File size: 48,032 Bytes
9c6594c |
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 |
# Copyright 2022 The HuggingFace Team. 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.
# Expectation:
# Provide a project dir name, then each type of logger gets stored in project/{`logging_dir`}
import json
import os
import time
from functools import wraps
from typing import Any, Optional, Union
import yaml
from packaging import version
from .logging import get_logger
from .state import PartialState
from .utils import (
LoggerType,
compare_versions,
is_aim_available,
is_clearml_available,
is_comet_ml_available,
is_dvclive_available,
is_mlflow_available,
is_swanlab_available,
is_tensorboard_available,
is_wandb_available,
listify,
)
_available_trackers = []
if is_tensorboard_available():
_available_trackers.append(LoggerType.TENSORBOARD)
if is_wandb_available():
_available_trackers.append(LoggerType.WANDB)
if is_comet_ml_available():
_available_trackers.append(LoggerType.COMETML)
if is_aim_available():
_available_trackers.append(LoggerType.AIM)
if is_mlflow_available():
_available_trackers.append(LoggerType.MLFLOW)
if is_clearml_available():
_available_trackers.append(LoggerType.CLEARML)
if is_dvclive_available():
_available_trackers.append(LoggerType.DVCLIVE)
if is_swanlab_available():
_available_trackers.append(LoggerType.SWANLAB)
logger = get_logger(__name__)
def on_main_process(function):
"""
Decorator to selectively run the decorated function on the main process only based on the `main_process_only`
attribute in a class.
Checks at function execution rather than initialization time, not triggering the initialization of the
`PartialState`.
"""
@wraps(function)
def execute_on_main_process(self, *args, **kwargs):
if getattr(self, "main_process_only", False):
return PartialState().on_main_process(function)(self, *args, **kwargs)
else:
return function(self, *args, **kwargs)
return execute_on_main_process
def get_available_trackers():
"Returns a list of all supported available trackers in the system"
return _available_trackers
class GeneralTracker:
"""
A base Tracker class to be used for all logging integration implementations.
Each function should take in `**kwargs` that will automatically be passed in from a base dictionary provided to
[`Accelerator`].
Should implement `name`, `requires_logging_directory`, and `tracker` properties such that:
`name` (`str`): String representation of the tracker class name, such as "TensorBoard" `requires_logging_directory`
(`bool`): Whether the logger requires a directory to store their logs. `tracker` (`object`): Should return internal
tracking mechanism used by a tracker class (such as the `run` for wandb)
Implementations can also include a `main_process_only` (`bool`) attribute to toggle if relevent logging, init, and
other functions should occur on the main process or across all processes (by default will use `True`)
"""
main_process_only = True
def __init__(self, _blank=False):
if not _blank:
err = ""
if not hasattr(self, "name"):
err += "`name`"
if not hasattr(self, "requires_logging_directory"):
if len(err) > 0:
err += ", "
err += "`requires_logging_directory`"
# as tracker is a @property that relies on post-init
if "tracker" not in dir(self):
if len(err) > 0:
err += ", "
err += "`tracker`"
if len(err) > 0:
raise NotImplementedError(
f"The implementation for this tracker class is missing the following "
f"required attributes. Please define them in the class definition: "
f"{err}"
)
def start(self):
"""
Lazy initialization of the tracker inside Accelerator to avoid initializing PartialState before
InitProcessGroupKwargs.
"""
pass
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Implementations should use the experiment configuration
functionality of a tracking API.
Args:
values (Dictionary `str` to `bool`, `str`, `float` or `int`):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, `int`, or `None`.
"""
pass
def log(self, values: dict, step: Optional[int], **kwargs):
"""
Logs `values` to the current run. Base `log` implementations of a tracking API should go in here, along with
special behavior for the `step parameter.
Args:
values (Dictionary `str` to `str`, `float`, or `int`):
Values to be logged as key-value pairs. The values need to have type `str`, `float`, or `int`.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
"""
pass
def finish(self):
"""
Should run any finalizing functions within the tracking API. If the API should not have one, just don't
overwrite that method.
"""
pass
class TensorBoardTracker(GeneralTracker):
"""
A `Tracker` class that supports `tensorboard`. Should be initialized at the start of your script.
