def from_pretrained(cls, pretrained_model_name_or_path:str):
"""
Setting up this method will enable to load directly from huggingface hub just like other HF models are loaded
"""
model_id = pretrained_model_name_or_path
if len(model_id.split("/")) == 1:
name = model_id
else:
username, name = model_id.split("/")
if name in os.listdir():
print("LOADING config & model weights from local directory")
config_file = os.path.join(name, "config.json")
model_file = os.path.join(name, "pytorch_model.bin")
else:
config_url = hf_bucket_url(model_id, filename="config.json")
config_file = cached_path(config_url)
# downloading & load only the adapter weights from huggingface hub
# and corresponding bert weights will be loaded when class is getting initiated
model_url = hf_bucket_url(model_id, filename="pytorch_model.bin")
model_file = cached_path(model_url)
with open(config_file, "r", encoding="utf-8") as f:
config = json.load(f)
config = Dict.from_nested_dict(config)
state_dict = torch.load(model_file, map_location="cpu")
# randomly initializing model from given config with bert weights restored
model = cls(config)
# now restoring adapter weights
model.load_state_dict(state_dict, strict=False)
model.eval()
return model
def test_revision_not_found(self):
# Valid file but missing revision
url = hf_bucket_url(MODEL_ID,
filename=CONFIG_NAME,
revision=REVISION_ID_INVALID)
with self.assertRaisesRegex(RevisionNotFoundError, "404 Client Error"):
_ = get_from_cache(url)
def model_file_from_short_name(short_name, model_type):
"""Get model weights file by short name"""
model_file = hf_bucket_url(
short_name,
filename=(TF2_WEIGHTS_NAME if model_type == 'tf' else WEIGHTS_NAME),
use_cdn=True)
return model_file
def test_lfs_object(self):
url = hf_bucket_url(MODEL_ID,
filename=WEIGHTS_NAME,
revision=REVISION_ID_DEFAULT)
filepath = get_from_cache(url, force_download=True)
metadata = filename_to_url(filepath)
self.assertEqual(metadata, (url, f'"{PINNED_SHA256}"'))
def test_standard_object_rev(self):
# Same object, but different revision
url = hf_bucket_url(MODEL_ID,
filename=CONFIG_NAME,
revision=REVISION_ID_ONE_SPECIFIC_COMMIT)
filepath = get_from_cache(url, force_download=True)
metadata = filename_to_url(filepath)
self.assertNotEqual(metadata[1], f'"{PINNED_SHA1}"')
def convert_pt_checkpoint_to_tf(model_type,
pytorch_checkpoint_path,
config_file,
tf_dump_path,
compare_with_pt_model=False,
use_cached_models=True):
if model_type not in MODEL_CLASSES:
raise ValueError(
"Unrecognized model type, should be one of {}.".format(
list(MODEL_CLASSES.keys())))
config_class, model_class, pt_model_class, aws_config_map = MODEL_CLASSES[
model_type]
# Initialise TF model
if config_file in aws_config_map:
config_file = cached_path(aws_config_map[config_file],
force_download=not use_cached_models)
config = config_class.from_json_file(config_file)
config.output_hidden_states = True
config.output_attentions = True
print("Building TensorFlow model from configuration: {}".format(
str(config)))
tf_model = model_class(config)
# Load weights from tf checkpoint
if pytorch_checkpoint_path in aws_config_map.keys():
pytorch_checkpoint_url = hf_bucket_url(pytorch_checkpoint_path,
filename=WEIGHTS_NAME)
pytorch_checkpoint_path = cached_path(
pytorch_checkpoint_url, force_download=not use_cached_models)
# Load PyTorch checkpoint in tf2 model:
tf_model = load_pytorch_checkpoint_in_tf2_model(tf_model,
pytorch_checkpoint_path)
if compare_with_pt_model:
tfo = tf_model(tf_model.dummy_inputs,
training=False) # build the network
state_dict = torch.load(pytorch_checkpoint_path, map_location="cpu")
pt_model = pt_model_class.from_pretrained(
pretrained_model_name_or_path=None,
config=config,
state_dict=state_dict)
with torch.no_grad():
pto = pt_model(**pt_model.dummy_inputs)
np_pt = pto[0].numpy()
np_tf = tfo[0].numpy()
diff = np.amax(np.abs(np_pt - np_tf))
print("Max absolute difference between models outputs {}".format(diff))
assert diff <= 2e-2, "Error, model absolute difference is >2e-2: {}".format(
diff)
# Save pytorch-model
print("Save TensorFlow model to {}".format(tf_dump_path))
tf_model.save_weights(tf_dump_path, save_format="h5")
def download_file_from_hf(pretrained_model_name_or_path: str,
file_name: str) -> str:
# Load model
if pretrained_model_name_or_path is not None:
if os.path.isdir(pretrained_model_name_or_path):
if os.path.isfile(
os.path.join(pretrained_model_name_or_path, file_name)):
# Load from a PyTorch checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
file_name)
else:
raise EnvironmentError(
"Error no file named {} found in directory {}".format(
file_name,
pretrained_model_name_or_path,
))
elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(
pretrained_model_name_or_path):
archive_file = pretrained_model_name_or_path
else:
archive_file = hf_bucket_url(
pretrained_model_name_or_path,
filename=file_name,
revision=None,
mirror=None,
)
try:
# Load from URL or cache if already cached
resolved_archive_file = cached_path(
archive_file,
cache_dir=None,
force_download=False,
proxies=None,
resume_download=False,
local_files_only=False,
)
except EnvironmentError as err:
logger.error(err)
msg = (
f"Can't load weights for '{pretrained_model_name_or_path}'. Make sure that:\n\n"
f"- '{pretrained_model_name_or_path}' is a correct model identifier listed on"
f"'https://huggingface.co/models'\n\n"
f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a"
f"file named one of {file_name}.\n\n")
raise EnvironmentError(msg)
if resolved_archive_file == archive_file:
logger.info("loading weights file {}".format(archive_file))
else:
logger.info("loading weights file {} from cache at {}".format(
archive_file, resolved_archive_file))
else:
resolved_archive_file = None
return resolved_archive_file
def load_cached_hf_parameters(model_name_or_path, cache_dir):
archive_file = hf_bucket_url(
model_name_or_path,
filename='pytorch_model.bin'
)
resolved_archive_file = cached_path(
archive_file,
cache_dir=cache_dir
)
