File size: 40,660 Bytes
fc0f7bd | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 | # Copyright 2019 DeepMind Technologies Limited. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradient transformations."""
import functools
from typing import NamedTuple, Optional
import chex
import jax
from jax import tree_util as jtu
import jax.numpy as jnp
from optax import tree_utils as otu
from optax._src import base
from optax._src import numerics
from optax._src import utils
from optax.transforms import _accumulation
from optax.transforms import _adding
abs_sq = numerics.abs_sq
def _reject_complex(params):
if any(jnp.iscomplexobj(x) for x in jtu.tree_leaves(params)):
raise ValueError('This transformation does not support complex parameters.')
class ScaleByRssState(NamedTuple):
"""State holding the sum of gradient squares to date."""
sum_of_squares: base.Updates
def scale_by_rss(
initial_accumulator_value: float = 0.1,
eps: float = 1e-7
) -> base.GradientTransformation:
"""Rescale updates by the root of the sum of all squared gradients to date.
References:
[Duchi et al, 2011](https://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)
[McMahan et al., 2010](https://arxiv.org/abs/1002.4908)
Args:
initial_accumulator_value: Starting value for accumulators, must be >= 0.
eps: A small floating point value to avoid zero denominator.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
return ScaleByRssState(
sum_of_squares=otu.tree_full_like(params, initial_accumulator_value))
def update_fn(updates, state, params=None):
del params
sum_of_squares = jtu.tree_map(
lambda g, t: abs_sq(g) + t, updates, state.sum_of_squares)
inv_sqrt_g_square = jtu.tree_map(
lambda t: jnp.where(t > 0, jax.lax.rsqrt(t + eps), 0.0), sum_of_squares)
updates = otu.tree_mul(inv_sqrt_g_square, updates)
return updates, ScaleByRssState(sum_of_squares=sum_of_squares)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByRmsState(NamedTuple):
"""State for exponential root mean-squared (RMS)-normalized updates."""
nu: base.Updates
def scale_by_rms(
decay: float = 0.9,
eps: float = 1e-8,
initial_scale: float = 0.
) -> base.GradientTransformation:
r"""Rescale updates by the root of the exp. moving avg of the square.
WARNING: PyTorch and optax's RMSprop implementations differ and could impact
performance. In the denominator, optax uses $\sqrt{v + \epsilon}$ whereas
PyTorch uses $\sqrt{v} + \epsilon$. See
https://github.com/google-deepmind/optax/issues/532 for more detail.
References:
[Hinton](www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)
Args:
decay: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
initial_scale: Initial value for second moment.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
nu = otu.tree_full_like(params, initial_scale) # second moment
return ScaleByRmsState(nu=nu)
def update_fn(updates, state, params=None):
del params
nu = otu.tree_update_moment_per_elem_norm(updates, state.nu, decay, 2)
updates = jtu.tree_map(
lambda g, n: g * jax.lax.rsqrt(n + eps), updates, nu)
return updates, ScaleByRmsState(nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByRStdDevState(NamedTuple):
"""State for centered exponential moving average of squares of updates."""
mu: base.Updates
nu: base.Updates
def scale_by_stddev(
decay: float = 0.9,
eps: float = 1e-8,
initial_scale: float = 0.
) -> base.GradientTransformation:
"""Rescale updates by the root of the centered exp. moving average of squares.
References:
[Hinton](www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf)
Args:
decay: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
initial_scale: Initial value for second moment.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
mu = otu.tree_zeros_like(params) # First moment
nu = otu.tree_full_like(params, initial_scale) # second moment
return ScaleByRStdDevState(mu=mu, nu=nu)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, decay, 1)
nu = otu.tree_update_moment_per_elem_norm(updates, state.nu, decay, 2)
updates = jtu.tree_map(
lambda g, m, n: g * jax.lax.rsqrt(n - abs_sq(m) + eps),
updates, mu, nu)
return updates, ScaleByRStdDevState(mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByAdamState(NamedTuple):
"""State for the Adam algorithm."""
count: chex.Array # shape=(), dtype=jnp.int32.
mu: base.Updates
nu: base.Updates
def scale_by_adam(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-8,
eps_root: float = 0.0,
mu_dtype: Optional[chex.ArrayDType] = None,
*,
nesterov: bool = False
) -> base.GradientTransformation:
"""Rescale updates according to the Adam algorithm.
References:
Kingma et al, `Adam: A Method for Stochastic Optimization
<https://arxiv.org/abs/1412.6980>`_, 2014
Dozat, `Incorporating Nesterov Momentum into Adam
<https://openreview.net/pdf?id=OM0jvwB8jIp57ZJjtNEZ>`_ 2016
.. warning::
PyTorch and optax's adam follow Algorithm 1 of the Kingma
and Ba's Adam paper, if reproducing old results note that TensorFlow
used instead the formulation just before Section 2.1 of the paper.