Args:
run_name (`str`):
The name of the experiment run
logging_dir (`str`, `os.PathLike`):
Location for TensorBoard logs to be stored.
**kwargs (additional keyword arguments, *optional*):
Additional key word arguments passed along to the `tensorboard.SummaryWriter.__init__` method.
"""
name = "tensorboard"
requires_logging_directory = True
def __init__(self, run_name: str, logging_dir: Union[str, os.PathLike], **kwargs):
super().__init__()
self.run_name = run_name
self.logging_dir_param = logging_dir
self.init_kwargs = kwargs
@on_main_process
def start(self):
try:
from torch.utils import tensorboard
except ModuleNotFoundError:
import tensorboardX as tensorboard
self.logging_dir = os.path.join(self.logging_dir_param, self.run_name)
self.writer = tensorboard.SummaryWriter(self.logging_dir, **self.init_kwargs)
logger.debug(f"Initialized TensorBoard project {self.run_name} logging to {self.logging_dir}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.writer
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment. Stores the
hyperparameters in a yaml file for future use.
Args:
values (Dictionary `str` to `bool`, `str`, `float` or `int`):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, `int`, or `None`.
"""
self.writer.add_hparams(values, metric_dict={})
self.writer.flush()
project_run_name = time.time()
dir_name = os.path.join(self.logging_dir, str(project_run_name))
os.makedirs(dir_name, exist_ok=True)
with open(os.path.join(dir_name, "hparams.yml"), "w") as outfile:
try:
yaml.dump(values, outfile)
except yaml.representer.RepresenterError:
logger.error("Serialization to store hyperparameters failed")
raise
logger.debug("Stored initial configuration hyperparameters to TensorBoard and hparams yaml file")
@on_main_process
def log(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `values` to the current run.
Args:
values (Dictionary `str` to `str`, `float`, `int` or `dict` of `str` to `float`/`int`):
Values to be logged as key-value pairs. The values need to have type `str`, `float`, `int` or `dict` of
`str` to `float`/`int`.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to either `SummaryWriter.add_scaler`,
`SummaryWriter.add_text`, or `SummaryWriter.add_scalers` method based on the contents of `values`.
"""
values = listify(values)
for k, v in values.items():
if isinstance(v, (int, float)):
self.writer.add_scalar(k, v, global_step=step, **kwargs)
elif isinstance(v, str):
self.writer.add_text(k, v, global_step=step, **kwargs)
elif isinstance(v, dict):
self.writer.add_scalars(k, v, global_step=step, **kwargs)
self.writer.flush()
logger.debug("Successfully logged to TensorBoard")
@on_main_process
def log_images(self, values: dict, step: Optional[int], **kwargs):
"""
Logs `images` to the current run.
Args:
values (Dictionary `str` to `List` of `np.ndarray` or `PIL.Image`):
Values to be logged as key-value pairs. The values need to have type `List` of `np.ndarray` or
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `SummaryWriter.add_image` method.
"""
for k, v in values.items():
self.writer.add_images(k, v, global_step=step, **kwargs)
logger.debug("Successfully logged images to TensorBoard")
@on_main_process
def finish(self):
"""
Closes `TensorBoard` writer
"""
self.writer.close()
logger.debug("TensorBoard writer closed")
class WandBTracker(GeneralTracker):
"""
A `Tracker` class that supports `wandb`. Should be initialized at the start of your script.
Args:
run_name (`str`):
The name of the experiment run.
**kwargs (additional keyword arguments, *optional*):
Additional key word arguments passed along to the `wandb.init` method.
"""
name = "wandb"
requires_logging_directory = False
main_process_only = False
def __init__(self, run_name: str, **kwargs):
super().__init__()
self.run_name = run_name
self.init_kwargs = kwargs
@on_main_process
def start(self):
import wandb
self.run = wandb.init(project=self.run_name, **self.init_kwargs)
logger.debug(f"Initialized WandB project {self.run_name}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.run
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment.
Args:
values (Dictionary `str` to `bool`, `str`, `float` or `int`):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, `int`, or `None`.
"""
import wandb
wandb.config.update(values, allow_val_change=True)
logger.debug("Stored initial configuration hyperparameters to WandB")
@on_main_process
def log(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `values` to the current run.