state_dict = torch.load(resolved_archive_file, map_location="cpu")
return state_dict
def _get_config_dict(cls, path, **kw):
local_files_only = kw.pop("local_files_only", False)
from_pipeline = kw.pop("_from_pipeline", None)
user_agent = {
"file_type": "config",
"from_auto_class": kw.pop("_from_auto", False)
}
if from_pipeline is not None:
user_agent["using_pipeline"] = from_pipeline
if is_offline_mode() and not local_files_only:
log.info("Offline mode: forcing local_files_only=True")
local_files_only = True
path = str(path)
if os.path.isfile(path) or is_remote_url(path):
x = path
else:
f = kw.pop("_configuration_file", CONFIG_NAME)
if os.path.isdir(path):
x = os.path.join(path, f)
else:
x = hf_bucket_url(path,
filename=f,
revision=kw.pop("revision", None),
mirror=None)
try:
x2 = cached_path(
x,
cache_dir=kw.pop("cache_dir", None),
force_download=kw.pop("force_download", False),
proxies=kw.pop("proxies", None),
resume_download=kw.pop("resume_download", False),
local_files_only=local_files_only,
use_auth_token=kw.pop("use_auth_token", None),
user_agent=user_agent,
)
except RepositoryNotFoundError as e:
raise OSError() from e
except RevisionNotFoundError as e:
raise OSError() from e
except EntryNotFoundError as e:
raise OSError() from e
except HTTPError as e:
raise OSError() from e
except OSError as e:
raise e
try:
y = cls._dict_from_json_file(x2)
except (json.JSONDecodeError, UnicodeDecodeError) as e:
raise OSError() from e
if x2 == x:
log.info(f"loading {x}")
else:
log.info(f"loading {x} from cache at {x2}")
return y, kw
def load_model_from_cache(model_name_or_path, model_arch, cache_dir, filename,
config):
url = hf_bucket_url(model_name_or_path, filename=filename)
path = cached_path(url, cache_dir=cache_dir) + "." + model_arch
xml_path = path + ".xml"
bin_path = path + ".bin"
model = None
if os.path.exists(xml_path) and os.path.exists(bin_path):
logger.info(f"Load OpenVINO model from cache: {xml_path}")
model = load_ov_model_from_ir(xml_path, bin_path, config)
return model, path
def is_pretrained_model(model_name):
# check if it's a built-in pre-trained config:
if model_name in ALL_PRETRAINED_CONFIG_ARCHIVE_MAP:
return True
# check if it's a model on the huggingface model hub:
url = hf_bucket_url(model_name, CONFIG_NAME)
r = requests.head(url)
if r.status_code == 200:
return True
return False
def tokenizer_files_from_short_name(short_name):
"""Get all possible files for a tokenizer model by short name"""
use_fast = False
config = AutoConfig.from_pretrained(short_name)
vocab_files = []
for config_class, (tokenizer_class_py,
tokenizer_class_fast) in TOKENIZER_MAPPING.items():
if isinstance(config, config_class):
tokenizer_class = tokenizer_class_fast if (
use_fast and tokenizer_class_fast) else tokenizer_class_py
vocab_files = list(tokenizer_class.vocab_files_names.values())
additional_files = [
ADDED_TOKENS_FILE, SPECIAL_TOKENS_MAP_FILE, TOKENIZER_CONFIG_FILE,
FULL_TOKENIZER_FILE
]
tokenizer_files = []
for filename in vocab_files + additional_files:
tokenizer_files.append(
hf_bucket_url(short_name, filename=filename, use_cdn=False))
return tokenizer_files
def _check_and_rename_pretrained_model_file(pretrained_model_dir, model_id,
file_name, use_cdn):
target_file_path = join_path(pretrained_model_dir, model_id, file_name)
if os.path.exists(target_file_path):
return True
file_url = hf_bucket_url(model_id, file_name, use_cdn=use_cdn)
file_dir_path = join_path(pretrained_model_dir, model_id)
url_file_name = url_to_filename(file_url)
matching_files = [
file
for file in fnmatch.filter(os.listdir(file_dir_path), url_file_name +
".*")
if not file.endswith(".json") and not file.endswith(".lock")
]
if len(matching_files) > 0:
found_file_name = join_path(file_dir_path, matching_files[-1])
os.rename(found_file_name, target_file_path)
return True
return False
import numpy as np
import torch
import subprocess
config_path = BART_PRETRAINED_CONFIG_ARCHIVE_MAP['bart-large-xsum']
vocab_path = vocab_url
merges_path = merges_url
weights_path = 'bart-large-xsum'
target_path = Path.home() / 'rustbert' / 'bart-large-xsum'
temp_config = get_from_cache(config_path)
temp_vocab = get_from_cache(vocab_path)
temp_merges = get_from_cache(merges_path)
temp_weights = get_from_cache(
hf_bucket_url(weights_path, filename="pytorch_model.bin", use_cdn=True))
os.makedirs(str(target_path), exist_ok=True)
config_path = str(target_path / 'config.json')
vocab_path = str(target_path / 'vocab.txt')
merges_path = str(target_path / 'merges.txt')
model_path = str(target_path / 'model.bin')
shutil.copy(temp_config, config_path)
shutil.copy(temp_vocab, vocab_path)
shutil.copy(temp_merges, merges_path)
shutil.copy(temp_weights, model_path)
weights = torch.load(temp_weights, map_location='cpu')
nps = {}
def from_pretrained(cls, pretrained_model_name_or_path, *model_args,
**kwargs):
"""Instantiate a pretrained pytorch model from a pre-trained model configuration.
The model is set in evaluation mode by default using ``model.eval()`` (Dropout modules are deactivated)
To train the model, you should first set it back in training mode with ``model.train()``
The warning ``Weights from XXX not initialized from pretrained model`` means that the weights of XXX do not come pre-trained with the rest of the model.
It is up to you to train those weights with a downstream fine-tuning task.
The warning ``Weights from XXX not used in YYY`` means that the layer XXX is not used by YYY, therefore those weights are discarded.
Parameters:
pretrained_model_name_or_path: either:
- a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
- a path to a `directory` containing model weights saved using :func:`~transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
- a path or url to a `tensorflow index checkpoint file` (e.g. `./tf_model/model.ckpt.index`). In this case, ``from_tf`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the TensorFlow checkpoint in a PyTorch model using the provided conversion scripts and loading the PyTorch model afterwards.
