import_utils.py 39.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
# 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.
"""
Import utilities: Utilities related to imports and our lazy inits.
"""

18
import importlib.metadata
19
20
21
import importlib.util
import json
import os
22
import shutil
玩火's avatar
玩火 committed
23
import subprocess
24
import sys
25
import warnings
26
from collections import OrderedDict
27
from functools import lru_cache
28
29
from itertools import chain
from types import ModuleType
30
from typing import Any, Tuple, Union
31
32
33
34
35
36
37
38

from packaging import version

from . import logging


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

39

Yih-Dar's avatar
Yih-Dar committed
40
# TODO: This doesn't work for all packages (`bs4`, `faiss`, etc.) Talk to Sylvain to see how to do with it better.
41
42
43
44
45
46
def _is_package_available(pkg_name: str, return_version: bool = False) -> Union[Tuple[bool, str], bool]:
    # Check we're not importing a "pkg_name" directory somewhere but the actual library by trying to grab the version
    package_exists = importlib.util.find_spec(pkg_name) is not None
    package_version = "N/A"
    if package_exists:
        try:
47
            package_version = importlib.metadata.version(pkg_name)
48
            package_exists = True
49
        except importlib.metadata.PackageNotFoundError:
50
51
52
53
54
55
56
57
            package_exists = False
        logger.debug(f"Detected {pkg_name} version {package_version}")
    if return_version:
        return package_exists, package_version
    else:
        return package_exists


58
59
60
61
62
63
64
ENV_VARS_TRUE_VALUES = {"1", "ON", "YES", "TRUE"}
ENV_VARS_TRUE_AND_AUTO_VALUES = ENV_VARS_TRUE_VALUES.union({"AUTO"})

USE_TF = os.environ.get("USE_TF", "AUTO").upper()
USE_TORCH = os.environ.get("USE_TORCH", "AUTO").upper()
USE_JAX = os.environ.get("USE_FLAX", "AUTO").upper()

65
66
FORCE_TF_AVAILABLE = os.environ.get("FORCE_TF_AVAILABLE", "AUTO").upper()

67
68
69
70
71
72
73
# This is the version of torch required to run torch.fx features and torch.onnx with dictionary inputs.
TORCH_FX_REQUIRED_VERSION = version.parse("1.10")


_accelerate_available, _accelerate_version = _is_package_available("accelerate", return_version=True)
_apex_available = _is_package_available("apex")
_bitsandbytes_available = _is_package_available("bitsandbytes")
74
# `importlib.metadata.version` doesn't work with `bs4` but `beautifulsoup4`. For `importlib.util.find_spec`, reversed.
Yih-Dar's avatar
Yih-Dar committed
75
_bs4_available = importlib.util.find_spec("bs4") is not None
76
77
78
79
_coloredlogs_available = _is_package_available("coloredlogs")
_datasets_available = _is_package_available("datasets")
_decord_available = importlib.util.find_spec("decord") is not None
_detectron2_available = _is_package_available("detectron2")
Yih-Dar's avatar
Yih-Dar committed
80
81
82
# We need to check both `faiss` and `faiss-cpu`.
_faiss_available = importlib.util.find_spec("faiss") is not None
try:
83
    _faiss_version = importlib.metadata.version("faiss")
Yih-Dar's avatar
Yih-Dar committed
84
    logger.debug(f"Successfully imported faiss version {_faiss_version}")
85
except importlib.metadata.PackageNotFoundError:
Yih-Dar's avatar
Yih-Dar committed
86
    try:
87
        _faiss_version = importlib.metadata.version("faiss-cpu")
Yih-Dar's avatar
Yih-Dar committed
88
        logger.debug(f"Successfully imported faiss version {_faiss_version}")
89
    except importlib.metadata.PackageNotFoundError:
Yih-Dar's avatar
Yih-Dar committed
90
        _faiss_available = False
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
_ftfy_available = _is_package_available("ftfy")
_ipex_available, _ipex_version = _is_package_available("intel_extension_for_pytorch", return_version=True)
_jieba_available = _is_package_available("jieba")
_kenlm_available = _is_package_available("kenlm")
_keras_nlp_available = _is_package_available("keras_nlp")
_librosa_available = _is_package_available("librosa")
_natten_available = _is_package_available("natten")
_onnx_available = _is_package_available("onnx")
_openai_available = _is_package_available("openai")
_optimum_available = _is_package_available("optimum")
_pandas_available = _is_package_available("pandas")
_peft_available = _is_package_available("peft")
_phonemizer_available = _is_package_available("phonemizer")
_psutil_available = _is_package_available("psutil")
_py3nvml_available = _is_package_available("py3nvml")
_pyctcdecode_available = _is_package_available("pyctcdecode")
_pytesseract_available = _is_package_available("pytesseract")
108
_pytest_available = _is_package_available("pytest")
109
110
111
112
113
114
_pytorch_quantization_available = _is_package_available("pytorch_quantization")
_rjieba_available = _is_package_available("rjieba")
_sacremoses_available = _is_package_available("sacremoses")
_safetensors_available = _is_package_available("safetensors")
_scipy_available = _is_package_available("scipy")
_sentencepiece_available = _is_package_available("sentencepiece")
115
_is_seqio_available = _is_package_available("seqio")
116
117
118
_sklearn_available = importlib.util.find_spec("sklearn") is not None
if _sklearn_available:
    try:
119
120
        importlib.metadata.version("scikit-learn")
    except importlib.metadata.PackageNotFoundError:
121
        _sklearn_available = False
122
_smdistributed_available = importlib.util.find_spec("smdistributed") is not None
123
124
125
126
127
128
129
130
131
132
133
134
135
_soundfile_available = _is_package_available("soundfile")
_spacy_available = _is_package_available("spacy")
_sudachipy_available = _is_package_available("sudachipy")
_tensorflow_probability_available = _is_package_available("tensorflow_probability")
_tensorflow_text_available = _is_package_available("tensorflow_text")
_tf2onnx_available = _is_package_available("tf2onnx")
_timm_available = _is_package_available("timm")
_tokenizers_available = _is_package_available("tokenizers")
_torchaudio_available = _is_package_available("torchaudio")
_torchdistx_available = _is_package_available("torchdistx")
_torchvision_available = _is_package_available("torchvision")