Args:
values (Dictionary `str` to `str`, `float`, `int` or `dict` of `str` to `float`/`int`):
Values to be logged as key-value pairs. The values need to have type `str`, `float`, `int` or `dict` of
`str` to `float`/`int`.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `wandb.log` method.
"""
self.run.log(values, step=step, **kwargs)
logger.debug("Successfully logged to WandB")
@on_main_process
def log_images(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `images` to the current run.
Args:
values (Dictionary `str` to `List` of `np.ndarray` or `PIL.Image`):
Values to be logged as key-value pairs. The values need to have type `List` of `np.ndarray` or
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `wandb.log` method.
"""
import wandb
for k, v in values.items():
self.log({k: [wandb.Image(image) for image in v]}, step=step, **kwargs)
logger.debug("Successfully logged images to WandB")
@on_main_process
def log_table(
self,
table_name: str,
columns: list[str] = None,
data: list[list[Any]] = None,
dataframe: Any = None,
step: Optional[int] = None,
**kwargs,
):
"""
Log a Table containing any object type (text, image, audio, video, molecule, html, etc). Can be defined either
with `columns` and `data` or with `dataframe`.
Args:
table_name (`str`):
The name to give to the logged table on the wandb workspace
columns (list of `str`, *optional*):
The name of the columns on the table
data (List of List of Any data type, *optional*):
The data to be logged in the table
dataframe (Any data type, *optional*):
The data to be logged in the table
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
"""
import wandb
values = {table_name: wandb.Table(columns=columns, data=data, dataframe=dataframe)}
self.log(values, step=step, **kwargs)
@on_main_process
def finish(self):
"""
Closes `wandb` writer
"""
self.run.finish()
logger.debug("WandB run closed")
class CometMLTracker(GeneralTracker):
"""
A `Tracker` class that supports `comet_ml`. Should be initialized at the start of your script.
API keys must be stored in a Comet config file.
Note:
For `comet_ml` versions < 3.41.0, additional keyword arguments are passed to `comet_ml.Experiment` instead:
https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment/#comet_ml.Experiment.__init__
Args:
run_name (`str`):
The name of the experiment run.
**kwargs (additional keyword arguments, *optional*):
Additional key word arguments passed along to the `comet_ml.start` method:
https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/start/
"""
name = "comet_ml"
requires_logging_directory = False
def __init__(self, run_name: str, **kwargs):
super().__init__()
self.run_name = run_name
self.init_kwargs = kwargs
@on_main_process
def start(self):
import comet_ml
comet_version = version.parse(comet_ml.__version__)
if compare_versions(comet_version, ">=", "3.41.0"):
self.writer = comet_ml.start(project_name=self.run_name, **self.init_kwargs)
else:
logger.info("Update `comet_ml` (>=3.41.0) for experiment reuse and offline support.")
self.writer = comet_ml.Experiment(project_name=self.run_name, **self.init_kwargs)
logger.debug(f"Initialized CometML project {self.run_name}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.writer
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment.
Args:
values (Dictionary `str` to `bool`, `str`, `float` or `int`):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, `int`, or `None`.
"""
self.writer.log_parameters(values)
logger.debug("Stored initial configuration hyperparameters to Comet")
@on_main_process
def log(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `values` to the current run.
Args:
values (Dictionary `str` to `str`, `float`, `int` or `dict` of `str` to `float`/`int`):
Values to be logged as key-value pairs. The values need to have type `str`, `float`, `int` or `dict` of
`str` to `float`/`int`.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to either `Experiment.log_metric`, `Experiment.log_other`,
or `Experiment.log_metrics` method based on the contents of `values`.
"""
if step is not None:
self.writer.set_step(step)
for k, v in values.items():
if isinstance(v, (int, float)):
self.writer.log_metric(k, v, step=step, **kwargs)
elif isinstance(v, str):
self.writer.log_other(k, v, **kwargs)
elif isinstance(v, dict):
self.writer.log_metrics(v, step=step, **kwargs)
logger.debug("Successfully logged to Comet")
@on_main_process
def finish(self):
"""
Flush `comet-ml` writer
"""
self.writer.end()
logger.debug("Comet run flushed")
class AimTracker(GeneralTracker):
"""
A `Tracker` class that supports `aim`. Should be initialized at the start of your script.
Args:
run_name (`str`):
The name of the experiment run.
