4096, percentile_clipping=100, block_wise=True):
diff --git a/bitsandbytes/optim/adamw.py b/bitsandbytes/optim/adamw.py
index 7761f3b..c4f0355 100644
--- a/bitsandbytes/optim/adamw.py
+++ b/bitsandbytes/optim/adamw.py
@@ -2,9 +2,7 @@
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
-import torch
from bitsandbytes.optim.optimizer import Optimizer2State
-import bitsandbytes.functional as F
class AdamW(Optimizer2State):
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8,
diff --git a/bitsandbytes/optim/lars.py b/bitsandbytes/optim/lars.py
index 40dede7..912520d 100644
--- a/bitsandbytes/optim/lars.py
+++ b/bitsandbytes/optim/lars.py
@@ -12,7 +12,7 @@ class LARS(Optimizer1State):
weight_decay=0, nesterov=False, optim_bits=32, args=None,
min_8bit_size=4096, percentile_clipping=100, max_unorm=0.02):
if momentum == 0:
- raise NotImplementError(f'LARS without momentum is not supported!')
+ raise NotImplementedError(f'LARS without momentum is not supported!')
super(LARS, self).__init__('lars', params, lr, (momentum, dampening), 0.0,
weight_decay, optim_bits, args, min_8bit_size, percentile_clipping, max_unorm=max_unorm, block_wise=False)
@@ -21,7 +21,7 @@ class LARS8bit(Optimizer1State):
weight_decay=0, nesterov=False, args=None,
min_8bit_size=4096, percentile_clipping=100, max_unorm=0.02):
if momentum == 0:
- raise NotImplementError(f'LARS without momentum is not supported!')
+ raise NotImplementedError(f'LARS without momentum is not supported!')
super(LARS8bit, self).__init__('lars', params, lr, (momentum, dampening), 0.0,
weight_decay, 8, args, min_8bit_size, percentile_clipping, max_unorm=max_unorm, block_wise=False)
@@ -30,7 +30,7 @@ class LARS32bit(Optimizer1State):
weight_decay=0, nesterov=False, args=None,
min_8bit_size=4096, percentile_clipping=100, max_unorm=0.02):
if momentum == 0:
- raise NotImplementError(f'LARS without momentum is not supported!')
+ raise NotImplementedError(f'LARS without momentum is not supported!')
super(LARS32bit, self).__init__('lars', params, lr, (momentum, dampening), 0.0,
weight_decay, 32, args, min_8bit_size, percentile_clipping, max_unorm=max_unorm, block_wise=False)
diff --git a/bitsandbytes/optim/rmsprop.py b/bitsandbytes/optim/rmsprop.py
index 99b718e..7909d5d 100644
--- a/bitsandbytes/optim/rmsprop.py
+++ b/bitsandbytes/optim/rmsprop.py
@@ -2,16 +2,15 @@
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
-import torch
from bitsandbytes.optim.optimizer import Optimizer1State
class RMSprop(Optimizer1State):
def __init__(self, params, lr=1e-2, alpha=0.99, eps=1e-8, weight_decay=0, momentum=0, centered=False, optim_bits=32, args=None,
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if alpha == 0:
- raise NotImplementError(f'RMSprop with alpha==0.0 is not supported!')
+ raise NotImplementedError(f'RMSprop with alpha==0.0 is not supported!')
if centered:
- raise NotImplementError(f'Centered RMSprop is not supported!')
+ raise NotImplementedError(f'Centered RMSprop is not supported!')
super(RMSprop, self).__init__('rmsprop', params, lr, (alpha, momentum), eps,
weight_decay, optim_bits, args, min_8bit_size, percentile_clipping, block_wise)
@@ -19,9 +18,9 @@ class RMSprop8bit(Optimizer1State):
def __init__(self, params, lr=1e-2, alpha=0.99, eps=1e-8, weight_decay=0, momentum=0, centered=False, args=None,
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if alpha == 0:
- raise NotImplementError(f'RMSprop with alpha==0.0 is not supported!')
+ raise NotImplementedError(f'RMSprop with alpha==0.0 is not supported!')
if centered:
- raise NotImplementError(f'Centered RMSprop is not supported!')
+ raise NotImplementedError(f'Centered RMSprop is not supported!')
super(RMSprop8bit, self).__init__('rmsprop', params, lr, (alpha, momentum), eps,
weight_decay, 8, args, min_8bit_size, percentile_clipping, block_wise)
@@ -30,7 +29,7 @@ class RMSprop32bit(Optimizer1State):
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if alpha == 0:
- raise NotImplementError(f'RMSprop with alpha==0.0 is not supported!')
+ raise NotImplementedError(f'RMSprop with alpha==0.0 is not supported!')
if centered:
raise NotImplementError(f'Centered RMSprop is not supported!')
super(RMSprop32bit, self).__init__('rmsprop', params, lr, (alpha, momentum), eps,
diff --git a/bitsandbytes/optim/sgd.py b/bitsandbytes/optim/sgd.py
index 926d804..0529879 100644
--- a/bitsandbytes/optim/sgd.py
+++ b/bitsandbytes/optim/sgd.py
@@ -9,7 +9,7 @@ class SGD(Optimizer1State):
weight_decay=0, nesterov=False, optim_bits=32, args=None,
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if momentum == 0:
- raise NotImplementError(f'SGD without momentum is not supported!')
+ 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)
@@ -18,7 +18,7 @@ class SGD8bit(Optimizer1State):
weight_decay=0, nesterov=False, args=None,
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if momentum == 0:
- raise NotImplementError(f'SGD without momentum is not supported!')
+ 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)
@@ -27,6 +27,6 @@ class SGD32bit(Optimizer1State):
weight_decay=0, nesterov=False, args=None,
min_8bit_size=4096, percentile_clipping=100, block_wise=True):
if momentum == 0:
- raise NotImplementError(f'SGD without momentum is not supported!')
+ 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)