136
_torch_version = "N/A"
137
_torch_available = False
138
if USE_TORCH in ENV_VARS_TRUE_AND_AUTO_VALUES and USE_TF not in ENV_VARS_TRUE_VALUES:
139
    _torch_available, _torch_version = _is_package_available("torch", return_version=True)
140
141
142
143
144
145
else:
    logger.info("Disabling PyTorch because USE_TF is set")
    _torch_available = False


_tf_version = "N/A"
146
_tf_available = False
147
148
if FORCE_TF_AVAILABLE in ENV_VARS_TRUE_VALUES:
    _tf_available = True
149
else:
150
    if USE_TF in ENV_VARS_TRUE_AND_AUTO_VALUES and USE_TORCH not in ENV_VARS_TRUE_VALUES:
151
152
153
        # Note: _is_package_available("tensorflow") fails for tensorflow-cpu. Please test any changes to the line below
        # with tensorflow-cpu to make sure it still works!
        _tf_available = importlib.util.find_spec("tensorflow") is not None
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
        if _tf_available:
            candidates = (
                "tensorflow",
                "tensorflow-cpu",
                "tensorflow-gpu",
                "tf-nightly",
                "tf-nightly-cpu",
                "tf-nightly-gpu",
                "intel-tensorflow",
                "intel-tensorflow-avx512",
                "tensorflow-rocm",
                "tensorflow-macos",
                "tensorflow-aarch64",
            )
            _tf_version = None
            # For the metadata, we have to look for both tensorflow and tensorflow-cpu
            for pkg in candidates:
                try:
172
                    _tf_version = importlib.metadata.version(pkg)
173
                    break
174
                except importlib.metadata.PackageNotFoundError:
175
176
177
178
179
180
181
182
183
184
                    pass
            _tf_available = _tf_version is not None
        if _tf_available:
            if version.parse(_tf_version) < version.parse("2"):
                logger.info(
                    f"TensorFlow found but with version {_tf_version}. Transformers requires version 2 minimum."
                )
                _tf_available = False
    else:
        logger.info("Disabling Tensorflow because USE_TORCH is set")
185
186


187
188
189
190
191
192
ccl_version = "N/A"
_is_ccl_available = (
    importlib.util.find_spec("torch_ccl") is not None
    or importlib.util.find_spec("oneccl_bindings_for_pytorch") is not None
)
try:
193
    ccl_version = importlib.metadata.version("oneccl_bind_pt")
194
    logger.debug(f"Detected oneccl_bind_pt version {ccl_version}")
195
except importlib.metadata.PackageNotFoundError:
196
    _is_ccl_available = False
197

198

199
200
201
202
203
204
205
206
207
208
_flax_available = False
if USE_JAX in ENV_VARS_TRUE_AND_AUTO_VALUES:
    _flax_available, _flax_version = _is_package_available("flax", return_version=True)
    if _flax_available:
        _jax_available, _jax_version = _is_package_available("jax", return_version=True)
        if _jax_available:
            logger.info(f"JAX version {_jax_version}, Flax version {_flax_version} available.")
        else:
            _flax_available = _jax_available = False
            _jax_version = _flax_version = "N/A"
209

210
211
212
213
214
215
216
217

_torch_fx_available = False
if _torch_available:
    torch_version = version.parse(_torch_version)
    _torch_fx_available = (torch_version.major, torch_version.minor) >= (
        TORCH_FX_REQUIRED_VERSION.major,
        TORCH_FX_REQUIRED_VERSION.minor,
    )
218
219


220
def is_kenlm_available():
221
    return _kenlm_available
222
223


224
225
226
227
def is_torch_available():
    return _torch_available


228
229
230
231
def get_torch_version():
    return _torch_version


NielsRogge's avatar
NielsRogge committed
232
def is_torchvision_available():
233
    return _torchvision_available
NielsRogge's avatar
NielsRogge committed
234
235


236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
def is_pyctcdecode_available():
    return _pyctcdecode_available


def is_librosa_available():
    return _librosa_available


def is_torch_cuda_available():
    if is_torch_available():
        import torch

        return torch.cuda.is_available()
    else:
        return False


253
254
255
256
257
258
259
260
261
def is_torch_mps_available():
    if is_torch_available():
        import torch

        if hasattr(torch.backends, "mps"):
            return torch.backends.mps.is_available()
    return False