**kwargs (additional keyword arguments, *optional*):
Additional key word arguments passed along to the `Run.__init__` method.
"""
name = "aim"
requires_logging_directory = True
def __init__(self, run_name: str, logging_dir: Optional[Union[str, os.PathLike]] = ".", **kwargs):
super().__init__()
self.run_name = run_name
self.aim_repo_path = logging_dir
self.init_kwargs = kwargs
@on_main_process
def start(self):
from aim import Run
self.writer = Run(repo=self.aim_repo_path, **self.init_kwargs)
self.writer.name = self.run_name
logger.debug(f"Initialized Aim project {self.run_name}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.writer
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment.
Args:
values (`dict`):
Values to be stored as initial hyperparameters as key-value pairs.
"""
self.writer["hparams"] = values
@on_main_process
def log(self, values: dict, step: Optional[int], **kwargs):
"""
Logs `values` to the current run.
Args:
values (`dict`):
Values to be logged as key-value pairs.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `Run.track` method.
"""
# Note: replace this with the dictionary support when merged
for key, value in values.items():
self.writer.track(value, name=key, step=step, **kwargs)
@on_main_process
def log_images(self, values: dict, step: Optional[int] = None, kwargs: Optional[dict[str, dict]] = None):
"""
Logs `images` to the current run.
Args:
values (`Dict[str, Union[np.ndarray, PIL.Image, Tuple[np.ndarray, str], Tuple[PIL.Image, str]]]`):
Values to be logged as key-value pairs. The values need to have type `np.ndarray` or PIL.Image. If a
tuple is provided, the first element should be the image and the second element should be the caption.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs (`Dict[str, dict]`):
Additional key word arguments passed along to the `Run.Image` and `Run.track` method specified by the
keys `aim_image` and `track`, respectively.
"""
import aim
aim_image_kw = {}
track_kw = {}
if kwargs is not None:
aim_image_kw = kwargs.get("aim_image", {})
track_kw = kwargs.get("track", {})
for key, value in values.items():
if isinstance(value, tuple):
img, caption = value
else:
img, caption = value, ""
aim_image = aim.Image(img, caption=caption, **aim_image_kw)
self.writer.track(aim_image, name=key, step=step, **track_kw)
@on_main_process
def finish(self):
"""
Closes `aim` writer
"""
self.writer.close()
class MLflowTracker(GeneralTracker):
"""
A `Tracker` class that supports `mlflow`. Should be initialized at the start of your script.
Args:
experiment_name (`str`, *optional*):
Name of the experiment. Environment variable MLFLOW_EXPERIMENT_NAME has priority over this argument.
logging_dir (`str` or `os.PathLike`, defaults to `"."`):
Location for mlflow logs to be stored.
run_id (`str`, *optional*):
If specified, get the run with the specified UUID and log parameters and metrics under that run. The runβs
end time is unset and its status is set to running, but the runβs other attributes (source_version,
source_type, etc.) are not changed. Environment variable MLFLOW_RUN_ID has priority over this argument.
tags (`Dict[str, str]`, *optional*):
An optional `dict` of `str` keys and values, or a `str` dump from a `dict`, to set as tags on the run. If a
run is being resumed, these tags are set on the resumed run. If a new run is being created, these tags are
set on the new run. Environment variable MLFLOW_TAGS has priority over this argument.
nested_run (`bool`, *optional*, defaults to `False`):
Controls whether run is nested in parent run. True creates a nested run. Environment variable
MLFLOW_NESTED_RUN has priority over this argument.
run_name (`str`, *optional*):
Name of new run (stored as a mlflow.runName tag). Used only when `run_id` is unspecified.
description (`str`, *optional*):
An optional string that populates the description box of the run. If a run is being resumed, the
description is set on the resumed run. If a new run is being created, the description is set on the new
run.