- None if you are both providing the configuration and state dictionary (resp. with keyword arguments ``config`` and ``state_dict``)
model_args: (`optional`) Sequence of positional arguments:
All remaning positional arguments will be passed to the underlying model's ``__init__`` method
config: (`optional`) instance of a class derived from :class:`~transformers.PretrainedConfig`:
Configuration for the model to use instead of an automatically loaded configuation. Configuration can be automatically loaded when:
- the model is a model provided by the library (loaded with the ``shortcut-name`` string of a pretrained model), or
- the model was saved using :func:`~transformers.PreTrainedModel.save_pretrained` and is reloaded by suppling the save directory.
- the model is loaded by suppling a local directory as ``pretrained_model_name_or_path`` and a configuration JSON file named `config.json` is found in the directory.
state_dict: (`optional`) dict:
an optional state dictionnary for the model to use instead of a state dictionary loaded from saved weights file.
This option can be used if you want to create a model from a pretrained configuration but load your own weights.
In this case though, you should check if using :func:`~transformers.PreTrainedModel.save_pretrained` and :func:`~transformers.PreTrainedModel.from_pretrained` is not a simpler option.
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
force_download: (`optional`) boolean, default False:
Force to (re-)download the model weights and configuration files and override the cached versions if they exists.
proxies: (`optional`) dict, default None:
A dictionary of proxy servers to use by protocol or endpoint, e.g.: {'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}.
The proxies are used on each request.
output_loading_info: (`optional`) boolean:
Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.
kwargs: (`optional`) Remaining dictionary of keyword arguments:
Can be used to update the configuration object (after it being loaded) and initiate the model. (e.g. ``output_attention=True``). Behave differently depending on whether a `config` is provided or automatically loaded:
- If a configuration is provided with ``config``, ``**kwargs`` will be directly passed to the underlying model's ``__init__`` method (we assume all relevant updates to the configuration have already been done)
- If a configuration is not provided, ``kwargs`` will be first passed to the configuration class initialization function (:func:`~transformers.PretrainedConfig.from_pretrained`). Each key of ``kwargs`` that corresponds to a configuration attribute will be used to override said attribute with the supplied ``kwargs`` value. Remaining keys that do not correspond to any configuration attribute will be passed to the underlying model's ``__init__`` function.
Examples::
model = BertModel.from_pretrained('bert-base-uncased') # Download model and configuration from S3 and cache.
model = BertModel.from_pretrained('./test/saved_model/') # E.g. model was saved using `save_pretrained('./test/saved_model/')`
model = BertModel.from_pretrained('bert-base-uncased', output_attention=True) # Update configuration during loading
assert model.config.output_attention == True
# Loading from a TF checkpoint file instead of a PyTorch model (slower)
config = BertConfig.from_json_file('./tf_model/my_tf_model_config.json')
model = BertModel.from_pretrained('./tf_model/my_tf_checkpoint.ckpt.index', from_tf=True, config=config)
"""
config = kwargs.pop('config', None)
state_dict = kwargs.pop('state_dict', None)
cache_dir = kwargs.pop('cache_dir', None)
from_tf = kwargs.pop('from_tf', False)
force_download = kwargs.pop('force_download', False)
proxies = kwargs.pop('proxies', None)
output_loading_info = kwargs.pop('output_loading_info', False)
random_init = kwargs.pop("random_init", False)
use_cdn = kwargs.pop("use_cdn", True)
local_files_only = kwargs.pop("local_files_only", False)
resume_download = kwargs.pop("resume_download", False)
proxies = kwargs.pop("proxies", None)
kwargs_config = kwargs.copy()
mapping_keys_state_dic = kwargs.pop("mapping_keys_state_dic", None)
kwargs_config.pop("mapping_keys_state_dic", None)
if config is None:
config, model_kwargs = cls.config_class.from_pretrained(
pretrained_model_name_or_path,
*model_args,
cache_dir=cache_dir,
return_unused_kwargs=True,
force_download=force_download,
**kwargs_config)
else:
model_kwargs = kwargs
# Load model
if pretrained_model_name_or_path is not None:
if os.path.isdir(pretrained_model_name_or_path):
if from_tf and os.path.isfile(
os.path.join(pretrained_model_name_or_path,
TF_WEIGHTS_NAME + ".index")):
# Load from a TF 1.0 checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
TF_WEIGHTS_NAME + ".index")
elif from_tf and os.path.isfile(
os.path.join(pretrained_model_name_or_path,
TF2_WEIGHTS_NAME)):
# Load from a TF 2.0 checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
TF2_WEIGHTS_NAME)
elif os.path.isfile(
os.path.join(pretrained_model_name_or_path,
WEIGHTS_NAME)):
# Load from a PyTorch checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
WEIGHTS_NAME)
else:
raise EnvironmentError(
"Error no file named {} found in directory {} or `from_tf` set to False"
.format(
[
WEIGHTS_NAME, TF2_WEIGHTS_NAME,
TF_WEIGHTS_NAME + ".index"
],
pretrained_model_name_or_path,
))
elif os.path.isfile(
pretrained_model_name_or_path) or is_remote_url(
pretrained_model_name_or_path):
archive_file = pretrained_model_name_or_path
elif os.path.isfile(pretrained_model_name_or_path + ".index"):
assert (
from_tf
), "We found a TensorFlow checkpoint at {}, please set from_tf to True to load from this checkpoint".format(
pretrained_model_name_or_path + ".index")
archive_file = pretrained_model_name_or_path + ".index"
else:
archive_file = hf_bucket_url(
pretrained_model_name_or_path,
filename=(TF2_WEIGHTS_NAME if from_tf else WEIGHTS_NAME),
use_cdn=use_cdn,
)
try:
# Load from URL or cache if already cached
resolved_archive_file = cached_path(
archive_file,
cache_dir=cache_dir,
force_download=force_download,
proxies=proxies,
resume_download=resume_download,
local_files_only=local_files_only,
)
if resolved_archive_file is None:
raise EnvironmentError
except EnvironmentError:
msg = (
f"Can't load weights for '{pretrained_model_name_or_path}'. Make sure that:\n\n"
f"- '{pretrained_model_name_or_path}' is a correct model identifier listed on 'https://huggingface.co/models'\n\n"
f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a file named one of {WEIGHTS_NAME}, {TF2_WEIGHTS_NAME}, {TF_WEIGHTS_NAME}.\n\n"
)
raise EnvironmentError(msg)
if resolved_archive_file == archive_file:
logger.info("loading weights file {}".format(archive_file))
else:
logger.info("loading weights file {} from cache at {}".format(
archive_file, resolved_archive_file))
else:
resolved_archive_file = None
# Instantiate model.
model = cls(config, *model_args, **model_kwargs)
if state_dict is None and not from_tf:
state_dict = torch.load(resolved_archive_file, map_location='cpu')
missing_keys = []
unexpected_keys = []
error_msgs = []
if from_tf:
if resolved_archive_file.endswith('.index'):
# Load from a TensorFlow 1.X checkpoint - provided by original authors
model = cls.load_tf_weights(
model, config,
resolved_archive_file[:-6]) # Remove the '.index'
else:
# Load from our TensorFlow 2.0 checkpoints
try:
from transformers import load_tf2_checkpoint_in_pytorch_model
model = load_tf2_checkpoint_in_pytorch_model(
model, resolved_archive_file, allow_missing_keys=True)
except ImportError as e:
logger.error(
"Loading a TensorFlow model in PyTorch, requires both PyTorch and TensorFlow to be installed. Please see "
"https://pytorch.org/ and https://www.tensorflow.org/install/ for installation instructions."
)
raise e
else:
# Convert old format to new format if needed from a PyTorch state_dict
old_keys = []
new_keys = []
for key in state_dict.keys():
new_key = None
if 'gamma' in key:
new_key = key.replace('gamma', 'weight')
if 'beta' in key:
new_key = key.replace('beta', 'bias')
if new_key:
old_keys.append(key)
new_keys.append(new_key)
for old_key, new_key in zip(old_keys, new_keys):
state_dict[new_key] = state_dict.pop(old_key)
# copy state_dict so _load_from_state_dict can modify it
metadata = getattr(state_dict, '_metadata', None)
state_dict = state_dict.copy()
if metadata is not None:
state_dict._metadata = metadata
# assert mapping_keys_state_dic is not None, "ERROR did not found mapping dicts for {} ".format(pretrained_model_name_or_path)
# mapping_keys_state_dic = {"roberta": "encoder", "lm_head": "head.mlm"}
if mapping_keys_state_dic is not None:
assert isinstance(mapping_keys_state_dic, dict), "ERROR "
print(
"INFO : from loading from pretrained method (assuming loading original google model : "
"need to rename some keys {})".format(
mapping_keys_state_dic))
state_dict = cls.adapt_state_dic_to_multitask(
state_dict,
keys_mapping=mapping_keys_state_dic,
add_prefix=pretrained_model_name_or_path ==
"asafaya/bert-base-arabic")
#pdb.set_trace()
def load(module, prefix=''):
local_metadata = {"version": 1}
if not prefix.startswith("head") or prefix.startswith(
"head.mlm"):
assert len(
missing_keys
) == 0, "ERROR {} missing keys in state_dict {}".format(
prefix, missing_keys)
else:
if len(missing_keys) == 0:
print(
"Warning {} missing keys in state_dict {} (warning expected for task-specific fine-tuning)"
.format(prefix, missing_keys))
module._load_from_state_dict(state_dict, prefix,
local_metadata, True,
missing_keys, unexpected_keys,
error_msgs)
for name, child in module._modules.items():
# load_params_only_ls = kwargs.get("load_params_only_ls ")
not_load_params_ls = kwargs.get(
"not_load_params_ls") if kwargs.get(
"not_load_params_ls") is not None else []
assert isinstance(
not_load_params_ls, list
), f"Argument error not_load_params_ls should be a list but is {not_load_params_ls}"
matching_not_load = []
# RANDOM-INIT
for pattern in not_load_params_ls:
matching = re.match(pattern, prefix + name)
if matching is not None:
matching_not_load.append(matching)
if len(matching_not_load) > 0:
# means there is at least one patter in not load pattern that matched --> so should load
print("MATCH not loading : {} parameters {} ".format(
prefix + name, not_load_params_ls))
if child is not None and len(matching_not_load) == 0:
#print("MODEL loading : child {} full {} ".format(name, prefix + name + '.'))
load(child, prefix + name + '.')
else:
print(
"MODEL not loading : child {} matching_not_load {} "
.format(child, matching_not_load))
# Make sure we are able to load base models as well as derived models (with heads)
start_prefix = ''
model_to_load = model
if not hasattr(model, cls.base_model_prefix) and any(
s.startswith(cls.base_model_prefix)
for s in state_dict.keys()):
start_prefix = cls.base_model_prefix + '.'
if hasattr(model, cls.base_model_prefix) and not any(
s.startswith(cls.base_model_prefix)
for s in state_dict.keys()):
model_to_load = getattr(model, cls.base_model_prefix)
if not random_init:
load(model_to_load, prefix=start_prefix)
else:
print("WARNING : RANDOM INTIALIZATION OF BERTMULTITASK")
if len(missing_keys) > 0:
logger.info(
"Weights of {} not initialized from pretrained model: {}".