See https://github.com/deepmind/optax/issues/571 for more detail.
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
eps_root: Term added to the denominator inside the square-root to improve
numerical stability when backpropagating gradients through the rescaling.
mu_dtype: Optional `dtype` to be used for the first order accumulator; if
`None` then the `dtype` is inferred from `params` and `updates`.
nesterov: Whether to use Nesterov momentum. The variant of Adam with
Nesterov momentum is described in [Dozat 2016]
Returns:
A `GradientTransformation` object.
"""
mu_dtype = utils.canonicalize_dtype(mu_dtype)
def init_fn(params):
mu = otu.tree_zeros_like(params, dtype=mu_dtype) # First moment
nu = otu.tree_zeros_like(params) # Second moment
return ScaleByAdamState(count=jnp.zeros([], jnp.int32), mu=mu, nu=nu)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
nu = otu.tree_update_moment_per_elem_norm(updates, state.nu, b2, 2)
count_inc = numerics.safe_int32_increment(state.count)
if nesterov:
mu_hat = jtu.tree_map(
lambda m, g: b1 * m + (1 - b1) * g,
otu.tree_bias_correction(
mu, b1, numerics.safe_int32_increment(count_inc)),
otu.tree_bias_correction(updates, b1, count_inc))
else:
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
# Dozat 2016 https://openreview.net/pdf?id=OM0jvwB8jIp57ZJjtNEZ
# Algorithm 2 further multiplies Adam's standard nu_hat by b2. It is
# unclear why. Other Nadam implementations also omit the extra b2 factor.
nu_hat = otu.tree_bias_correction(nu, b2, count_inc)
updates = jtu.tree_map(
lambda m, v: m / (jnp.sqrt(v + eps_root) + eps), mu_hat, nu_hat)
mu = otu.tree_cast(mu, mu_dtype)
return updates, ScaleByAdamState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByAmsgradState(NamedTuple):
"""State for the AMSGrad algorithm."""
count: chex.Array # shape=(), dtype=jnp.int32.
mu: base.Updates
nu: base.Updates
nu_max: base.Updates
def scale_by_amsgrad(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-8,
eps_root: float = 0.0,
mu_dtype: Optional[chex.ArrayDType] = None,
) -> base.GradientTransformation:
"""Rescale updates according to the AMSGrad algorithm.
References:
[Reddi et al, 2018](https://openreview.net/forum?id=ryQu7f-RZ)
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
eps_root: Term added to the denominator inside the square-root to improve
numerical stability when backpropagating gradients through the rescaling.
mu_dtype: Optional `dtype` to be used for the first order accumulator; if
`None` then the `dtype` is inferred from `params` and `updates`.
Returns:
A `GradientTransformation` object.
"""
mu_dtype = utils.canonicalize_dtype(mu_dtype)
def init_fn(params):
mu = otu.tree_zeros_like(params, dtype=mu_dtype) # First moment
nu = otu.tree_zeros_like(params) # Second moment
nu_max = otu.tree_zeros_like(params)
return ScaleByAmsgradState(
count=jnp.zeros([], jnp.int32),
mu=mu, nu=nu, nu_max=nu_max)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
nu = otu.tree_update_moment_per_elem_norm(updates, state.nu, b2, 2)
count_inc = numerics.safe_int32_increment(state.count)
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
nu_hat = otu.tree_bias_correction(nu, b2, count_inc)
nu_max = jtu.tree_map(jnp.maximum, state.nu_max, nu_hat)
updates = jtu.tree_map(
lambda m, v: m / (jnp.sqrt(v + eps_root) + eps), mu_hat, nu_max)
mu = otu.tree_cast(mu, mu_dtype)
return updates, ScaleByAmsgradState(
count=count_inc,
mu=mu, nu=nu, nu_max=nu_max)
return base.GradientTransformation(init_fn, update_fn)
def scale_by_adamax(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-8
) -> base.GradientTransformation:
"""Rescale updates according to the Adamax algorithm.
References:
[Kingma et al, 2014](https://arxiv.org/abs/1412.6980)
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted maximum of grads.
eps: Term added to the denominator to improve numerical stability.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
mu = otu.tree_zeros_like(params) # First moment
nu = otu.tree_zeros_like(params) # Infinite moment
return ScaleByAdamState(count=jnp.zeros([], jnp.int32), mu=mu, nu=nu)
def update_fn(updates, state, params=None):
del params
count_inc = numerics.safe_int32_increment(state.count)
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
nu = otu.tree_update_infinity_moment(updates, state.nu, b2, eps)
# Bias correction for mean. No bias correction needed for infinity moment.
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
updates = jtu.tree_map(lambda m, v: m / v, mu_hat, nu)
return updates, ScaleByAdamState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByLionState(NamedTuple):
"""State for the Lion algorithm."""
count: chex.Array # shape=(), dtype=jnp.int32.
mu: base.Updates
def scale_by_lion(
b1: float = 0.9,
b2: float = 0.99,
mu_dtype: Optional[chex.ArrayDType] = None,
) -> base.GradientTransformation:
"""Rescale updates according to the Lion algorithm.