"""
name = "mlflow"
requires_logging_directory = False
def __init__(
self,
experiment_name: str = None,
logging_dir: Optional[Union[str, os.PathLike]] = None,
run_id: Optional[str] = None,
tags: Optional[Union[dict[str, Any], str]] = None,
nested_run: Optional[bool] = False,
run_name: Optional[str] = None,
description: Optional[str] = None,
):
experiment_name = os.environ.get("MLFLOW_EXPERIMENT_NAME", experiment_name)
run_id = os.environ.get("MLFLOW_RUN_ID", run_id)
tags = os.environ.get("MLFLOW_TAGS", tags)
if isinstance(tags, str):
tags = json.loads(tags)
nested_run = os.environ.get("MLFLOW_NESTED_RUN", nested_run)
self.experiment_name = experiment_name
self.logging_dir = logging_dir
self.run_id = run_id
self.tags = tags
self.nested_run = nested_run
self.run_name = run_name
self.description = description
@on_main_process
def start(self):
import mlflow
exps = mlflow.search_experiments(filter_string=f"name = '{self.experiment_name}'")
if len(exps) > 0:
if len(exps) > 1:
logger.warning("Multiple experiments with the same name found. Using first one.")
experiment_id = exps[0].experiment_id
else:
experiment_id = mlflow.create_experiment(
name=self.experiment_name,
artifact_location=self.logging_dir,
tags=self.tags,
)
self.active_run = mlflow.start_run(
run_id=self.run_id,
experiment_id=experiment_id,
run_name=self.run_name,
nested=self.nested_run,
tags=self.tags,
description=self.description,
)
logger.debug(f"Initialized mlflow experiment {self.experiment_name}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.active_run
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment.
Args:
values (`dict`):
Values to be stored as initial hyperparameters as key-value pairs.
"""
import mlflow
for name, value in list(values.items()):
# internally, all values are converted to str in MLflow
if len(str(value)) > mlflow.utils.validation.MAX_PARAM_VAL_LENGTH:
logger.warning_once(
f'Accelerate is attempting to log a value of "{value}" for key "{name}" as a parameter. MLflow\'s'
f" log_param() only accepts values no longer than {mlflow.utils.validation.MAX_PARAM_VAL_LENGTH} characters so we dropped this attribute."
)
del values[name]
values_list = list(values.items())
# MLflow cannot log more than 100 values in one go, so we have to split it
for i in range(0, len(values_list), mlflow.utils.validation.MAX_PARAMS_TAGS_PER_BATCH):
mlflow.log_params(dict(values_list[i : i + mlflow.utils.validation.MAX_PARAMS_TAGS_PER_BATCH]))
logger.debug("Stored initial configuration hyperparameters to MLflow")
@on_main_process
def log(self, values: dict, step: Optional[int]):
"""
Logs `values` to the current run.
Args:
values (`dict`):
Values to be logged as key-value pairs.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
"""
metrics = {}
for k, v in values.items():
if isinstance(v, (int, float)):
metrics[k] = v
else:
logger.warning_once(
f'MLflowTracker is attempting to log a value of "{v}" of type {type(v)} for key "{k}" as a metric. '
"MLflow's log_metric() only accepts float and int types so we dropped this attribute."
)
import mlflow
mlflow.log_metrics(metrics, step=step)
logger.debug("Successfully logged to mlflow")
@on_main_process
def log_figure(self, figure: Any, artifact_file: str, **save_kwargs):
"""
Logs an figure to the current run.
Args:
figure (Any):
The figure to be logged.
artifact_file (`str`, *optional*):
The run-relative artifact file path in posixpath format to which the image is saved.
If not provided, the image is saved to a default location.
**kwargs:
Additional keyword arguments passed to the underlying mlflow.log_image function.
"""
import mlflow
mlflow.log_figure(figure=figure, artifact_file=artifact_file, **save_kwargs)
logger.debug("Successfully logged image to mlflow")
@on_main_process
def log_artifacts(self, local_dir: str, artifact_path: Optional[str] = None):
"""
Logs an artifacts (all content of a dir) to the current run.
local_dir (`str`):
Path to the directory to be logged as an artifact.
artifact_path (`str`, *optional*):
Directory within the run's artifact directory where the artifact will be logged. If omitted, the
artifact will be logged to the root of the run's artifact directory. The run step. If included, the
artifact will be affiliated with this step.
"""
import mlflow
mlflow.log_artifacts(local_dir=local_dir, artifact_path=artifact_path)
logger.debug("Successfully logged artofact to mlflow")
@on_main_process
def log_artifact(self, local_path: str, artifact_path: Optional[str] = None):
"""
Logs an artifact (file) to the current run.
local_path (`str`):
Path to the file to be logged as an artifact.
artifact_path (`str`, *optional*):
Directory within the run's artifact directory where the artifact will be logged. If omitted, the
artifact will be logged to the root of the run's artifact directory. The run step. If included, the
artifact will be affiliated with this step.