format(model.__class__.__name__, missing_keys))
if len(unexpected_keys) > 0:
logger.info(
"Weights from pretrained model not used in {}: {}".format(
model.__class__.__name__, unexpected_keys))
if len(error_msgs) > 0:
raise RuntimeError(
'Error(s) in loading state_dict for {}:\n\t{}'.format(
model.__class__.__name__, "\n\t".join(error_msgs)))
if hasattr(model, 'tie_weights'):
model.tie_weights(
) # make sure word embedding weights are still tied
# Set model in evaluation mode to desactivate DropOut modules by default
model.eval()
if output_loading_info:
loading_info = {
"missing_keys": missing_keys,
"unexpected_keys": unexpected_keys,
"error_msgs": error_msgs
}
return model, loading_info
return model
def from_pretrained(cls, model_name_or_path, *model_args, **kwargs):
cache_dir = kwargs.get("cache_dir", None)
from_pt = kwargs.pop("from_pt", False)
from_tf = kwargs.pop("from_tf", False)
from_ov = kwargs.get("from_ov", not (from_pt | from_tf))
force_download = kwargs.get("force_download", False)
resume_download = kwargs.get("resume_download", False)
proxies = kwargs.get("proxies", None)
local_files_only = kwargs.get("local_files_only", False)
use_auth_token = kwargs.get("use_auth_token", None)
revision = kwargs.get("revision", None)
from_pipeline = kwargs.get("_from_pipeline", None)
from_auto_class = kwargs.get("_from_auto", False)
config = kwargs.get(
"config") if "config" in kwargs else AutoConfig.from_pretrained(
model_name_or_path)
if from_pt:
model = cls._pt_auto_model.from_pretrained(model_name_or_path,
*model_args, **kwargs)
net = load_ov_model_from_pytorch(model)
return OVPreTrainedModel(net, model.config)
elif from_tf:
model, cache_path = load_model_from_cache(model_name_or_path,
cls.__name__, cache_dir,
TF2_WEIGHTS_NAME, config)
if model is not None:
return model
model = cls._tf_auto_model.from_pretrained(model_name_or_path,
*model_args, **kwargs)
return load_ov_model_from_tf(model, cache_path)
user_agent = {
"file_type": "model",
"framework": "openvino",
"from_auto_class": from_auto_class
}
if from_pipeline is not None:
user_agent["using_pipeline"] = from_pipeline
# Load model
OV_BIN_NAME = OV_WEIGHTS_NAME.replace(".xml", ".bin")
if model_name_or_path is not None:
if os.path.isdir(model_name_or_path):
if (from_ov and os.path.isfile(
os.path.join(model_name_or_path, OV_WEIGHTS_NAME))
and os.path.isfile(
os.path.join(model_name_or_path, OV_BIN_NAME))):
# Load from an OpenVINO IR
archive_files = [
os.path.join(model_name_or_path, name)
for name in [OV_WEIGHTS_NAME, OV_BIN_NAME]
]
else:
raise EnvironmentError(
f"Error no files named {[OV_WEIGHTS_NAME, OV_BIN_NAME]} found in directory "
f"{model_name_or_path} or `from_ov` set to False")
# elif os.path.isfile(model_name_or_path) or is_remote_url(model_name_or_path):
# archive_file = model_name_or_path
else:
names = [OV_WEIGHTS_NAME, OV_BIN_NAME]
archive_files = [
hf_bucket_url(
model_name_or_path,
filename=name,
revision=revision,
) for name in names
]
# redirect to the cache, if necessary
try:
resolved_archive_files = [
cached_path(
archive_file,
cache_dir=cache_dir,
force_download=force_download,
proxies=proxies,
resume_download=resume_download,
local_files_only=local_files_only,
use_auth_token=use_auth_token,
user_agent=user_agent,
) for archive_file in archive_files
]
except EnvironmentError as err:
logger.error(err)
name = model_name_or_path
msg = (
f"Can't load weights for '{name}'. Make sure that:\n\n"
f"- '{name}' is a correct model identifier listed on 'https://huggingface.co/models'\n"
f" (make sure '{name}' is not a path to a local directory with something else, in that case)\n\n"
f"- or '{name}' is the correct path to a directory containing a file named {OV_WEIGHTS_NAME}.\n\n"
)
raise EnvironmentError(msg)
if resolved_archive_files == archive_files:
logger.info(f"loading weights file {archive_files}")
else:
logger.info(
f"loading weights file {archive_files} from cache at {resolved_archive_files}"
)
else:
resolved_archive_files = None
return load_ov_model_from_ir(*resolved_archive_files, config=config)
def get_pretrained_state_dict(pretrained_model_name_or_path, *model_args,
**kwargs):
"""Get PyTorch state dict via HuggingFace transformers library."""
config = kwargs.pop("config", None)
state_dict = kwargs.pop("state_dict", None)
cache_dir = kwargs.pop("cache_dir", None)
# from_tf = kwargs.pop("from_tf", False)
force_download = kwargs.pop("force_download", False)
resume_download = kwargs.pop("resume_download", False)
proxies = kwargs.pop("proxies", None)
output_loading_info = kwargs.pop("output_loading_info", False)
local_files_only = kwargs.pop("local_files_only", False)
use_cdn = kwargs.pop("use_cdn", True)
mirror = kwargs.pop("mirror", None)
if pretrained_model_name_or_path is not None:
if os.path.isdir(pretrained_model_name_or_path):
if os.path.isfile(
os.path.join(pretrained_model_name_or_path, WEIGHTS_NAME)):
# Load from a PyTorch checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
WEIGHTS_NAME)
else:
raise EnvironmentError(
"Error no file named {} found in directory {}".format(
WEIGHTS_NAME,
pretrained_model_name_or_path,
))
elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(
pretrained_model_name_or_path):
archive_file = pretrained_model_name_or_path
elif os.path.isfile(pretrained_model_name_or_path + ".index"):
assert False, "Loading TensorFlow checkpoints is not supported"
else:
archive_file = hf_bucket_url(
pretrained_model_name_or_path,
filename=WEIGHTS_NAME,
use_cdn=use_cdn,
mirror=mirror,
)
try:
# Load from URL or cache if already cached
resolved_archive_file = cached_path(
archive_file,
cache_dir=cache_dir,
force_download=force_download,
proxies=proxies,
resume_download=resume_download,
local_files_only=local_files_only,
)
if resolved_archive_file is None:
raise EnvironmentError
except EnvironmentError:
msg = (
f"Can't load weights for '{pretrained_model_name_or_path}'. Make sure that:\n\n"
f"- '{pretrained_model_name_or_path}' is a correct model identifier listed on 'https://huggingface.co/models'\n\n"
f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a file named {WEIGHTS_NAME}.\n\n"
)
raise EnvironmentError(msg)
if resolved_archive_file == archive_file:
print("loading weights file {}".format(archive_file))
else:
print("loading weights file {} from cache at {}".format(
archive_file, resolved_archive_file))
else:
resolved_archive_file = None
if state_dict is None:
try:
state_dict = torch.load(resolved_archive_file, map_location="cpu")
except Exception:
raise OSError(
"Unable to load weights from pytorch checkpoint file.")
return state_dict
def test_file_not_found(self):
# Valid revision (None) but missing file.
url = hf_bucket_url(MODEL_ID, filename="missing.bin")
with self.assertRaisesRegex(requests.exceptions.HTTPError,
"404 Client Error"):
_ = get_from_cache(url)
def get_config_dict(cls,
pretrained_model_name_or_path: str,
pretrained_config_archive_map: Optional[Dict] = None,
**kwargs) -> Tuple[Dict, Dict]:
"""
From a `pretrained_model_name_or_path`, resolve to a dictionary of parameters, to be used
for instantiating a Config using `from_dict`.
Parameters:
pretrained_model_name_or_path (:obj:`string`):
The identifier of the pre-trained checkpoint from which we want the dictionary of parameters.
pretrained_config_archive_map: (:obj:`Dict[str, str]`, `optional`) Dict:
A map of `shortcut names` to `url`. By default, will use the current class attribute.
Returns:
:obj:`Tuple[Dict, Dict]`: The dictionary that will be used to instantiate the configuration object.