References:
[Chen et al, 2023](https://arxiv.org/abs/2302.06675)
Args:
b1: Rate for combining the momentum and the current grad.
b2: Decay rate for the exponentially weighted average of grads.
mu_dtype: Optional `dtype` to be used for the momentum; if
`None` then the `dtype is inferred from `params` and `updates`.
Returns:
A `GradientTransformation` object.
"""
mu_dtype = utils.canonicalize_dtype(mu_dtype)
def init_fn(params):
mu = otu.tree_zeros_like(params, dtype=mu_dtype) # moment
return ScaleByLionState(count=jnp.zeros([], jnp.int32), mu=mu)
def update_fn(updates, state, params=None):
del params
updates_new = jtu.tree_map(
lambda g, m: jnp.sign((1. - b1) * g + b1 * m), updates, state.mu)
mu = otu.tree_update_moment(updates, state.mu, b2, 1)
mu = otu.tree_cast(mu, mu_dtype)
count_inc = numerics.safe_int32_increment(state.count)
return updates_new, ScaleByLionState(count=count_inc, mu=mu)
return base.GradientTransformation(init_fn, update_fn)
def scale(
step_size: float
) -> base.GradientTransformation:
"""Scale updates by some fixed scalar `step_size`.
Args:
step_size: A scalar corresponding to a fixed scaling factor for updates.
Returns:
A `GradientTransformation` object.
"""
def update_fn(updates, state, params=None):
del params
updates = jtu.tree_map(lambda g: step_size * g, updates)
return updates, state
return base.GradientTransformation(base.init_empty_state, update_fn)
def scale_by_param_block_norm(
min_scale: float = 1e-3
) -> base.GradientTransformation:
"""Scale updates for each param block by the norm of that block's parameters.
A `block` is here a weight vector (e.g. in a Linear layer) or a weight matrix
(e.g. in a convolutional layer) appearing as a leaf in the grads/param pytree.
Args:
min_scale: Minimum scaling factor.
Returns:
A `GradientTransformation` object.
"""
def update_fn(updates, state, params):
if params is None:
raise ValueError(base.NO_PARAMS_MSG)
updates = jtu.tree_map(
lambda u, p: u * numerics.safe_norm(p, min_scale),
updates, params)
return updates, state
return base.GradientTransformation(base.init_empty_state, update_fn)
def scale_by_param_block_rms(
min_scale: float = 1e-3
) -> base.GradientTransformation:
"""Scale updates by rms of the gradient for each param vector or matrix.
A `block` is here a weight vector (e.g. in a Linear layer) or a weight matrix
(e.g. in a convolutional layer) appearing as a leaf in the grads/param pytree.
Args:
min_scale: Minimum scaling factor.
Returns:
A `GradientTransformation` object.
"""
def update_fn(updates, state, params):
if params is None:
raise ValueError(base.NO_PARAMS_MSG)
updates = jtu.tree_map(
lambda u, p: u * numerics.safe_root_mean_squares(p, min_scale),
updates, params)
return updates, state
return base.GradientTransformation(base.init_empty_state, update_fn)
class ScaleByAdaDeltaState(NamedTuple):
"""State for the rescaling by Adadelta algoritm."""
e_g: base.Updates
e_x: base.Updates
def scale_by_adadelta(
rho: float = 0.9, eps: float = 1e-6
) -> base.GradientTransformation:
"""Rescale updates according to the Adadelta algorithm.
References:
[Matthew D. Zeiler, 2012](https://arxiv.org/pdf/1212.5701.pdf)
Args:
rho: A coefficient used for computing a running average of squared
gradients.
eps: Term added to the denominator to improve numerical stability.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
e_g = otu.tree_zeros_like(params) # E[squared gradient]
e_x = otu.tree_zeros_like(params) # E[squared update]
return ScaleByAdaDeltaState(e_g=e_g, e_x=e_x)
def update_fn(updates, state, params=None):
del params
e_g = otu.tree_update_moment(updates, state.e_g, rho, 2)
updates = jtu.tree_map(
lambda g, cur_e_g, prev_e_x: (
jnp.sqrt(prev_e_x + eps) / jnp.sqrt(cur_e_g + eps)
)
* g,
updates,
e_g,
state.e_x,
)
e_x = otu.tree_update_moment(updates, state.e_x, rho, 2)
return updates, ScaleByAdaDeltaState(e_g=e_g, e_x=e_x)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByBeliefState(NamedTuple):
"""State for the rescaling by AdaBelief algorithm."""
count: chex.Array # shape=(), dtype=jnp.int32.
mu: base.Updates
nu: base.Updates
def scale_by_belief(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-16,
eps_root: float = 1e-16
) -> base.GradientTransformation:
"""Rescale updates according to the AdaBelief algorithm.