"""
import mlflow
mlflow.log_artifact(local_path=local_path, artifact_path=artifact_path)
logger.debug("Successfully logged artofact to mlflow")
@on_main_process
def finish(self):
"""
End the active MLflow run.
"""
import mlflow
mlflow.end_run()
class ClearMLTracker(GeneralTracker):
"""
A `Tracker` class that supports `clearml`. Should be initialized at the start of your script.
Args:
run_name (`str`, *optional*):
Name of the experiment. Environment variables `CLEARML_PROJECT` and `CLEARML_TASK` have priority over this
argument.
**kwargs (additional keyword arguments, *optional*):
Kwargs passed along to the `Task.__init__` method.
"""
name = "clearml"
requires_logging_directory = False
def __init__(self, run_name: str = None, **kwargs):
super().__init__()
self.user_provided_run_name = run_name
self._initialized_externally = False
self.init_kwargs = kwargs
@on_main_process
def start(self):
from clearml import Task
current_task = Task.current_task()
if current_task:
self._initialized_externally = True
self.task = current_task
return
task_init_args = {**self.init_kwargs}
task_init_args.setdefault("project_name", os.environ.get("CLEARML_PROJECT", self.user_provided_run_name))
task_init_args.setdefault("task_name", os.environ.get("CLEARML_TASK", self.user_provided_run_name))
self.task = Task.init(**task_init_args)
@property
def tracker(self):
return self.task
@on_main_process
def store_init_configuration(self, values: dict):
"""
Connect configuration dictionary to the Task object. Should be run at the beginning of your experiment.
Args:
values (`dict`):
Values to be stored as initial hyperparameters as key-value pairs.
"""
return self.task.connect_configuration(values)
@on_main_process
def log(self, values: dict[str, Union[int, float]], step: Optional[int] = None, **kwargs):
"""
Logs `values` dictionary to the current run. The dictionary keys must be strings. The dictionary values must be
ints or floats
Args:
values (`Dict[str, Union[int, float]]`):
Values to be logged as key-value pairs. If the key starts with 'eval_'/'test_'/'train_', the value will
be reported under the 'eval'/'test'/'train' series and the respective prefix will be removed.
Otherwise, the value will be reported under the 'train' series, and no prefix will be removed.
step (`int`, *optional*):
If specified, the values will be reported as scalars, with the iteration number equal to `step`.
Otherwise they will be reported as single values.
kwargs:
Additional key word arguments passed along to the `clearml.Logger.report_single_value` or
`clearml.Logger.report_scalar` methods.
"""
clearml_logger = self.task.get_logger()
for k, v in values.items():
if not isinstance(v, (int, float)):
logger.warning_once(
"Accelerator is attempting to log a value of "
f'"{v}" of type {type(v)} for key "{k}" as a scalar. '
"This invocation of ClearML logger's report_scalar() "
"is incorrect so we dropped this attribute."
)
continue
if step is None:
clearml_logger.report_single_value(name=k, value=v, **kwargs)
continue
title, series = ClearMLTracker._get_title_series(k)
clearml_logger.report_scalar(title=title, series=series, value=v, iteration=step, **kwargs)
@on_main_process
def log_images(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `images` to the current run.
Args:
values (`Dict[str, List[Union[np.ndarray, PIL.Image]]`):
Values to be logged as key-value pairs. The values need to have type `List` of `np.ndarray` or
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `clearml.Logger.report_image` method.
"""
clearml_logger = self.task.get_logger()
for k, v in values.items():
title, series = ClearMLTracker._get_title_series(k)
clearml_logger.report_image(title=title, series=series, iteration=step, image=v, **kwargs)
@on_main_process
def log_table(
self,
table_name: str,
columns: list[str] = None,
data: list[list[Any]] = None,
dataframe: Any = None,
step: Optional[int] = None,
**kwargs,
):
"""
Log a Table to the task. Can be defined eitherwith `columns` and `data` or with `dataframe`.