"""
cache_dir = kwargs.pop("cache_dir", None)
force_download = kwargs.pop("force_download", False)
resume_download = kwargs.pop("resume_download", False)
proxies = kwargs.pop("proxies", None)
local_files_only = kwargs.pop("local_files_only", False)
if pretrained_config_archive_map is None:
pretrained_config_archive_map = cls.pretrained_config_archive_map
if pretrained_model_name_or_path in pretrained_config_archive_map:
config_file = pretrained_config_archive_map[
pretrained_model_name_or_path]
elif os.path.isdir(pretrained_model_name_or_path):
config_file = os.path.join(pretrained_model_name_or_path,
CONFIG_NAME)
elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(
pretrained_model_name_or_path):
config_file = pretrained_model_name_or_path
else:
config_file = hf_bucket_url(pretrained_model_name_or_path,
postfix=CONFIG_NAME)
try:
# Load from URL or cache if already cached
resolved_config_file = cached_path(
config_file,
cache_dir=cache_dir,
force_download=force_download,
proxies=proxies,
resume_download=resume_download,
local_files_only=local_files_only,
)
# Load config dict
if resolved_config_file is None:
raise EnvironmentError
config_dict = cls._dict_from_json_file(resolved_config_file)
except EnvironmentError:
if pretrained_model_name_or_path in pretrained_config_archive_map:
msg = "Couldn't reach server at '{}' to download pretrained model configuration file.".format(
config_file)
else:
msg = (
"Can't load '{}'. Make sure that:\n\n"
"- '{}' is a correct model identifier listed on 'https://huggingface.co/models'\n\n"
"- or '{}' is the correct path to a directory containing a '{}' file\n\n"
.format(
pretrained_model_name_or_path,
pretrained_model_name_or_path,
pretrained_model_name_or_path,
CONFIG_NAME,
))
raise EnvironmentError(msg)
except json.JSONDecodeError:
msg = (
"Couldn't reach server at '{}' to download configuration file or "
"configuration file is not a valid JSON file. "
"Please check network or file content here: {}.".format(
config_file, resolved_config_file))
raise EnvironmentError(msg)
if resolved_config_file == config_file:
logger.info("loading configuration file {}".format(config_file))
else:
logger.info(
"loading configuration file {} from cache at {}".format(
config_file, resolved_config_file))
return config_dict, kwargs
from pathlib import Path
import shutil
import os
import numpy as np
import torch
import subprocess
config_path = T5_PRETRAINED_CONFIG_ARCHIVE_MAP['t5-base']
vocab_path = PRETRAINED_VOCAB_FILES_MAP['vocab_file']['t5-base']
weights_path = 't5-base'
target_path = Path.home() / 'rustbert' / 't5-base'
temp_config = get_from_cache(config_path)
temp_vocab = get_from_cache(vocab_path)
temp_weights = get_from_cache(hf_bucket_url(weights_path, filename="pytorch_model.bin", use_cdn=True))
os.makedirs(str(target_path), exist_ok=True)
config_path = str(target_path / 'config.json')
vocab_path = str(target_path / 'spiece.model')
model_path = str(target_path / 'model.bin')
shutil.copy(temp_config, config_path)
shutil.copy(temp_vocab, vocab_path)
shutil.copy(temp_weights, model_path)
weights = torch.load(temp_weights, map_location='cpu')
nps = {}
for k, v in weights.items():
k = k.replace("gamma", "weight").replace("beta", "bias")
def config_file_from_short_name(short_name):
return hf_bucket_url(short_name, filename=CONFIG_NAME, use_cdn=False)
import shutil
import os
import numpy as np
import torch
import subprocess
config_path = BERT_PRETRAINED_CONFIG_ARCHIVE_MAP["bert-base-uncased"]
vocab_path = PRETRAINED_VOCAB_FILES_MAP["vocab_file"]["bert-base-uncased"]
weights_path = "bert-base-uncased"
target_path = Path().absolute()
temp_config = get_from_cache(config_path)
temp_vocab = get_from_cache(vocab_path)
temp_weights = get_from_cache(
hf_bucket_url(weights_path, filename="pytorch_model.bin"))
os.makedirs(str(target_path), exist_ok=True)
config_path = str(target_path / 'config.json')
vocab_path = str(target_path / 'vocab.txt')
model_path = str(target_path / 'model.bin')
shutil.copy(temp_config, config_path)
shutil.copy(temp_vocab, vocab_path)
shutil.copy(temp_weights, model_path)
weights = torch.load(temp_weights, map_location='cpu')
nps = {}
for k, v in weights.items():
k = k.replace("gamma", "weight").replace("beta", "bias")
def from_pretrained_detailed(model_class, pretrained_model_name_or_path,
*model_args, **kwargs):
r"""Instantiate a pretrained TF 2.0 model from a pre-trained model configuration.
The warning ``Weights from XXX not initialized from pretrained model`` means that the weights of XXX do not come pre-trained with the rest of the model.
It is up to you to train those weights with a downstream fine-tuning task.
The warning ``Weights from XXX not used in YYY`` means that the layer XXX is not used by YYY, therefore those weights are discarded.
Parameters:
pretrained_model_name_or_path: either:
- a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
- a string with the `identifier name` of a pre-trained model that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
- a path to a `directory` containing model weights saved using :func:`~transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
- a path or url to a `PyTorch state_dict save file` (e.g. `./pt_model/pytorch_model.bin`). In this case, ``from_pt`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the PyTorch checkpoint in a TensorFlow model using the provided conversion scripts and loading the TensorFlow model afterwards.
model_args: (`optional`) Sequence of positional arguments:
All remaning positional arguments will be passed to the underlying model's ``__init__`` method
config: (`optional`) one of:
- an instance of a class derived from :class:`~transformers.PretrainedConfig`, or
- a string valid as input to :func:`~transformers.PretrainedConfig.from_pretrained()`
Configuration for the model to use instead of an automatically loaded configuation. Configuration can be automatically loaded when:
- the model is a model provided by the library (loaded with the ``shortcut-name`` string of a pretrained model), or
- the model was saved using :func:`~transformers.PreTrainedModel.save_pretrained` and is reloaded by suppling the save directory.
- the model is loaded by suppling a local directory as ``pretrained_model_name_or_path`` and a configuration JSON file named `config.json` is found in the directory.
from_pt: (`optional`) boolean, default False:
Load the model weights from a PyTorch state_dict save file (see docstring of pretrained_model_name_or_path argument).