References:
[Zhuang et al, 2020](https://arxiv.org/abs/2010.07468)
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of variance of grads.
eps: Term added to the denominator to improve numerical stability.
eps_root: Term added to the second moment of the prediction error to
improve numerical stability. If backpropagating gradients through the
gradient transformation (e.g. for meta-learning), this must be non-zero.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
mu = otu.tree_zeros_like(params) # First moment
s = otu.tree_zeros_like(params) # Second Central moment
return ScaleByBeliefState(count=jnp.zeros([], jnp.int32), mu=mu, nu=s)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
prediction_error = jtu.tree_map(
lambda g, m: g-m, updates, state.mu)
nu = otu.tree_update_moment_per_elem_norm(prediction_error, state.nu, b2, 2)
nu = jtu.tree_map(lambda v: v + eps_root, nu)
count_inc = numerics.safe_int32_increment(state.count)
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
nu_hat = otu.tree_bias_correction(nu, b2, count_inc)
updates = jtu.tree_map(
lambda m, v: m / (jnp.sqrt(v) + eps), mu_hat, nu_hat)
return updates, ScaleByBeliefState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
def scale_by_yogi(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-3,
eps_root: float = 0.0,
initial_accumulator_value: float = 1e-6
) -> base.GradientTransformation:
"""Rescale updates according to the Yogi algorithm.
Supports complex numbers, see
https://gist.github.com/wdphy16/118aef6fb5f82c49790d7678cf87da29
References:
[Zaheer et al, 2018](https://papers.nips.cc/paper/2018/hash/90365351ccc7437a1309dc64e4db32a3-Abstract.html) #pylint:disable=line-too-long
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of variance of grads.
eps: Term added to the denominator to improve numerical stability.
eps_root: Term added to the denominator inside the square-root to improve
numerical stability when backpropagating gradients through the rescaling.
initial_accumulator_value: The starting value for accumulators.
Only positive values are allowed.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
mu = otu.tree_full_like(params, initial_accumulator_value) # First moment
nu = otu.tree_full_like(params, initial_accumulator_value) # Second moment
return ScaleByAdamState(count=jnp.zeros([], jnp.int32), mu=mu, nu=nu)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
nu = jtu.tree_map(
lambda g, v: v - (1 - b2) * jnp.sign(v - abs_sq(g)) * abs_sq(g),
updates, state.nu)
count_inc = numerics.safe_int32_increment(state.count)
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
nu_hat = otu.tree_bias_correction(nu, b2, count_inc)
updates = jtu.tree_map(
lambda m, v: m / (jnp.sqrt(v + eps_root) + eps), mu_hat, nu_hat)
return updates, ScaleByAdamState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
def scale_by_radam(
b1: float = 0.9,
b2: float = 0.999,
eps: float = 1e-8,
eps_root: float = 0.0,
threshold: float = 5.0,
*,
nesterov: bool = False,
) -> base.GradientTransformation:
"""Rescale updates according to the Rectified Adam algorithm.
References:
[Liu et al, 2020](https://arxiv.org/abs/1908.03265)
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
eps_root: Term added to the denominator inside the square-root to improve
numerical stability when backpropagating gradients through the rescaling.
threshold: Threshold for variance tractability.
nesterov: Whether to use Nesterov momentum.
Returns:
A `GradientTransformation` object.