262
def is_torch_bf16_gpu_available():
263
264
265
266
267
268
269
270
    if not is_torch_available():
        return False

    import torch

    # since currently no utility function is available we build our own.
    # some bits come from https://github.com/pytorch/pytorch/blob/2289a12f21c54da93bf5d696e3f9aea83dd9c10d/torch/testing/_internal/common_cuda.py#L51
    # with additional check for torch version
Yih-Dar's avatar
Yih-Dar committed
271
272
273
274
    # to succeed: (torch is required to be >= 1.10 anyway)
    # 1. the hardware needs to support bf16 (GPU arch >= Ampere, or CPU)
    # 2. if using gpu, CUDA >= 11
    # 3. torch.autocast exists
275
276
    # XXX: one problem here is that it may give invalid results on mixed gpus setup, so it's
    # really only correct for the 0th gpu (or currently set default device if different from 0)
277
278
    if torch.cuda.is_available() and torch.version.cuda is not None:
        if torch.cuda.get_device_properties(torch.cuda.current_device()).major < 8:
279
            return False
280
        if int(torch.version.cuda.split(".")[0]) < 11:
281
            return False
282
        if not hasattr(torch.cuda.amp, "autocast"):
283
            return False
284
    else:
285
286
287
288
289
290
291
292
293
294
295
        return False

    return True


def is_torch_bf16_cpu_available():
    if not is_torch_available():
        return False

    import torch

296
297
298
299
    try:
        # multiple levels of AttributeError depending on the pytorch version so do them all in one check
        _ = torch.cpu.amp.autocast
    except AttributeError:
300
        return False
301

302
303
304
305
    return True


def is_torch_bf16_available():
306
307
308
309
310
311
312
313
    # the original bf16 check was for gpu only, but later a cpu/bf16 combo has emerged so this util
    # has become ambiguous and therefore deprecated
    warnings.warn(
        "The util is_torch_bf16_available is deprecated, please use is_torch_bf16_gpu_available "
        "or is_torch_bf16_cpu_available instead according to whether it's used with cpu or gpu",
        FutureWarning,
    )
    return is_torch_bf16_gpu_available()
314
315
316
317
318
319
320
321
322
323
324
325
326
327


def is_torch_tf32_available():
    if not is_torch_available():
        return False

    import torch

    if not torch.cuda.is_available() or torch.version.cuda is None:
        return False
    if torch.cuda.get_device_properties(torch.cuda.current_device()).major < 8:
        return False
    if int(torch.version.cuda.split(".")[0]) < 11:
        return False
328
    if version.parse(version.parse(torch.__version__).base_version) < version.parse("1.7"):
329
330
331
332
333
334
335
336
337
        return False

    return True


def is_torch_fx_available():
    return _torch_fx_available


338
def is_peft_available():
339
    return _peft_available
340
341


NielsRogge's avatar
NielsRogge committed
342
def is_bs4_available():
343
    return _bs4_available
NielsRogge's avatar
NielsRogge committed
344
345


346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
def is_tf_available():
    return _tf_available


def is_coloredlogs_available():
    return _coloredlogs_available


def is_tf2onnx_available():
    return _tf2onnx_available


def is_onnx_available():
    return _onnx_available


Sylvain Gugger's avatar
Sylvain Gugger committed
362
def is_openai_available():
363
    return _openai_available
Sylvain Gugger's avatar
Sylvain Gugger committed
364
365


366
367
368
369
370
371
372
373
def is_flax_available():
    return _flax_available


def is_ftfy_available():
    return _ftfy_available


374
@lru_cache()
375
376
def is_torch_tpu_available(check_device=True):
    "Checks if `torch_xla` is installed and potentially if a TPU is in the environment"
377
378
    if not _torch_available:
        return False
379
380
381
382
383
    if importlib.util.find_spec("torch_xla") is not None:
        if check_device:
            # We need to check if `xla_device` can be found, will raise a RuntimeError if not
            try:
                import torch_xla.core.xla_model as xm
384

385
386
387
388
                _ = xm.xla_device()
                return True
            except RuntimeError:
                return False
389
        return True
390
    return False
391
392


393
394
395
396
397
398
399
@lru_cache()
def is_torch_neuroncore_available(check_device=True):
    if importlib.util.find_spec("torch_neuronx") is not None:
        return is_torch_tpu_available(check_device)
    return False


400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
@lru_cache()
def is_torch_npu_available(check_device=False):
    "Checks if `torch_npu` is installed and potentially if a NPU is in the environment"
    if not _torch_available or importlib.util.find_spec("torch_npu") is None:
        return False

    import torch
    import torch_npu  # noqa: F401

    if check_device:
        try:
            # Will raise a RuntimeError if no NPU is found
            _ = torch.npu.device_count()
            return torch.npu.is_available()
        except RuntimeError:
            return False
    return hasattr(torch, "npu") and torch.npu.is_available()


419
def is_torchdynamo_available():
420
421
422
423
424
425
426
427
    if not is_torch_available():
        return False
    try:
        import torch._dynamo as dynamo  # noqa: F401

        return True
    except Exception:
        return False
428
429


430
431
432
433
434
435
def is_torch_compile_available():
    if not is_torch_available():
        return False

    import torch

436
437
    # We don't do any version check here to support nighlies marked as 1.14. Ultimately needs to check version against
    # 2.0 but let's do it later.
438
439
440
    return hasattr(torch, "compile")


441
442
443
444
445
446
def is_torch_tensorrt_fx_available():
    if importlib.util.find_spec("torch_tensorrt") is None:
        return False
    return importlib.util.find_spec("torch_tensorrt.fx") is not None