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
force_download: (`optional`) boolean, default False:
Force to (re-)download the model weights and configuration files and override the cached versions if they exists.
resume_download: (`optional`) boolean, default False:
Do not delete incompletely recieved file. Attempt to resume the download if such a file exists.
proxies: (`optional`) dict, default None:
A dictionary of proxy servers to use by protocol or endpoint, e.g.: {'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}.
The proxies are used on each request.
output_loading_info: (`optional`) boolean:
Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.
kwargs: (`optional`) Remaining dictionary of keyword arguments:
Can be used to update the configuration object (after it being loaded) and initiate the model. (e.g. ``output_attention=True``). Behave differently depending on whether a `config` is provided or automatically loaded:
- If a configuration is provided with ``config``, ``**kwargs`` will be directly passed to the underlying model's ``__init__`` method (we assume all relevant updates to the configuration have already been done)
- If a configuration is not provided, ``kwargs`` will be first passed to the configuration class initialization function (:func:`~transformers.PretrainedConfig.from_pretrained`). Each key of ``kwargs`` that corresponds to a configuration attribute will be used to override said attribute with the supplied ``kwargs`` value. Remaining keys that do not correspond to any configuration attribute will be passed to the underlying model's ``__init__`` function.
- If layer pruning is supported, ``layer_pruning`` will passed as a dictionary contains layer pruning configurations as follows:
- strategy:
can be one of these values: {`top`, `buttom`, `symmetric`, `alternate`, `custom`}
- k:
is the number of layers to prune. mandatory if strategy is one of {`top`, `buttom`, `symmetric`, `alternate`}
- layers_indexes:
is array of layers indexs to prune. mandatory if strategy is `custom`
- is_odd:
is odd alternate or not. mandatory if strategy is `alternate`
Examples::
# For example purposes. Not runnable.
model = BertModel.from_pretrained('bert-base-uncased') # Download model and configuration from S3 and cache.
model = BertModel.from_pretrained('./test/saved_model/') # E.g. model was saved using `save_pretrained('./test/saved_model/')`
model = BertModel.from_pretrained('bert-base-uncased', output_attention=True) # Update configuration during loading
assert model.config.output_attention == True
# Loading from a TF checkpoint file instead of a PyTorch model (slower)
config = BertConfig.from_json_file('./tf_model/my_tf_model_config.json')
model = BertModel.from_pretrained('./tf_model/my_tf_checkpoint.ckpt.index', from_pt=True, config=config)
"""
config = kwargs.pop("config", None)
cache_dir = kwargs.pop("cache_dir", None)
from_pt = kwargs.pop("from_pt", False)
force_download = kwargs.pop("force_download", False)
resume_download = kwargs.pop("resume_download", False)
proxies = kwargs.pop("proxies", None)
output_loading_info = kwargs.pop("output_loading_info", False)
local_files_only = kwargs.pop("local_files_only", False)
use_cdn = kwargs.pop("use_cdn", True)
# mwahdan: Read layer_pruning config if exist
layer_pruning = kwargs.pop("layer_pruning", None)
# Load config if we don't provide a configuration
if not isinstance(config, PretrainedConfig):
config_path = config if config is not None else pretrained_model_name_or_path
config, model_kwargs = model_class.config_class.from_pretrained(
config_path,
*model_args,
cache_dir=cache_dir,
return_unused_kwargs=True,
force_download=force_download,
resume_download=resume_download,
proxies=proxies,
local_files_only=local_files_only,
**kwargs,
)
else:
model_kwargs = kwargs
# Load model
if pretrained_model_name_or_path is not None:
if os.path.isdir(pretrained_model_name_or_path):
if os.path.isfile(
os.path.join(pretrained_model_name_or_path,
TF2_WEIGHTS_NAME)):
# Load from a TF 2.0 checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
TF2_WEIGHTS_NAME)
elif from_pt and os.path.isfile(
os.path.join(pretrained_model_name_or_path, WEIGHTS_NAME)):
# Load from a PyTorch checkpoint
archive_file = os.path.join(pretrained_model_name_or_path,
WEIGHTS_NAME)
else:
raise EnvironmentError(
"Error no file named {} found in directory {} or `from_pt` set to False"
.format([WEIGHTS_NAME, TF2_WEIGHTS_NAME],
pretrained_model_name_or_path))
elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(
pretrained_model_name_or_path):
archive_file = pretrained_model_name_or_path
elif os.path.isfile(pretrained_model_name_or_path + ".index"):
archive_file = pretrained_model_name_or_path + ".index"
else:
archive_file = hf_bucket_url(
pretrained_model_name_or_path,
filename=(WEIGHTS_NAME if from_pt else TF2_WEIGHTS_NAME),
)
try:
# Load from URL or cache if already cached
resolved_archive_file = cached_path(
archive_file,
cache_dir=cache_dir,
force_download=force_download,
proxies=proxies,
resume_download=resume_download,
local_files_only=local_files_only,
)
if resolved_archive_file is None:
raise EnvironmentError
except EnvironmentError:
msg = (
f"Can't load weights for '{pretrained_model_name_or_path}'. Make sure that:\n\n"
f"- '{pretrained_model_name_or_path}' is a correct model identifier listed on 'https://huggingface.co/models'\n\n"
f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a file named one of {TF2_WEIGHTS_NAME}, {WEIGHTS_NAME}.\n\n"
)
raise EnvironmentError(msg)
if resolved_archive_file == archive_file:
logger.info("loading weights file {}".format(archive_file))
else:
logger.info("loading weights file {} from cache at {}".format(
archive_file, resolved_archive_file))
else:
resolved_archive_file = None
# mwahdan: Modify config
if layer_pruning:
layer_pruning_k = layer_pruning_layers_indexes = layer_pruning_is_odd = None
layer_pruning_strategy = get_mandatory_parameter(
'strategy', layer_pruning)
if layer_pruning_strategy in {'top', 'buttom', 'symmetric'}:
layer_pruning_k = get_mandatory_parameter('k', layer_pruning)
config, original_num_layers = modify_num_of_layers(
config, k=layer_pruning_k)
elif layer_pruning_strategy == 'custom':
layer_pruning_layers_indexes = get_mandatory_parameter(
'layers_indexes', layer_pruning)
config, original_num_layers = modify_num_of_layers(
config, layers_indexes=layer_pruning_layers_indexes)
elif layer_pruning_strategy == 'alternate':
layer_pruning_k = get_mandatory_parameter('k', layer_pruning)
layer_pruning_is_odd = get_mandatory_parameter(
'is_odd', layer_pruning)
config, original_num_layers = modify_num_of_layers(
config, k=layer_pruning_k, is_alternate=True)
else:
raise Exception('`%s` is not a supported layer pruning strategy' %
layer_pruning_strategy)