"""
ro_inf = 2./(1 - b2) - 1
def _radam_update(params):
ro = params[0]
mu_hat = params[1]
nu_hat = params[2]
r = jnp.sqrt((ro - 4)*(ro - 2)*ro_inf/((ro_inf - 4)*(ro_inf - 2)*ro))
updates = jtu.tree_map(
lambda m, v: r*m / (jnp.sqrt(v + eps_root) + eps), mu_hat, nu_hat)
return updates
def init_fn(params):
mu = otu.tree_zeros_like(params) # First moment
nu = otu.tree_zeros_like(params) # Second moment
return ScaleByAdamState(count=jnp.zeros([], jnp.int32), mu=mu, nu=nu)
def update_fn(updates, state, params=None):
del params
mu = otu.tree_update_moment(updates, state.mu, b1, 1)
nu = otu.tree_update_moment_per_elem_norm(updates, state.nu, b2, 2)
count_inc = numerics.safe_int32_increment(state.count)
b2t = b2**count_inc
ro = ro_inf - 2 * count_inc * b2t / (1 - b2t)
if nesterov:
mu_hat = jtu.tree_map(
lambda m, g: b1 * m + (1 - b1) * g,
otu.tree_bias_correction(
mu, b1, numerics.safe_int32_increment(count_inc)),
otu.tree_bias_correction(updates, b1, count_inc))
else:
mu_hat = otu.tree_bias_correction(mu, b1, count_inc)
nu_hat = otu.tree_bias_correction(nu, b2, count_inc)
updates = jax.lax.cond(
ro >= threshold, _radam_update, lambda _: mu_hat,
(ro, mu_hat, nu_hat))
return updates, ScaleByAdamState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByRpropState(NamedTuple):
step_sizes: base.Updates
prev_updates: base.Updates
def scale_by_rprop(
learning_rate: float,
eta_minus: float = 0.5,
eta_plus: float = 1.2,
min_step_size: float = 1e-6,
max_step_size: float = 50.0,
) -> base.GradientTransformation:
"""Scale with the Rprop optimizer.
Rprop, short for resillient backpropogation, is a first order variant of
gradient descent. It responds only to the sign of the gradient by increasing
or decreasing the step size selected per parameter exponentially to speed up
convergence and avoid oscillations.
References:
Riedmiller and Braun. `A direct adaptive method for faster backpropagation
learning: the RPROP algorithm
<https://ieeexplore.ieee.org/document/298623>`_, 1993
Igel and Hüsken. `Empirical evaluation of the improved Rprop learning
algorithms
<https://www.sciencedirect.com/science/article/abs/pii/S0925231201007007>`_,
2003
Args:
learning_rate: The initial step size.
eta_minus: Multiplicative factor for decreasing step size. This is applied
when the gradient changes sign from one step to the next.
eta_plus: Multiplicative factor for increasing step size. This is applied
when the gradient has the same sign from one step to the next.
min_step_size: Minimum allowed step size. Smaller steps will be clipped to
this value.
max_step_size: Maximum allowed step size. Larger steps will be clipped to
this value.
Returns:
The corresponding `GradientTransformation`.
"""
def init_fn(params):
step_sizes = otu.tree_full_like(params, learning_rate)
prev_updates = otu.tree_zeros_like(params)
return ScaleByRpropState(step_sizes, prev_updates)
def update_fn(updates, state, params=None):
del params
sign = jtu.tree_map(
lambda g, prev_g: g * prev_g, updates, state.prev_updates)
step_sizes = jtu.tree_map(
lambda s, step_size: jnp.where(
s == 0,
step_size,
jnp.clip(
step_size * jnp.where(s > 0, eta_plus, eta_minus),
min=min_step_size, max=max_step_size
)
),
sign, state.step_sizes
)
prev_updates = jtu.tree_map(
lambda s, g, step_size: jnp.where(
s < 0, jnp.zeros_like(g), step_size * jnp.sign(g)),
sign, updates, step_sizes)
updates = jtu.tree_map(
lambda s, g, prev_g: jnp.where(s < 0, jnp.zeros_like(prev_g), prev_g),
sign, prev_updates, state.prev_updates)
return updates, ScaleByRpropState(step_sizes, prev_updates)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByScheduleState(NamedTuple):
"""Maintains count for scale scheduling."""
count: chex.Array # shape=(), dtype=jnp.int32
def scale_by_learning_rate(
learning_rate: base.ScalarOrSchedule,
*,
flip_sign: bool = True,
) -> base.GradientTransformation:
"""Scale by the (negative) learning rate (either as scalar or as schedule).
Args:
learning_rate: Can either be a scalar or a schedule (i.e. a callable that
maps an (int) step to a float).
flip_sign: When set to True (the default) this corresponds to scaling by the
negative learning rate.
Returns:
An optax.GradientTransformation that corresponds to multiplying the gradient
with `-learning_rate` (if flip_sign is True) or with `learning_rate` (if
flip_sign is False).
"""
m = -1 if flip_sign else 1
if callable(learning_rate):
return scale_by_schedule(lambda count: m * learning_rate(count))
return scale(m * learning_rate)
def scale_by_schedule(
step_size_fn: base.Schedule
) -> base.GradientTransformation:
"""Scale updates using a custom schedule for the `step_size`.