447
448
449
450
451
452
453
454
455
def is_datasets_available():
    return _datasets_available


def is_detectron2_available():
    return _detectron2_available


def is_rjieba_available():
456
    return _rjieba_available
457
458
459


def is_psutil_available():
460
    return _psutil_available
461
462
463


def is_py3nvml_available():
464
    return _py3nvml_available
465
466


467
def is_sacremoses_available():
468
    return _sacremoses_available
469
470


471
def is_apex_available():
472
    return _apex_available
473
474


475
def is_ninja_available():
玩火's avatar
玩火 committed
476
477
478
479
480
481
482
483
484
485
    r"""
    Code comes from *torch.utils.cpp_extension.is_ninja_available()*. Returns `True` if the
    [ninja](https://ninja-build.org/) build system is available on the system, `False` otherwise.
    """
    try:
        subprocess.check_output("ninja --version".split())
    except Exception:
        return False
    else:
        return True
486
487


488
def is_ipex_available():
489
490
491
    def get_major_and_minor_from_version(full_version):
        return str(version.parse(full_version).major) + "." + str(version.parse(full_version).minor)

492
    if not is_torch_available() or not _ipex_available:
493
        return False
494

495
496
497
498
499
500
501
502
503
    torch_major_and_minor = get_major_and_minor_from_version(_torch_version)
    ipex_major_and_minor = get_major_and_minor_from_version(_ipex_version)
    if torch_major_and_minor != ipex_major_and_minor:
        logger.warning(
            f"Intel Extension for PyTorch {ipex_major_and_minor} needs to work with PyTorch {ipex_major_and_minor}.*,"
            f" but PyTorch {_torch_version} is found. Please switch to the matching version and run again."
        )
        return False
    return True
504
505


506
def is_bitsandbytes_available():
507
    return _bitsandbytes_available
508
509


510
def is_torchdistx_available():
511
    return _torchdistx_available
512
513


514
515
516
517
518
def is_faiss_available():
    return _faiss_available


def is_scipy_available():
519
    return _scipy_available
520
521
522


def is_sklearn_available():
523
    return _sklearn_available
524
525
526


def is_sentencepiece_available():
527
    return _sentencepiece_available
528
529


530
531
532
533
def is_seqio_available():
    return _is_seqio_available


534
535
536
537
538
539
def is_protobuf_available():
    if importlib.util.find_spec("google") is None:
        return False
    return importlib.util.find_spec("google.protobuf") is not None


540
541
542
def is_accelerate_available(min_version: str = None):
    if min_version is not None:
        return _accelerate_available and version.parse(_accelerate_version) >= version.parse(min_version)
543
    return _accelerate_available
544
545


546
def is_optimum_available():
547
    return _optimum_available
548
549


550
def is_optimum_neuron_available():
551
    return _optimum_available and _is_package_available("optimum.neuron")
552
553


554
def is_safetensors_available():
555
    return _safetensors_available
556
557


558
def is_tokenizers_available():
559
    return _tokenizers_available
560
561
562


def is_vision_available():
563
564
565
    _pil_available = importlib.util.find_spec("PIL") is not None
    if _pil_available:
        try:
566
567
            package_version = importlib.metadata.version("Pillow")
        except importlib.metadata.PackageNotFoundError:
Yih-Dar's avatar
Yih-Dar committed
568
569
570
571
            try:
                package_version = importlib.metadata.version("Pillow-SIMD")
            except importlib.metadata.PackageNotFoundError:
                return False
572
573
        logger.debug(f"Detected PIL version {package_version}")
    return _pil_available
574
575
576


def is_pytesseract_available():
577
    return _pytesseract_available
578
579


580
581
582
583
def is_pytest_available():
    return _pytest_available


584
def is_spacy_available():
585
    return _spacy_available
586
587


588
def is_tensorflow_text_available():
589
    return is_tf_available() and _tensorflow_text_available
590
591


592
def is_keras_nlp_available():
593
    return is_tensorflow_text_available() and _keras_nlp_available
594
595


596
597
598
599
600
601
602
603
def is_in_notebook():
    try:
        # Test adapted from tqdm.autonotebook: https://github.com/tqdm/tqdm/blob/master/tqdm/autonotebook.py
        get_ipython = sys.modules["IPython"].get_ipython
        if "IPKernelApp" not in get_ipython().config:
            raise ImportError("console")
        if "VSCODE_PID" in os.environ:
            raise ImportError("vscode")
604
605
606
        if "DATABRICKS_RUNTIME_VERSION" in os.environ and os.environ["DATABRICKS_RUNTIME_VERSION"] < "11.0":
            # Databricks Runtime 11.0 and above uses IPython kernel by default so it should be compatible with Jupyter notebook
            # https://docs.microsoft.com/en-us/azure/databricks/notebooks/ipython-kernel
607
            raise ImportError("databricks")
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622

        return importlib.util.find_spec("IPython") is not None
    except (AttributeError, ImportError, KeyError):
        return False


def is_pytorch_quantization_available():
    return _pytorch_quantization_available


def is_tensorflow_probability_available():
    return _tensorflow_probability_available


def is_pandas_available():
623
    return _pandas_available
624
625
626
627
628
629
630
631
632
633
634
635
636


def is_sagemaker_dp_enabled():
    # Get the sagemaker specific env variable.
    sagemaker_params = os.getenv("SM_FRAMEWORK_PARAMS", "{}")
    try:
        # Parse it and check the field "sagemaker_distributed_dataparallel_enabled".
        sagemaker_params = json.loads(sagemaker_params)
        if not sagemaker_params.get("sagemaker_distributed_dataparallel_enabled", False):
            return False
    except json.JSONDecodeError:
        return False
    # Lastly, check if the `smdistributed` module is present.
637
    return _smdistributed_available
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660


def is_sagemaker_mp_enabled():
    # Get the sagemaker specific mp parameters from smp_options variable.
    smp_options = os.getenv("SM_HP_MP_PARAMETERS", "{}")
    try:
        # Parse it and check the field "partitions" is included, it is required for model parallel.
        smp_options = json.loads(smp_options)
        if "partitions" not in smp_options:
            return False
    except json.JSONDecodeError:
        return False