Args:
table_name (`str`):
The name of the table
columns (list of `str`, *optional*):
The name of the columns on the table
data (List of List of Any data type, *optional*):
The data to be logged in the table. If `columns` is not specified, then the first entry in data will be
the name of the columns of the table
dataframe (Any data type, *optional*):
The data to be logged in the table
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `clearml.Logger.report_table` method.
"""
to_report = dataframe
if dataframe is None:
if data is None:
raise ValueError(
"`ClearMLTracker.log_table` requires that `data` to be supplied if `dataframe` is `None`"
)
to_report = [columns] + data if columns else data
title, series = ClearMLTracker._get_title_series(table_name)
self.task.get_logger().report_table(title=title, series=series, table_plot=to_report, iteration=step, **kwargs)
@on_main_process
def finish(self):
"""
Close the ClearML task. If the task was initialized externally (e.g. by manually calling `Task.init`), this
function is a noop
"""
if self.task and not self._initialized_externally:
self.task.close()
@staticmethod
def _get_title_series(name):
for prefix in ["eval", "test", "train"]:
if name.startswith(prefix + "_"):
return name[len(prefix) + 1 :], prefix
return name, "train"
class DVCLiveTracker(GeneralTracker):
"""
A `Tracker` class that supports `dvclive`. Should be initialized at the start of your script.
Args:
run_name (`str`, *optional*):
Ignored for dvclive. See `kwargs` instead.
kwargs:
Additional key word arguments passed along to [`dvclive.Live()`](https://dvc.org/doc/dvclive/live).
Example:
```py
from accelerate import Accelerator
accelerator = Accelerator(log_with="dvclive")
accelerator.init_trackers(project_name="my_project", init_kwargs={"dvclive": {"dir": "my_directory"}})
```
"""
name = "dvclive"
requires_logging_directory = False
def __init__(self, run_name: Optional[str] = None, live: Optional[Any] = None, **kwargs):
super().__init__()
self.live = live
self.init_kwargs = kwargs
@on_main_process
def start(self):
from dvclive import Live
self.live = self.live if self.live is not None else Live(**self.init_kwargs)
@property
def tracker(self):
return self.live
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment. Stores the
hyperparameters in a yaml file for future use.
Args:
values (Dictionary `str` to `bool`, `str`, `float`, `int`, or a List or Dict of those types):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, or `int`.
"""
self.live.log_params(values)
@on_main_process
def log(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `values` to the current run.
Args:
values (Dictionary `str` to `str`, `float`, or `int`):
Values to be logged as key-value pairs. The values need to have type `str`, `float`, or `int`.
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to `dvclive.Live.log_metric()`.
"""
from dvclive.plots import Metric
if step is not None:
self.live.step = step
for k, v in values.items():
if Metric.could_log(v):
self.live.log_metric(k, v, **kwargs)
else:
logger.warning_once(
"Accelerator attempted to log a value of "
f'"{v}" of type {type(v)} for key "{k}" as a scalar. '
"This invocation of DVCLive's Live.log_metric() "
"is incorrect so we dropped this attribute."
)
self.live.next_step()
@on_main_process
def finish(self):
"""
Closes `dvclive.Live()`.
"""
self.live.end()
class SwanLabTracker(GeneralTracker):
"""
A `Tracker` class that supports `swanlab`. Should be initialized at the start of your script.
Args:
run_name (`str`):
The name of the experiment run.
**kwargs (additional keyword arguments, *optional*):
Additional key word arguments passed along to the `swanlab.init` method.
"""
name = "swanlab"
requires_logging_directory = False
main_process_only = False
def __init__(self, run_name: str, **kwargs):
super().__init__()
self.run_name = run_name
self.init_kwargs = kwargs
@on_main_process
def start(self):
import swanlab
self.run = swanlab.init(project=self.run_name, **self.init_kwargs)
swanlab.config["FRAMEWORK"] = "π€Accelerate" # add accelerate logo in config
logger.debug(f"Initialized SwanLab project {self.run_name}")
logger.debug(
"Make sure to log any initial configurations with `self.store_init_configuration` before training!"
)
@property
def tracker(self):
return self.run
@on_main_process
def store_init_configuration(self, values: dict):
"""
Logs `values` as hyperparameters for the run. Should be run at the beginning of your experiment.
Args:
values (Dictionary `str` to `bool`, `str`, `float` or `int`):
Values to be stored as initial hyperparameters as key-value pairs. The values need to have type `bool`,
`str`, `float`, `int`, or `None`.