Args:
step_size_fn: A function that takes an update count as input and proposes
the step_size to multiply the updates by.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
del params
return ScaleByScheduleState(count=jnp.zeros([], jnp.int32))
def update_fn(updates, state, params=None):
del params
step_size = step_size_fn(state.count)
updates = jtu.tree_map(
lambda g: jnp.array(step_size, dtype=g.dtype) * g, updates)
return updates, ScaleByScheduleState(
count=numerics.safe_int32_increment(state.count))
return base.GradientTransformation(init_fn, update_fn)
def scale_by_trust_ratio(
min_norm: float = 0.0,
trust_coefficient: float = 1.,
eps: float = 0.,
) -> base.GradientTransformation:
"""Scale updates by `trust ratio`.
References:
[You et. al 2020](https://arxiv.org/abs/1904.00962)
Args:
min_norm: Minimum norm for params and gradient norms; by default is zero.
trust_coefficient: A multiplier for the trust ratio.
eps: Additive constant added to the denominator for numerical stability.
Returns:
A `GradientTransformation` object.
"""
def update_fn(updates, state, params):
if params is None:
raise ValueError(base.NO_PARAMS_MSG)
def _scale_update(update, param):
# Clip norms to minimum value, by default no clipping.
param_norm = numerics.safe_norm(param, min_norm)
update_norm = numerics.safe_norm(update, min_norm)
trust_ratio = trust_coefficient * param_norm / (update_norm + eps)
# If no minimum norm clipping is used
# Set trust_ratio to 1 in case where parameters would never be updated.
zero_norm = jnp.logical_or(param_norm == 0., update_norm == 0.)
safe_trust_ratio = jnp.where(
zero_norm, jnp.array(1.0, dtype=param.dtype), trust_ratio)
return update * safe_trust_ratio
updates = jtu.tree_map(_scale_update, updates, params)
return updates, state
return base.GradientTransformation(base.init_empty_state, update_fn)
class ApplyEvery(NamedTuple):
"""Contains a counter and a gradient accumulator."""
count: chex.Array
grad_acc: base.Updates
def apply_every(
k: int = 1
) -> base.GradientTransformation:
"""Accumulate gradients and apply them every k steps.
Note that if this transformation is part of a chain, the states of the other
transformations will still be updated at every step. In particular, using
`apply_every` with a batch size of N/2 and k=2 is not necessarily equivalent
to not using `apply_every` with a batch size of N. If this equivalence is
important for you, consider using the `optax.MultiSteps`.
Args:
k: Emit non-zero gradients every k steps, otherwise accumulate them.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
grad_acc = otu.tree_zeros_like(params)
return ApplyEvery(count=jnp.zeros([], jnp.int32), grad_acc=grad_acc)
def update_fn(updates, state, params=None):
del params
c = state.count % k
acc = c != 0
grad_acc = jtu.tree_map(
lambda g, ga: acc * ga + g, updates, state.grad_acc)
emit = c == (k - 1)
updates = jtu.tree_map(lambda ga: emit * ga, grad_acc)
count_inc = numerics.safe_int32_increment(state.count)
return updates, ApplyEvery(count=count_inc % k, grad_acc=grad_acc)
return base.GradientTransformation(init_fn, update_fn)
def _subtract_mean(g):
if len(g.shape) > 1:
return g - g.mean(tuple(range(1, len(g.shape))), keepdims=True)
else:
return g
CentralState = base.EmptyState
def centralize() -> base.GradientTransformation:
"""Centralize gradients.
References:
[Yong et al, 2020](https://arxiv.org/abs/2004.01461)
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
del params
return CentralState()
def update_fn(updates, state, params=None):
del params
updates = jtu.tree_map(_subtract_mean, updates)
return updates, state
return base.GradientTransformation(init_fn, update_fn)
class ScaleBySM3State(NamedTuple):
"""State for the SM3 algorithm."""
mu: base.Updates
nu: base.Updates
def scale_by_sm3(
b1: float = 0.9,
b2: float = 1.0,
eps: float = 1e-8
) -> base.GradientTransformation:
"""Scale updates by `sm3`.
References:
[Anil et. al 2019](https://arxiv.org/abs/1901.11150)
Args:
b1: Decay rate for the exponentially weighted average of grads.
b2: Decay rate for the exponentially weighted average of squared grads.
eps: Term added to the denominator to improve numerical stability.
Returns:
A `GradientTransformation` object.