    # Get the sagemaker specific framework parameters from mpi_options variable.
    mpi_options = os.getenv("SM_FRAMEWORK_PARAMS", "{}")
    try:
        # Parse it and check the field "sagemaker_distributed_dataparallel_enabled".
        mpi_options = json.loads(mpi_options)
        if not mpi_options.get("sagemaker_mpi_enabled", False):
            return False
    except json.JSONDecodeError:
        return False
    # Lastly, check if the `smdistributed` module is present.
661
    return _smdistributed_available
662
663
664
665
666
667
668
669
670
671
672
673
674
675


def is_training_run_on_sagemaker():
    return "SAGEMAKER_JOB_NAME" in os.environ


def is_soundfile_availble():
    return _soundfile_available


def is_timm_available():
    return _timm_available


676
677
678
679
def is_natten_available():
    return _natten_available


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
def is_torchaudio_available():
    return _torchaudio_available


def is_speech_available():
    # For now this depends on torchaudio but the exact dependency might evolve in the future.
    return _torchaudio_available


def is_phonemizer_available():
    return _phonemizer_available


def torch_only_method(fn):
    def wrapper(*args, **kwargs):
        if not _torch_available:
            raise ImportError(
                "You need to install pytorch to use this method or class, "
                "or activate it with environment variables USE_TORCH=1 and USE_TF=0."
            )
        else:
            return fn(*args, **kwargs)

    return wrapper


706
707
708
709
def is_ccl_available():
    return _is_ccl_available


710
def is_decord_available():
711
    return _decord_available
712
713


714
def is_sudachi_available():
715
    return _sudachipy_available
716
717
718


def is_jumanpp_available():
Hao Wang's avatar
Hao Wang committed
719
    return (importlib.util.find_spec("rhoknp") is not None) and (shutil.which("jumanpp") is not None)
720
721


722
723
724
725
def is_cython_available():
    return importlib.util.find_spec("pyximport") is not None


726
727
728
729
def is_jieba_available():
    return _jieba_available


730
731
732
733
734
735
736
737
738
739
740
741
742
743
# docstyle-ignore
DATASETS_IMPORT_ERROR = """
{0} requires the 🤗 Datasets library but it was not found in your environment. You can install it with:
```
pip install datasets
```
In a notebook or a colab, you can install it by executing a cell with
```
!pip install datasets
```
then restarting your kernel.

Note that if you have a local folder named `datasets` or a local python file named `datasets.py` in your current
working directory, python may try to import this instead of the 🤗 Datasets library. You should rename this folder or
744
that python file if that's the case. Please note that you may need to restart your runtime after installation.
745
746
747
748
749
750
751
752
753
754
755
756
757
"""


# docstyle-ignore
TOKENIZERS_IMPORT_ERROR = """
{0} requires the 🤗 Tokenizers library but it was not found in your environment. You can install it with:
```
pip install tokenizers
```
In a notebook or a colab, you can install it by executing a cell with
```
!pip install tokenizers
```
758
Please note that you may need to restart your runtime after installation.
759
760
761
762
763
764
765
"""


# docstyle-ignore
SENTENCEPIECE_IMPORT_ERROR = """
{0} requires the SentencePiece library but it was not found in your environment. Checkout the instructions on the
installation page of its repo: https://github.com/google/sentencepiece#installation and follow the ones
766
that match your environment. Please note that you may need to restart your runtime after installation.
767
768
769
770
771
772
773
"""


# docstyle-ignore
PROTOBUF_IMPORT_ERROR = """
{0} requires the protobuf library but it was not found in your environment. Checkout the instructions on the
installation page of its repo: https://github.com/protocolbuffers/protobuf/tree/master/python#installation and follow the ones
774
that match your environment. Please note that you may need to restart your runtime after installation.
775
776
777
778
779
780
781
"""


# docstyle-ignore
FAISS_IMPORT_ERROR = """
{0} requires the faiss library but it was not found in your environment. Checkout the instructions on the
installation page of its repo: https://github.com/facebookresearch/faiss/blob/master/INSTALL.md and follow the ones
782
that match your environment. Please note that you may need to restart your runtime after installation.
783
784
785
786
787
788
789
"""


# docstyle-ignore
PYTORCH_IMPORT_ERROR = """
{0} requires the PyTorch library but it was not found in your environment. Checkout the instructions on the
installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
790
Please note that you may need to restart your runtime after installation.
791
792
"""

NielsRogge's avatar
NielsRogge committed
793
794
795
796
797
798
799
800

# docstyle-ignore
TORCHVISION_IMPORT_ERROR = """
{0} requires the Torchvision library but it was not found in your environment. Checkout the instructions on the
installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
Please note that you may need to restart your runtime after installation.
"""

801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
# docstyle-ignore
PYTORCH_IMPORT_ERROR_WITH_TF = """
{0} requires the PyTorch library but it was not found in your environment.
However, we were able to find a TensorFlow installation. TensorFlow classes begin
with "TF", but are otherwise identically named to our PyTorch classes. This
means that the TF equivalent of the class you tried to import would be "TF{0}".
If you want to use TensorFlow, please use TF classes instead!

