def load(model: str):
model_file = None
# news-english-forward
if model.lower() == 'ner':
base_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models/ner-conll03.pt'
model_file = cached_path(base_path)
if model.lower() == 'chunk':
base_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models/chunk-conll2000.pt'
model_file = cached_path(base_path)
if model_file is not None:
tagger: SequenceTaggerLSTM = torch.load(model_file, map_location={'cuda:0': 'cpu'})
tagger.eval()
if torch.cuda.is_available():
tagger = tagger.cuda()
return tagger
def __init__(
self,
base_path: Union[str, Path] = None,
tag_to_bioes: str = "ner",
in_memory: bool = True,
):
if type(base_path) == str:
base_path: Path = Path(base_path)
# column format
columns = {0: "text", 1: "ner"}
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
wnut_path = "https://noisy-text.github.io/2017/files/"
cached_path(f"{wnut_path}wnut17train.conll",
Path("datasets") / dataset_name)
cached_path(f"{wnut_path}emerging.dev.conll",
Path("datasets") / dataset_name)
cached_path(f"{wnut_path}emerging.test.annotated",
Path("datasets") / dataset_name)
super(WNUT_17, self).__init__(data_folder,
columns,
tag_to_bioes=tag_to_bioes,
in_memory=in_memory)
def __init__(self,
base_path: Union[str, Path] = None,
in_memory: bool = True):
if type(base_path) == str:
base_path: Path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
ud_path = "https://raw.githubusercontent.com/UniversalDependencies/UD_Bulgarian-BTB/master"
cached_path(f"{ud_path}/bg_btb-ud-dev.conllu",
Path("datasets") / dataset_name)
cached_path(f"{ud_path}/bg_btb-ud-test.conllu",
Path("datasets") / dataset_name)
cached_path(f"{ud_path}/bg_btb-ud-train.conllu",
Path("datasets") / dataset_name)
super(UD_BULGARIAN, self).__init__(data_folder, in_memory=in_memory)
def __init__(self,
base_path: Union[str, Path] = None,
in_memory: bool = True):
if type(base_path) == str:
base_path: Path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
web_path = "https://raw.githubusercontent.com/UniversalDependencies/UD_Classical_Chinese-Kyoto/master"
cached_path(f"{web_path}/lzh_kyoto-ud-dev.conllu",
Path("datasets") / dataset_name)
cached_path(f"{web_path}/lzh_kyoto-ud-test.conllu",
Path("datasets") / dataset_name)
cached_path(f"{web_path}/lzh_kyoto-ud-train.conllu",
Path("datasets") / dataset_name)
super(UD_CHINESE_KYOTO, self).__init__(data_folder,
in_memory=in_memory)
def __init__(
self,
base_path: Union[str, Path] = None,
tag_to_bioes: str = "ner",
in_memory: bool = True,
):
if type(base_path) == str:
base_path: Path = Path(base_path)
# column format
columns = {1: 'text', 3: 'pos', 9: 'ner'}
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
data_path = Path(flair.cache_root) / "datasets" / dataset_name
train_data_file = data_path / "ddt.train.conllu"
if not train_data_file.is_file():
temp_file = cached_path(
'https://danlp.s3.eu-central-1.amazonaws.com/datasets/ddt.zip',
Path("datasets") / dataset_name
)
from zipfile import ZipFile
with ZipFile(temp_file, 'r') as zip_file:
zip_file.extractall(path=data_path)
# Remove CoNLL-U meta information in the last column
for part in ['train', 'dev', 'test']:
lines = []
data_file = "ddt.{}.conllu".format(part)
with open(data_path / data_file, 'r') as file:
for line in file:
if line.startswith("#") or line == "\n":
lines.append(line)
lines.append(line.replace("name=", "").replace("|SpaceAfter=No", ""))
with open(data_path / data_file, 'w') as file:
file.writelines(lines)
print(data_path / data_file)
super(DANE, self).__init__(
data_folder, columns, tag_to_bioes=tag_to_bioes,
in_memory=in_memory, comment_symbol="#"
)
def __init__(
self,
base_path: Union[str, Path] = None,
in_memory: bool = True,
sentence_splitter: SentenceSplitter = SegtokSentenceSplitter(),
):
if type(base_path) == str:
base_path: Path = Path(base_path)
self.sentence_splitter = sentence_splitter
# this dataset name
dataset_name = self.__class__.__name__.lower() + "_" + type(
self.sentence_splitter).__name__
# default dataset folder is the cache root
if not base_path:
base_path = flair.cache_root / "datasets"
data_folder = base_path / dataset_name
drugprot_url = (
"https://zenodo.org/record/5042151/files/drugprot-gs-training-development.zip"
)
data_file = data_folder / "drugprot-train.conllu"
if not data_file.is_file():
source_data_folder = data_folder / "original"
cached_path(drugprot_url, source_data_folder)
self.extract_and_convert_to_conllu(
data_file=source_data_folder /
"drugprot-gs-training-development.zip",
data_folder=data_folder,
)
super(DrugProt, self).__init__(
data_folder,
in_memory=in_memory,
sample_missing_splits=False,
)
def load(model: str):
model_file = None
if model.lower() == 'ner':
base_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models/ner-conll03.pt'
model_file = cached_path(base_path)
if model.lower() == 'chunk':
base_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models/chunk-conll2000.pt'
model_file = cached_path(base_path)
if model.lower() == 'pos':
base_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models/pos-ontonotes-small.pt'
model_file = cached_path(base_path)
if model_file is not None:
tagger: SequenceTaggerLSTM = torch.load(
model_file, map_location={'cuda:0': 'cpu'})
tagger.eval()
if torch.cuda.is_available():
tagger = tagger.cuda()
return tagger
def load(model: str):
model_file = None
aws_resource_path = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
)
cache_dir = Path("models")
if model.lower() == "de-offensive-language":
base_path = "/".join([
aws_resource_path,
"TEXT-CLASSIFICATION_germ-eval-2018_task-1",
"germ-eval-2018-task-1.pt",
])
model_file = cached_path(base_path, cache_dir=cache_dir)
elif model.lower() == "en-sentiment":
base_path = "/".join(
[aws_resource_path, "TEXT-CLASSIFICATION_imdb", "imdb.pt"])
model_file = cached_path(base_path, cache_dir=cache_dir)
if model_file is not None:
return TextClassifier.load_from_file(model_file)
def __init__(
self,
base_path: Union[str, Path] = None,
tag_to_bioes: str = "ner",
in_memory: bool = False,
):
"""
Initialize the LER_GERMAN (Legal Entity Recognition) corpus. The first time you call this constructor it will automatically
download the dataset.
:param base_path: Default is None, meaning that corpus gets auto-downloaded and loaded. You can override this
to point to a different folder but typically this should not be necessary.
:param in_memory: If True, keeps dataset in memory giving speedups in training. Not recommended due to heavy RAM usage.