"""
def zeros_for_dim(p):
return [jnp.zeros([s]) for s in p.shape]
def init_fn(params):
_reject_complex(params)
mu = jtu.tree_map(zeros_for_dim, params)
nu = otu.tree_zeros_like(params)
return ScaleBySM3State(mu, nu)
def _expanded_shape(shape, axis):
# Replaces a `shape` of [M, N, K] with 1 in all dimensions except for i.
# For eg: i = 1 returns [1, N, 1].
rank = len(shape)
return [1] * axis + [shape[axis]] + [1] * (rank - axis - 1)
def _new_accum(g, v):
coeffs = ((1.0 - b2) if b2 != 1.0 else 1.0, b2)
if g.ndim < 2:
return coeffs[0]*g**2 + coeffs[1]*v[0]
else:
return coeffs[0]*g**2 + coeffs[1]*functools.reduce(jnp.minimum, v)
def _new_mu(g, i):
if g.ndim < 2:
return g
else:
return jnp.max(g, axis=other_axes(i, g.ndim))
def other_axes(idx, ndim):
return list(range(idx)) + list(range(idx+1, ndim))
def update_fn(updates, state, params=None):
del params
mu = jtu.tree_map(
lambda g, v: # pylint:disable=g-long-lambda
[jnp.reshape(v[i], _expanded_shape(g.shape, i)) for i in range(g.ndim)],
updates, state.mu)
accum = jtu.tree_map(_new_accum, updates, mu)
accum_inv_sqrt = jtu.tree_map(
lambda t: jnp.where(t > 0, jax.lax.rsqrt(t + eps), 0.0), accum)
up = jtu.tree_map(lambda g, a: g*a, updates, accum_inv_sqrt)
nu = otu.tree_update_moment(up, state.nu, b1, 1)
mu = jtu.tree_map(
lambda g: [_new_mu(g, i) for i in range(g.ndim)], accum)
return nu, ScaleBySM3State(mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByNovogradState(NamedTuple):
"""State for Novograd."""
count: chex.Array
mu: base.Updates
nu: base.Updates
def scale_by_novograd(
b1: float = 0.9,
b2: float = 0.25,
eps: float = 1e-8,
eps_root: float = 0.0,
weight_decay: float = 0.0,
mu_dtype: Optional[chex.ArrayDType] = None,
) -> base.GradientTransformation:
"""Computes NovoGrad updates.
References:
[Ginsburg et al, 2019](https://arxiv.org/abs/1905.11286)
Args:
b1: A decay rate for the exponentially weighted average of grads.
b2: A decay rate for the exponentially weighted average of squared grads.
eps: A term added to the denominator to improve numerical stability.
eps_root: A term added to the denominator inside the square-root to improve
numerical stability when backpropagating gradients through the rescaling.
weight_decay: A scalar weight decay rate.
mu_dtype: An optional `dtype` to be used for the first order accumulator; if
`None` then the `dtype` is inferred from `params` and `updates`.
Returns:
The corresponding `GradientTransformation`.
"""
mu_dtype = utils.canonicalize_dtype(mu_dtype)
def init_fn(params):
mu = otu.tree_zeros_like(params, dtype=mu_dtype) # First moment
nu = jtu.tree_map(lambda _: 0.0, params) # Second moment
return ScaleByNovogradState(count=jnp.zeros([], jnp.int32), mu=mu, nu=nu)
def nu_addition(grads):
return jnp.linalg.norm(grads)**2
def mu_addition(grads, params, nu):
return grads / (jnp.sqrt(nu + eps_root) + eps) + weight_decay * params
def init_nu(grads, nu):
del nu
return jtu.tree_map(nu_addition, grads)
def update_nu(grads, nu):
updates = jtu.tree_map(nu_addition, grads)
return otu.tree_update_moment(updates, nu, b2, 1)
def init_mu(grads, params, mu, nu):
del mu
return jtu.tree_map(mu_addition, grads, params, nu)
def update_mu(grads, params, mu, nu):
updates = jtu.tree_map(mu_addition, grads, params, nu)
return jtu.tree_map(lambda m, u: b1 * m + u, mu, updates)
def update_fn(updates, state, params):
count_inc = numerics.safe_int32_increment(state.count)
nu = jax.lax.cond(
count_inc == 1, init_nu, update_nu, updates, state.nu)
mu = jax.lax.cond(
count_inc == 1, init_mu, update_mu, updates, params, state.mu, nu)
mu = otu.tree_cast(mu, mu_dtype)
updates = mu
return updates, ScaleByNovogradState(count=count_inc, mu=mu, nu=nu)
return base.GradientTransformation(init_fn, update_fn)
def scale_by_optimistic_gradient(
alpha: float = 1.0,
beta: float = 1.0
) -> base.GradientTransformation:
"""Compute generalized optimistic gradients.
References:
[Mokhtari et al, 2019](https://arxiv.org/abs/1901.08511v2)
Args:
alpha: Coefficient for generalized optimistic gradient descent.
beta: Coefficient for negative momentum.
Returns:
A `GradientTransformation` object.