If you really do want to use PyTorch please go to
https://pytorch.org/get-started/locally/ and follow the instructions that
match your environment.
"""

# docstyle-ignore
TF_IMPORT_ERROR_WITH_PYTORCH = """
{0} requires the TensorFlow library but it was not found in your environment.
However, we were able to find a PyTorch installation. PyTorch classes do not begin
with "TF", but are otherwise identically named to our TF classes.
If you want to use PyTorch, please use those classes instead!

If you really do want to use TensorFlow, please follow the instructions on the
installation page https://www.tensorflow.org/install that match your environment.
"""

NielsRogge's avatar
NielsRogge committed
825
826
827
# docstyle-ignore
BS4_IMPORT_ERROR = """
{0} requires the Beautiful Soup library but it was not found in your environment. You can install it with pip:
828
`pip install beautifulsoup4`. Please note that you may need to restart your runtime after installation.
NielsRogge's avatar
NielsRogge committed
829
830
"""

831
832
833
834
835
836
837
838
839
840
841

# docstyle-ignore
SKLEARN_IMPORT_ERROR = """
{0} requires the scikit-learn library but it was not found in your environment. You can install it with:
```
pip install -U scikit-learn
```
In a notebook or a colab, you can install it by executing a cell with
```
!pip install -U scikit-learn
```
842
Please note that you may need to restart your runtime after installation.
843
844
845
846
847
848
849
"""


# docstyle-ignore
TENSORFLOW_IMPORT_ERROR = """
{0} requires the TensorFlow library but it was not found in your environment. Checkout the instructions on the
installation page: https://www.tensorflow.org/install and follow the ones that match your environment.
850
Please note that you may need to restart your runtime after installation.
851
852
853
854
855
856
857
"""


# docstyle-ignore
DETECTRON2_IMPORT_ERROR = """
{0} requires the detectron2 library but it was not found in your environment. Checkout the instructions on the
installation page: https://github.com/facebookresearch/detectron2/blob/master/INSTALL.md and follow the ones
858
that match your environment. Please note that you may need to restart your runtime after installation.
859
860
861
862
863
864
865
"""


# docstyle-ignore
FLAX_IMPORT_ERROR = """
{0} requires the FLAX library but it was not found in your environment. Checkout the instructions on the
installation page: https://github.com/google/flax and follow the ones that match your environment.
866
Please note that you may need to restart your runtime after installation.
867
868
869
870
871
872
"""

# docstyle-ignore
FTFY_IMPORT_ERROR = """
{0} requires the ftfy library but it was not found in your environment. Checkout the instructions on the
installation section: https://github.com/rspeer/python-ftfy/tree/master#installing and follow the ones
873
that match your environment. Please note that you may need to restart your runtime after installation.
874
875
876
877
878
879
"""

# docstyle-ignore
PYTORCH_QUANTIZATION_IMPORT_ERROR = """
{0} requires the pytorch-quantization library but it was not found in your environment. You can install it with pip:
`pip install pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com`
880
Please note that you may need to restart your runtime after installation.
881
882
883
884
885
"""

# docstyle-ignore
TENSORFLOW_PROBABILITY_IMPORT_ERROR = """
{0} requires the tensorflow_probability library but it was not found in your environment. You can install it with pip as
886
explained here: https://github.com/tensorflow/probability. Please note that you may need to restart your runtime after installation.
887
888
"""

889
890
891
892
# docstyle-ignore
TENSORFLOW_TEXT_IMPORT_ERROR = """
{0} requires the tensorflow_text library but it was not found in your environment. You can install it with pip as
explained here: https://www.tensorflow.org/text/guide/tf_text_intro.
893
Please note that you may need to restart your runtime after installation.
894
895
"""

896
897
898
899
900

# docstyle-ignore
PANDAS_IMPORT_ERROR = """
{0} requires the pandas library but it was not found in your environment. You can install it with pip as
explained here: https://pandas.pydata.org/pandas-docs/stable/getting_started/install.html.
901
Please note that you may need to restart your runtime after installation.
902
903
904
905
906
907
"""


# docstyle-ignore
PHONEMIZER_IMPORT_ERROR = """
{0} requires the phonemizer library but it was not found in your environment. You can install it with pip:
908
`pip install phonemizer`. Please note that you may need to restart your runtime after installation.
909
910
911
"""


912
913
914
# docstyle-ignore
SACREMOSES_IMPORT_ERROR = """
{0} requires the sacremoses library but it was not found in your environment. You can install it with pip:
915
`pip install sacremoses`. Please note that you may need to restart your runtime after installation.
916
917
918
"""


919
920
921
# docstyle-ignore
SCIPY_IMPORT_ERROR = """
{0} requires the scipy library but it was not found in your environment. You can install it with pip:
922
`pip install scipy`. Please note that you may need to restart your runtime after installation.
923
924
925
926
927
928
"""