:param document_as_sequence: If True, all sentences of a document are read into a single Sentence object
"""
if type(base_path) == str:
base_path: Path = Path(base_path)
# column format
columns = {0: "text", 1: "ner"}
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
ler_path = "https://raw.githubusercontent.com/elenanereiss/Legal-Entity-Recognition/master/data/"
cached_path(f"{ler_path}ler.conll", Path("datasets") / dataset_name)
super(LER_GERMAN, self).__init__(data_folder,
columns,
tag_to_bioes=tag_to_bioes,
in_memory=in_memory,
train_file='ler.conll')
def _fetch_model(model_name) -> str:
model_map = {}
hu_path: str = "https://nlp.informatik.hu-berlin.de/resources/models"
model_map["relations-fast"] = "/".join([hu_path, "relations-fast", "relations-fast.pt"])
model_map["relations"] = "/".join([hu_path, "relations", "relations.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name], cache_dir=cache_dir)
return model_name
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4'
model_map['de-offensive-language'] = '/'.join([
aws_resource_path, 'TEXT-CLASSIFICATION_germ-eval-2018_task-1',
'germ-eval-2018-task-1.pt'
])
model_map['en-sentiment'] = '/'.join(
[aws_resource_path, 'TEXT-CLASSIFICATION_imdb', 'imdb.pt'])
cache_dir = Path('models')
if (model_name in model_map):
model_name = cached_path(model_map[model_name],
cache_dir=cache_dir)
return model_name
def __init__(self, base_path: Union[str, Path] = None, in_memory: bool = True):
if type(base_path) == str:
base_path: Path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
web_path = "https://raw.githubusercontent.com/UniversalDependencies/UD_North_Sami-Giella/master"
cached_path(
f"{web_path}/sme_giella-ud-test.conllu", Path("datasets") / dataset_name
)
cached_path(
f"{web_path}/sme_giella-ud-train.conllu", Path("datasets") / dataset_name
)
super(UD_NORTH_SAMI, self).__init__(data_folder, in_memory=in_memory)
def _fetch_model(model_name) -> str:
model_map = {}
hu_path: str = "https://nlp.informatik.hu-berlin.de/resources/models"
model_map["tars-base"] = "/".join(
[hu_path, "tars-base", "tars-base-v8.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name],
cache_dir=cache_dir)
return model_name
def __init__(
self,
embeddings: str,
use_local: bool = True,
use_gensim: bool = False,
field: str = None,
):
"""
Initializes fasttext word embeddings. Constructor downloads required embedding file and stores in cache
if use_local is False.
:param embeddings: path to your embeddings '.bin' file
:param use_local: set this to False if you are using embeddings from a remote source
:param use_gensim: set this to true if your fasttext embedding is trained with fasttext version below 0.9.1
"""
cache_dir = Path("embeddings")
if use_local:
if not Path(embeddings).exists():
raise ValueError(
f'The given embeddings "{embeddings}" is not available or is not a valid path.'
)
else:
embeddings = cached_path(f"{embeddings}", cache_dir=cache_dir)
self.embeddings = embeddings
self.name: str = str(embeddings)
self.static_embeddings = True
self.use_gensim = use_gensim
if use_gensim:
self.precomputed_word_embeddings = gensim.models.FastText.load_fasttext_format(
str(embeddings))
self.__embedding_length: int = self.precomputed_word_embeddings.vector_size
else:
self.precomputed_word_embeddings = ft.load_model(str(embeddings))
self.__embedding_length: int = self.precomputed_word_embeddings.get_dimension(
)
self.field = field
super().__init__()
def __init__(
self,
base_path: Union[str, Path] = None,
tag_to_bioes: str = "ner",
in_memory: bool = True,
document_as_sequence: bool = False,
):
"""
Initialize the CoNLL-03 corpus for Dutch. The first time you call this constructor it will automatically
download the dataset.
:param base_path: Default is None, meaning that corpus gets auto-downloaded and loaded. You can override this
to point to a different folder but typically this should not be necessary.
:param tag_to_bioes: NER by default, need not be changed, but you could also select 'pos' to predict
POS tags instead
:param in_memory: If True, keeps dataset in memory giving speedups in training.