"""
import swanlab
swanlab.config.update(values, allow_val_change=True)
logger.debug("Stored initial configuration hyperparameters to SwanLab")
@on_main_process
def log(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `values` to the current run.
Args:
data : Dict[str, DataType]
Data must be a dict. The key must be a string with 0-9, a-z, A-Z, " ", "_", "-", "/". The value must be a
`float`, `float convertible object`, `int` or `swanlab.data.BaseType`.
step : int, optional
The step number of the current data, if not provided, it will be automatically incremented.
If step is duplicated, the data will be ignored.
kwargs:
Additional key word arguments passed along to the `swanlab.log` method. Likes:
print_to_console : bool, optional
Whether to print the data to the console, the default is False.
"""
self.run.log(values, step=step, **kwargs)
logger.debug("Successfully logged to SwanLab")
@on_main_process
def log_images(self, values: dict, step: Optional[int] = None, **kwargs):
"""
Logs `images` to the current run.
Args:
values (Dictionary `str` to `List` of `np.ndarray` or `PIL.Image`):
Values to be logged as key-value pairs. The values need to have type `List` of `np.ndarray` or
step (`int`, *optional*):
The run step. If included, the log will be affiliated with this step.
kwargs:
Additional key word arguments passed along to the `swanlab.log` method. Likes:
print_to_console : bool, optional
Whether to print the data to the console, the default is False.
"""
import swanlab
for k, v in values.items():
self.log({k: [swanlab.Image(image) for image in v]}, step=step, **kwargs)
logger.debug("Successfully logged images to SwanLab")
@on_main_process
def finish(self):
"""
Closes `swanlab` writer
"""
self.run.finish()
logger.debug("SwanLab run closed")
LOGGER_TYPE_TO_CLASS = {
"aim": AimTracker,
"comet_ml": CometMLTracker,
"mlflow": MLflowTracker,
"tensorboard": TensorBoardTracker,
"wandb": WandBTracker,
"clearml": ClearMLTracker,
"dvclive": DVCLiveTracker,
"swanlab": SwanLabTracker,
}
def filter_trackers(
log_with: list[Union[str, LoggerType, GeneralTracker]],
logging_dir: Union[str, os.PathLike] = None,
):
"""
Takes in a list of potential tracker types and checks that:
- The tracker wanted is available in that environment
- Filters out repeats of tracker types
- If `all` is in `log_with`, will return all trackers in the environment
- If a tracker requires a `logging_dir`, ensures that `logging_dir` is not `None`
Args:
log_with (list of `str`, [`~utils.LoggerType`] or [`~tracking.GeneralTracker`], *optional*):
A list of loggers to be setup for experiment tracking. Should be one or several of:
- `"all"`
- `"tensorboard"`
- `"wandb"`
- `"comet_ml"`
- `"mlflow"`
- `"dvclive"`
- `"swanlab"`
If `"all"` is selected, will pick up all available trackers in the environment and initialize them. Can
also accept implementations of `GeneralTracker` for custom trackers, and can be combined with `"all"`.
logging_dir (`str`, `os.PathLike`, *optional*):
A path to a directory for storing logs of locally-compatible loggers.
"""
loggers = []
if log_with is not None:
if not isinstance(log_with, (list, tuple)):
log_with = [log_with]
if "all" in log_with or LoggerType.ALL in log_with:
loggers = [o for o in log_with if issubclass(type(o), GeneralTracker)] + get_available_trackers()
else:
for log_type in log_with:
if log_type not in LoggerType and not issubclass(type(log_type), GeneralTracker):
raise ValueError(f"Unsupported logging capability: {log_type}. Choose between {LoggerType.list()}")
if issubclass(type(log_type), GeneralTracker):
loggers.append(log_type)
else:
log_type = LoggerType(log_type)
if log_type not in loggers:
if log_type in get_available_trackers():
tracker_init = LOGGER_TYPE_TO_CLASS[str(log_type)]
if tracker_init.requires_logging_directory:
if logging_dir is None:
raise ValueError(
f"Logging with `{log_type}` requires a `logging_dir` to be passed in."
)
loggers.append(log_type)
else:
logger.debug(f"Tried adding logger {log_type}, but package is unavailable in the system.")
return loggers
|