# docstyle-ignore
SPEECH_IMPORT_ERROR = """
{0} requires the torchaudio library but it was not found in your environment. You can install it with pip:
929
`pip install torchaudio`. Please note that you may need to restart your runtime after installation.
930
931
932
933
934
"""

# docstyle-ignore
TIMM_IMPORT_ERROR = """
{0} requires the timm library but it was not found in your environment. You can install it with pip:
935
`pip install timm`. Please note that you may need to restart your runtime after installation.
936
937
"""

938
939
940
941
942
943
944
# docstyle-ignore
NATTEN_IMPORT_ERROR = """
{0} requires the natten library but it was not found in your environment. You can install it by referring to:
shi-labs.com/natten . You can also install it with pip (may take longer to build):
`pip install natten`. Please note that you may need to restart your runtime after installation.
"""

945
946
947
# docstyle-ignore
VISION_IMPORT_ERROR = """
{0} requires the PIL library but it was not found in your environment. You can install it with pip:
948
`pip install pillow`. Please note that you may need to restart your runtime after installation.
949
950
951
952
953
954
"""


# docstyle-ignore
PYTESSERACT_IMPORT_ERROR = """
{0} requires the PyTesseract library but it was not found in your environment. You can install it with pip:
955
`pip install pytesseract`. Please note that you may need to restart your runtime after installation.
956
957
958
959
960
"""

# docstyle-ignore
PYCTCDECODE_IMPORT_ERROR = """
{0} requires the pyctcdecode library but it was not found in your environment. You can install it with pip:
961
`pip install pyctcdecode`. Please note that you may need to restart your runtime after installation.
962
963
"""

964
965
966
# docstyle-ignore
ACCELERATE_IMPORT_ERROR = """
{0} requires the accelerate library but it was not found in your environment. You can install it with pip:
967
`pip install accelerate`. Please note that you may need to restart your runtime after installation.
968
969
"""

970
971
972
973
# docstyle-ignore
CCL_IMPORT_ERROR = """
{0} requires the torch ccl library but it was not found in your environment. You can install it with pip:
`pip install oneccl_bind_pt -f https://developer.intel.com/ipex-whl-stable`
974
Please note that you may need to restart your runtime after installation.
975
"""
976

977
978
979
980
981
DECORD_IMPORT_ERROR = """
{0} requires the decord library but it was not found in your environment. You can install it with pip: `pip install
decord`. Please note that you may need to restart your runtime after installation.
"""

Clémentine Fourrier's avatar
Clémentine Fourrier committed
982
983
984
985
986
CYTHON_IMPORT_ERROR = """
{0} requires the Cython library but it was not found in your environment. You can install it with pip: `pip install
Cython`. Please note that you may need to restart your runtime after installation.
"""

987
988
989
990
991
JIEBA_IMPORT_ERROR = """
{0} requires the jieba library but it was not found in your environment. You can install it with pip: `pip install
jieba`. Please note that you may need to restart your runtime after installation.
"""

992
993
BACKENDS_MAPPING = OrderedDict(
    [
NielsRogge's avatar
NielsRogge committed
994
        ("bs4", (is_bs4_available, BS4_IMPORT_ERROR)),
995
996
997
998
999
1000
1001
1002
1003
1004
        ("datasets", (is_datasets_available, DATASETS_IMPORT_ERROR)),
        ("detectron2", (is_detectron2_available, DETECTRON2_IMPORT_ERROR)),
        ("faiss", (is_faiss_available, FAISS_IMPORT_ERROR)),
        ("flax", (is_flax_available, FLAX_IMPORT_ERROR)),
        ("ftfy", (is_ftfy_available, FTFY_IMPORT_ERROR)),
        ("pandas", (is_pandas_available, PANDAS_IMPORT_ERROR)),
        ("phonemizer", (is_phonemizer_available, PHONEMIZER_IMPORT_ERROR)),
        ("protobuf", (is_protobuf_available, PROTOBUF_IMPORT_ERROR)),
        ("pyctcdecode", (is_pyctcdecode_available, PYCTCDECODE_IMPORT_ERROR)),
        ("pytesseract", (is_pytesseract_available, PYTESSERACT_IMPORT_ERROR)),
1005
        ("sacremoses", (is_sacremoses_available, SACREMOSES_IMPORT_ERROR)),
1006
1007
1008
1009
1010
1011
        ("pytorch_quantization", (is_pytorch_quantization_available, PYTORCH_QUANTIZATION_IMPORT_ERROR)),
        ("sentencepiece", (is_sentencepiece_available, SENTENCEPIECE_IMPORT_ERROR)),
        ("sklearn", (is_sklearn_available, SKLEARN_IMPORT_ERROR)),
        ("speech", (is_speech_available, SPEECH_IMPORT_ERROR)),
        ("tensorflow_probability", (is_tensorflow_probability_available, TENSORFLOW_PROBABILITY_IMPORT_ERROR)),
        ("tf", (is_tf_available, TENSORFLOW_IMPORT_ERROR)),
1012
        ("tensorflow_text", (is_tensorflow_text_available, TENSORFLOW_TEXT_IMPORT_ERROR)),
1013
        ("timm", (is_timm_available, TIMM_IMPORT_ERROR)),
1014
        ("natten", (is_natten_available, NATTEN_IMPORT_ERROR)),
1015
1016
        ("tokenizers", (is_tokenizers_available, TOKENIZERS_IMPORT_ERROR)),
        ("torch", (is_torch_available, PYTORCH_IMPORT_ERROR)),
NielsRogge's avatar
NielsRogge committed
1017
        ("torchvision", (is_torchvision_available, TORCHVISION_IMPORT_ERROR)),
1018
1019
        ("vision", (is_vision_available, VISION_IMPORT_ERROR)),
        ("scipy", (is_scipy_available, SCIPY_IMPORT_ERROR)),
1020
        ("accelerate", (is_accelerate_available, ACCELERATE_IMPORT_ERROR)),
1021
        ("oneccl_bind_pt", (is_ccl_available, CCL_IMPORT_ERROR)),
1022
        ("decord", (is_decord_available, DECORD_IMPORT_ERROR)),
Clémentine Fourrier's avatar
Clémentine Fourrier committed
1023
        ("cython", (is_cython_available, CYTHON_IMPORT_ERROR)),
1024
        ("jieba", (is_jieba_available, JIEBA_IMPORT_ERROR)),
1025
1026
1027
1028
1029
1030
1031
1032
1033
    ]
)