:param document_as_sequence: If True, all sentences of a document are read into a single Sentence object
"""
if type(base_path) == str:
base_path: Path = Path(base_path)
# column format
columns = {0: "text", 1: "pos", 2: "ner"}
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
conll_02_path = "https://www.clips.uantwerpen.be/conll2002/ner/data/"
cached_path(f"{conll_02_path}ned.testa",
Path("datasets") / dataset_name)
cached_path(f"{conll_02_path}ned.testb",
Path("datasets") / dataset_name)
cached_path(f"{conll_02_path}ned.train",
Path("datasets") / dataset_name)
super(CONLL_03_DUTCH, self).__init__(
data_folder,
columns,
tag_to_bioes=tag_to_bioes,
encoding="latin-1",
in_memory=in_memory,
document_separator_token=None
if not document_as_sequence else "-DOCSTART-",
)
def __init__(self,
base_path: Union[str, Path] = None,
in_memory: bool = False):
if type(base_path) == str:
base_path: Path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
ud_path = (
"https://raw.githubusercontent.com/UniversalDependencies/UD_German-HDT/dev"
)
cached_path(f"{ud_path}/de_hdt-ud-dev.conllu",
Path("datasets") / dataset_name)
cached_path(f"{ud_path}/de_hdt-ud-test.conllu",
Path("datasets") / dataset_name)
train_filenames = [
"de_hdt-ud-train-a-1.conllu",
"de_hdt-ud-train-a-2.conllu",
"de_hdt-ud-train-b-1.conllu",
"de_hdt-ud-train-b-2.conllu",
]
for train_file in train_filenames:
cached_path(f"{ud_path}/{train_file}",
Path("datasets") / dataset_name / "original")
data_path = Path(flair.cache_root) / "datasets" / dataset_name
new_train_file: Path = data_path / "de_hdt-ud-train-all.conllu"
if not new_train_file.is_file():
with open(new_train_file, "wt") as f_out:
for train_filename in train_filenames:
with open(data_path / "original" / train_filename,
"rt") as f_in:
f_out.write(f_in.read())
super(UD_GERMAN_HDT, self).__init__(data_folder, in_memory=in_memory)
def evaluate(test_file, model_file, dataset_format='macss', semeval_scoring=False):
if semeval_scoring:
eval_script = cached_path(
'https://raw.githubusercontent.com/vzhong/semeval/master/dataset/SemEval2010_task8_scorer-v1.2/semeval2010_task8_scorer-v1.2.pl',
cache_dir='scripts')
chmod(eval_script, 0o777)
classifier: TextClassifier = TextClassifier.load_from_file(model_file)
#sentences_test: List[Sentence] = load_sentences_jsonl(test_file, attach_id=True)
idx2item = load_idx2item(join(dirname(test_file), 'vocabulary/embeddings.csv'))
load_dataset = dataset_loader[dataset_format]
sentences_test: List[Sentence] = load_dataset(test_file, idx2item, is_test=False, attach_id=True)
sentences_pred: List[Sentence] = load_dataset(test_file, idx2item, is_test=True, attach_id=True)
sentences_pred = classifier.predict(sentences_pred)
if semeval_scoring:
id_labels_true = [(sentence.id_, sentence.labels[0]) for sentence in sentences_test]
id_labels_pred = [(sentence.id_, sentence.labels[0]) for sentence in sentences_pred]
input_files = []
for id_labels in [id_labels_true, id_labels_pred]:
tmp_file = NamedTemporaryFile(delete=True)
input_files.append(tmp_file)
with open(tmp_file.name, 'w') as f:
for id_, label in id_labels:
f.write('{}\t{}\n'.format(id_, label.name))
tmp_file.file.close()
p = run([eval_script, input_files[0].name, input_files[1].name], stdout=PIPE, encoding='utf-8')
main_result = p.stdout
print(main_result)
else:
y_true = [sentence.labels[0].name for sentence in sentences_test]
y_pred = [sentence.labels[0].name for sentence in sentences_pred]
print(classification_report(y_true, y_pred))
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
)
model_map["de-offensive-language"] = "/".join([
aws_resource_path,
"classy-offensive-de-rnn-cuda%3A0",
"germ-eval-2018-task-1-v0.4.pt",
])
model_map["en-sentiment"] = "/".join(
[aws_resource_path, "classy-imdb-en-rnn-cuda%3A0", "imdb-v0.4.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name],
cache_dir=cache_dir)
return model_name
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
)
model_map["de-offensive-language"] = "/".join([