# Instantiate model.
model = model_class(config, *model_args, **model_kwargs)
# mwahdan: Rename layers
if layer_pruning:
model = rename_layers_in_strategy(model, layer_pruning_strategy,
original_num_layers, layer_pruning_k,
layer_pruning_layers_indexes,
layer_pruning_is_odd)
if from_pt:
# Load from a PyTorch checkpoint
model = load_pytorch_checkpoint_in_tf2_model(model,
resolved_archive_file,
allow_missing_keys=True)
# mwahdan: Rename layers
if layer_pruning is not None:
model = rename_layers(model)
return model
model(model.dummy_inputs,
training=False) # build the network with dummy inputs
assert os.path.isfile(
resolved_archive_file), "Error retrieving file {}".format(
resolved_archive_file)
# 'by_name' allow us to do transfer learning by skipping/adding layers
# see https://github.com/tensorflow/tensorflow/blob/00fad90125b18b80fe054de1055770cfb8fe4ba3/tensorflow/python/keras/engine/network.py#L1339-L1357
try:
# added skip_mismatch=True because we will prune full layers
model.load_weights(resolved_archive_file,
by_name=True,
skip_mismatch=True)
# mwahdan: Rename layers
except OSError:
raise OSError(
"Unable to load weights from h5 file. "
"If you tried to load a TF 2.0 model from a PyTorch checkpoint, please set from_pt=True. "
)
model(model.dummy_inputs, training=False) # Make sure restore ops are run
# mwahdan: Rename layers
if layer_pruning is not None:
model = rename_layers(model)
# Check if the models are the same to output loading informations
with h5py.File(resolved_archive_file, "r") as f:
if "layer_names" not in f.attrs and "model_weights" in f:
f = f["model_weights"]
hdf5_layer_names = set(
hdf5_format.load_attributes_from_hdf5_group(f, "layer_names"))
model_layer_names = set(layer.name for layer in model.layers)
missing_keys = list(model_layer_names - hdf5_layer_names)
unexpected_keys = list(hdf5_layer_names - model_layer_names)
error_msgs = []
if len(unexpected_keys) > 0:
logger.warning(
f"Some weights of the model checkpoint at {pretrained_model_name_or_path} were not used when "
f"initializing {model.__class__.__name__}: {unexpected_keys}\n"
f"- This IS expected if you are initializing {model.__class__.__name__} from the checkpoint of a model trained on another task "
f"or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n"
f"- This IS NOT expected if you are initializing {model.__class__.__name__} from the checkpoint of a model that you expect "
f"to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model)."
)
else:
logger.warning(
f"All model checkpoint weights were used when initializing {model.__class__.__name__}.\n"
)
if len(missing_keys) > 0:
logger.warning(
f"Some weights of {model.__class__.__name__} were not initialized from the model checkpoint at {pretrained_model_name_or_path} "
f"and are newly initialized: {missing_keys}\n"
f"You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference."
)
else:
logger.warning(
f"All the weights of {model.__class__.__name__} were initialized from the model checkpoint at {pretrained_model_name_or_path}.\n"
f"If your task is similar to the task the model of the ckeckpoint was trained on, "
f"you can already use {model.__class__.__name__} for predictions without further training."
)
if len(error_msgs) > 0:
raise RuntimeError("Error(s) in loading weights for {}:\n\t{}".format(
model.__class__.__name__, "\n\t".join(error_msgs)))
if output_loading_info:
loading_info = {
"missing_keys": missing_keys,
"unexpected_keys": unexpected_keys,
"error_msgs": error_msgs
}
return model, loading_info
return model
else:
output_dict = outputs
else:
output_dict = {}
for k, v in outputs.items():
if format_columns is not None and k not in format_columns and not output_all_columns:
continue
if format_columns is None or k in format_columns:
v = map_nested(command, v, **map_nested_kwargs)
output_dict[k] = v
return output_dict
ds = FixedDataset.from_file('../WRITTEN/dataset.arrow')
ds.set_format(type='tensorflow', columns=['input_ids'], shape=[2048])
mirrored_strategy = tf.distribute.MirroredStrategy(
devices=["/gpu:0", "/gpu:1"])
with mirrored_strategy.scope():
config_name = 'gpt2'
model = TFGPT2LMHeadModel.from_pretrained(config_name)
gpt2_weights_file_url = hf_bucket_url(config_name,
filename=TF2_WEIGHTS_NAME)
gpt2_weights_file = cached_path(gpt2_weights_file_url)
model.load_weights(gpt2_weights_file, by_name=True)
optimizer = tf.keras.optimizers.Adam(learning_rate=3e-5)
loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
model.compile(optimizer=optimizer, loss=loss)
model.fit(tf.data.Dataset.from_tensor_slices(ds['input_ids']),
epochs=2,
steps_per_epoch=115)
#! /usr/bin/python3 -i
# coding=utf-8
import os
PACKAGE_DIR=os.path.abspath(os.path.dirname(__file__))
DOWNLOAD_DIR=os.path.join(PACKAGE_DIR,"models")
from transformers.file_utils import hf_bucket_url
MODEL_URL=hf_bucket_url("KoichiYasuoka/SuPar-Kanbun","suparkanbun/models/")
import numpy
from spacy.language import Language
from spacy.symbols import LANG,NORM,LEMMA,POS,TAG,DEP,HEAD
from spacy.tokens import Doc,Span,Token
from spacy.util import get_lang_class
class SuParKanbunLanguage(Language):
lang="lzh"
max_length=10**6
def __init__(self,BERT,Danku):
self.Defaults.lex_attr_getters[LANG]=lambda _text:"lzh"
try:
self.vocab=self.Defaults.create_vocab()
self.pipeline=[]
except:
from spacy.vocab import create_vocab
self.vocab=create_vocab("lzh",self.Defaults)
self._components=[]
self._disabled=set()
self.tokenizer=SuParKanbunTokenizer(BERT,Danku,self.vocab)
self._meta={
def test_model_not_found(self):
# Invalid model file.
url = hf_bucket_url("bert-base", filename="pytorch_model.bin")
with self.assertRaisesRegex(RepositoryNotFoundError,
"404 Client Error"):
_ = get_from_cache(url)