def requires_backends(obj, backends):
    if not isinstance(backends, (list, tuple)):
        backends = [backends]

    name = obj.__name__ if hasattr(obj, "__name__") else obj.__class__.__name__
1034
1035
1036
1037
1038
1039
1040
1041
1042

    # Raise an error for users who might not realize that classes without "TF" are torch-only
    if "torch" in backends and "tf" not in backends and not is_torch_available() and is_tf_available():
        raise ImportError(PYTORCH_IMPORT_ERROR_WITH_TF.format(name))

    # Raise the inverse error for PyTorch users trying to load TF classes
    if "tf" in backends and "torch" not in backends and is_torch_available() and not is_tf_available():
        raise ImportError(TF_IMPORT_ERROR_WITH_PYTORCH.format(name))

1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
    checks = (BACKENDS_MAPPING[backend] for backend in backends)
    failed = [msg.format(name) for available, msg in checks if not available()]
    if failed:
        raise ImportError("".join(failed))


class DummyObject(type):
    """
    Metaclass for the dummy objects. Any class inheriting from it will return the ImportError generated by
    `requires_backend` each time a user tries to access any method of that class.
    """

1055
    def __getattribute__(cls, key):
1056
        if key.startswith("_") and key != "_from_config":
1057
            return super().__getattribute__(key)
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
        requires_backends(cls, cls._backends)


def is_torch_fx_proxy(x):
    if is_torch_fx_available():
        import torch.fx

        return isinstance(x, torch.fx.Proxy)
    return False


class _LazyModule(ModuleType):
    """
    Module class that surfaces all objects but only performs associated imports when the objects are requested.
    """

    # Very heavily inspired by optuna.integration._IntegrationModule
    # https://github.com/optuna/optuna/blob/master/optuna/integration/__init__.py
    def __init__(self, name, module_file, import_structure, module_spec=None, extra_objects=None):
        super().__init__(name)
        self._modules = set(import_structure.keys())
        self._class_to_module = {}
        for key, values in import_structure.items():
            for value in values:
                self._class_to_module[value] = key
        # Needed for autocompletion in an IDE
        self.__all__ = list(import_structure.keys()) + list(chain(*import_structure.values()))
        self.__file__ = module_file
        self.__spec__ = module_spec
        self.__path__ = [os.path.dirname(module_file)]
        self._objects = {} if extra_objects is None else extra_objects
        self._name = name
        self._import_structure = import_structure

    # Needed for autocompletion in an IDE
    def __dir__(self):
        result = super().__dir__()
        # The elements of self.__all__ that are submodules may or may not be in the dir already, depending on whether
        # they have been accessed or not. So we only add the elements of self.__all__ that are not already in the dir.
        for attr in self.__all__:
            if attr not in result:
                result.append(attr)
        return result

    def __getattr__(self, name: str) -> Any:
        if name in self._objects:
            return self._objects[name]
        if name in self._modules:
            value = self._get_module(name)
        elif name in self._class_to_module.keys():
            module = self._get_module(self._class_to_module[name])
            value = getattr(module, name)
        else:
            raise AttributeError(f"module {self.__name__} has no attribute {name}")

        setattr(self, name, value)
        return value

    def _get_module(self, module_name: str):
        try:
            return importlib.import_module("." + module_name, self.__name__)
        except Exception as e:
            raise RuntimeError(
Sylvain Gugger's avatar
Sylvain Gugger committed
1121
1122
                f"Failed to import {self.__name__}.{module_name} because of the following error (look up to see its"
                f" traceback):\n{e}"
1123
1124
1125
1126
            ) from e

    def __reduce__(self):
        return (self.__class__, (self._name, self.__file__, self._import_structure))
1127
1128
1129
1130


class OptionalDependencyNotAvailable(BaseException):
    """Internally used error class for signalling an optional dependency was not found."""
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149


def direct_transformers_import(path: str, file="__init__.py") -> ModuleType:
    """Imports transformers directly

    Args:
        path (`str`): The path to the source file
        file (`str`, optional): The file to join with the path. Defaults to "__init__.py".

    Returns:
        `ModuleType`: The resulting imported module
    """
    name = "transformers"
    location = os.path.join(path, file)
    spec = importlib.util.spec_from_file_location(name, location, submodule_search_locations=[path])
    module = importlib.util.module_from_spec(spec)
    spec.loader.exec_module(module)
    module = sys.modules[name]
    return module