File size: 15,539 Bytes
dc9bb20 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 | # 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 Optimizer2State
class AdamW(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
is_paged=False,
):
"""
Base AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
is_paged (`bool`, defaults to `False`):
Whether the optimizer is a paged optimizer or not.
"""
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
optim_bits,
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=is_paged,
)
class AdamW8bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
is_paged=False,
):
"""
8-bit AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
Note: This parameter is not supported in AdamW8bit and must be False.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
Note: This parameter is not used in AdamW8bit as it always uses 8-bit optimization.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
is_paged (`bool`, defaults to `False`):
Whether the optimizer is a paged optimizer or not.
"""
# Validate unsupported parameters
if amsgrad:
raise ValueError("AdamW8bit does not support amsgrad=True")
if optim_bits != 32:
# We allow the default value of 32 to maintain compatibility with the function signature,
# but any other value is invalid since AdamW8bit always uses 8-bit optimization
raise ValueError("AdamW8bit only supports optim_bits=32 (default value for compatibility)")
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
8, # Hardcoded to 8 bits
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=is_paged,
)
class AdamW32bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
is_paged=False,
):
"""
32-bit AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
is_paged (`bool`, defaults to `False`):
Whether the optimizer is a paged optimizer or not.
"""
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
32,
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=is_paged,
)
class PagedAdamW(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
"""
Paged AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
"""
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
optim_bits,
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=True,
)
class PagedAdamW8bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
"""
Paged 8-bit AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
Note: This parameter is not supported in PagedAdamW8bit and must be False.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
Note: This parameter is not used in PagedAdamW8bit as it always uses 8-bit optimization.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
"""
# Validate unsupported parameters
if amsgrad:
raise ValueError("PagedAdamW8bit does not support amsgrad=True")
if optim_bits != 32:
# We allow the default value of 32 to maintain compatibility with the function signature,
# but any other value is invalid since PagedAdamW8bit always uses 8-bit optimization
raise ValueError("PagedAdamW8bit only supports optim_bits=32 (default value for compatibility)")
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
8, # Hardcoded to 8 bits
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=True,
)
class PagedAdamW32bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=1e-2,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
"""
Paged 32-bit AdamW optimizer.
Arguments:
params (`torch.Tensor`):
The input parameters to optimize.
lr (`float`, defaults to 1e-3):
The learning rate.
betas (`tuple(float, float)`, defaults to (0.9, 0.999)):
The beta values are the decay rates of the first and second-order moment of the optimizer.
eps (`float`, defaults to 1e-8):
The epsilon value prevents division by zero in the optimizer.
weight_decay (`float`, defaults to 1e-2):
The weight decay value for the optimizer.
amsgrad (`bool`, defaults to `False`):
Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
optim_bits (`int`, defaults to 32):
The number of bits of the optimizer state.
args (`object`, defaults to `None`):
An object with additional arguments.
min_8bit_size (`int`, defaults to 4096):
The minimum number of elements of the parameter tensors for 8-bit optimization.
percentile_clipping (`int`, defaults to 100):
Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability.
block_wise (`bool`, defaults to `True`):
Whether to independently quantize each block of tensors to reduce outlier effects and improve stability.
"""
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
32,
args,
min_8bit_size,
percentile_clipping,
block_wise,
is_paged=True,
)
|