"""
def init_fn(params):
return TraceState(trace=otu.tree_zeros_like(params))
def update_fn(updates, state, params=None):
del params
new_updates = jtu.tree_map(
lambda grad_t, grad_tm1: (alpha + beta) * grad_t - beta * grad_tm1,
updates, state.trace)
return new_updates, TraceState(trace=updates)
return base.GradientTransformation(init_fn, update_fn)
class ScaleByDistanceOverGradientsState(NamedTuple):
"""State for scale_by_distance_over_gradients."""
max_dist: base.OptState
grad_sum_of_squares: base.OptState
init_params: base.OptState
def scale_by_distance_over_gradients(
reps_rel=1e-6, eps=1e-8, param_dtype=jnp.float32, global_scale=1.0
) -> base.GradientTransformation:
"""Distance-over-gradients learning rate-free optimizer.
This implementation stores a single copy of the model parameters, plus two
scalars per parameter array. It is equivalent to "Layer-wise DoG" (LDoG)
in the paper.
The authors recommend using model averaging with this optimizer.
References:
["DoG is SGD's Best Friend: A Parameter-Free Dynamic Step Size
Schedule"](https://arxiv.org/pdf/2302.12022.pdf)
Args:
reps_rel: Used to compute initial learning rate. Recommended values are 1e-4
for models using batch norm, 1e-6 otherwise.
eps: Small loading term to avoid divide-by-zero errors.
param_dtype: dtype for storing initial parameters.
global_scale: Global scale factor, typically 1.0 or -1.0
Returns:
A `GradientTransformation` object.
"""
def _l2(x, y=0.0):
return jnp.sqrt(jnp.square(x - y).sum())
def init_fn(params):
return ScaleByDistanceOverGradientsState(
# Initial distance (needed to prevent zero step sizes).
jtu.tree_map(lambda x: reps_rel * (1 + _l2(x)), params),
# Initial gradient sum-of-squares.
jtu.tree_map(lambda x: jnp.zeros(1), params),
# Initial params, cast to preferred precision.
otu.tree_cast(params, param_dtype),
)
def update_fn(updates, state: ScaleByDistanceOverGradientsState, params):
# update max distance
max_dist = jtu.tree_map(
lambda d, x, y: jnp.maximum(d, _l2(x, y)),
state.max_dist,
params,
state.init_params,
)
# update gradient sum-of-squares
g_sos = jtu.tree_map(
lambda x, y: x + jnp.square(y).sum(), state.grad_sum_of_squares, updates
)
def _tx(g, d, g_sos):
"""Apply the transformation."""
eta = global_scale * (d / jnp.sqrt(g_sos + eps))
return eta * g
updates = jtu.tree_map(_tx, max_dist, g_sos, updates)
# new state
state = ScaleByDistanceOverGradientsState(
max_dist, g_sos, state.init_params
)
return updates, state
return base.GradientTransformation(init_fn, update_fn)
def scale_by_polyak(
f_min: float = 0.0,
max_learning_rate: float = 1.0,
eps: float = 0.0,
) -> base.GradientTransformationExtraArgs:
"""Scales the update by Polyak's step-size."""
def update_fn(
updates: base.Updates,
state: base.EmptyState,
params: Optional[base.Params] = None,
*,
value: float,
**extra_args,
) -> tuple[base.Updates, base.EmptyState]:
"""Scales the update by the Polyak step-size.
Args:
updates: the updates to be scaled.
state: the state of the transformation.
params: the parameters of the model.
value: the value of the loss function.
**extra_args: additional keyword arguments. They are ignored by this
transformation.
Returns:
The scaled updates and the state of the transformation.
"""
del params, extra_args
grad_sq_norm = otu.tree_l2_norm(updates, squared=True)
# avoid division by zero
step = jnp.where(
grad_sq_norm + eps <= jnp.finfo(float).eps,
jnp.array(0.0),
jnp.minimum(
(value - f_min) / (grad_sq_norm + eps), max_learning_rate
),
)
updates = otu.tree_scalar_mul(step, updates)
return updates, state
return base.GradientTransformationExtraArgs(base.init_empty_state, update_fn)
### Legacy symbols to be removed. ###
@functools.partial(
chex.warn_deprecated_function,
replacement='optax.tree_utils.tree_cast')
def cast_tree(
tree: chex.ArrayTree,
dtype: Optional[chex.ArrayDType]
) -> chex.ArrayTree:
return otu.tree_cast(tree, dtype)
trace = _accumulation.trace
TraceState = _accumulation.TraceState
ema = _accumulation.ema
EmaState = _accumulation.EmaState
add_noise = _adding.add_noise
AddNoiseState = _adding.AddNoiseState
add_decayed_weights = _adding.add_decayed_weights
AddDecayedWeightsState = base.EmptyState
ScaleState = base.EmptyState
ScaleByTrustRatioState = base.EmptyState
|