aws_resource_path,
"TEXT-CLASSIFICATION_germ-eval-2018_task-1",
"germ-eval-2018-task-1.pt",
])
model_map["en-sentiment"] = "/".join(
[aws_resource_path, "TEXT-CLASSIFICATION_imdb", "imdb.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name],
cache_dir=cache_dir)
return model_name
def __init__(self, **kwargs):
dataset = "feidegger"
# cache Feidegger config file
json_link = "https://raw.githubusercontent.com/zalandoresearch/feidegger/master/data/FEIDEGGER_release_1.1.json"
json_local_path = cached_path(json_link, Path("datasets") / dataset)
# cache Feidegger images
dataset_info = json.load(open(json_local_path, "r"))
images_cache_folder = os.path.join(os.path.dirname(json_local_path),
"images")
if not os.path.isdir(images_cache_folder):
os.mkdir(images_cache_folder)
for image_info in tqdm(dataset_info):
name = os.path.basename(image_info["url"])
filename = os.path.join(images_cache_folder, name)
if not os.path.isfile(filename):
urllib.request.urlretrieve(image_info["url"], filename)
# replace image URL with local cached file
image_info["url"] = filename
feidegger_dataset: Dataset = FeideggerDataset(dataset_info, **kwargs)
train_indices = list(
np.where(np.in1d(feidegger_dataset.split, list(range(8))))[0])
train = torch.utils.data.dataset.Subset(feidegger_dataset,
train_indices)
dev_indices = list(np.where(np.in1d(feidegger_dataset.split, [8]))[0])
dev = torch.utils.data.dataset.Subset(feidegger_dataset, dev_indices)
test_indices = list(np.where(np.in1d(feidegger_dataset.split, [9]))[0])
test = torch.utils.data.dataset.Subset(feidegger_dataset, test_indices)
super(FeideggerCorpus, self).__init__(train,
dev,
test,
name="feidegger")
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path = "https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
hu_path: str = "https://nlp.informatik.hu-berlin.de/resources/models"
model_map["de-offensive-language"] = "/".join([
aws_resource_path,
"classy-offensive-de-rnn-cuda%3A0",
"germ-eval-2018-task-1-v0.4.pt",
])
# English sentiment models
model_map["sentiment"] = "/".join([
hu_path, "sentiment-curated-distilbert",
"sentiment-en-mix-distillbert.pt"
])
model_map["en-sentiment"] = "/".join([
hu_path, "sentiment-curated-distilbert",
"sentiment-en-mix-distillbert.pt"
])
model_map["sentiment-fast"] = "/".join([
hu_path, "sentiment-curated-fasttext-rnn",
"sentiment-en-mix-ft-rnn.pt"
])
#Communicative Functions Model
model_map["communicative-functions"] = "/".join(
[hu_path, "comfunc", "communicative-functions-v0.5b.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name],
cache_dir=cache_dir)
return model_name
def _fetch_model(model_name) -> str:
model_map = {}
hu_path: str = "https://nlp.informatik.hu-berlin.de/resources/models"
model_map["de-offensive-language"] = "/".join(
[hu_path, "de-offensive-language", "germ-eval-2018-task-1-v0.8.pt"]
)
# English sentiment models
model_map["sentiment"] = "/".join(
[
hu_path,
"sentiment-curated-distilbert",
"sentiment-en-mix-distillbert_4.pt",
]
)
model_map["en-sentiment"] = "/".join(
[
hu_path,
"sentiment-curated-distilbert",
"sentiment-en-mix-distillbert_4.pt",
]
)
model_map["sentiment-fast"] = "/".join(
[hu_path, "sentiment-curated-fasttext-rnn", "sentiment-en-mix-ft-rnn_v8.pt"]
)
# Communicative Functions Model
model_map["communicative-functions"] = "/".join([hu_path, "comfunc", "communicative-functions.pt"])
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name], cache_dir=cache_dir)
return model_name
def download_dataset(task: NLPTask):
# conll 2000 chunking task
if task == NLPTask.CONLL_2000:
conll_2000_path = 'https://www.clips.uantwerpen.be/conll2000/chunking/'
data_file = Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'train.txt'
if not data_file.is_file():
cached_path(f'{conll_2000_path}train.txt.gz', Path('datasets') / task.value)
cached_path(f'{conll_2000_path}test.txt.gz', Path('datasets') / task.value)
import gzip, shutil
with gzip.open(Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'train.txt.gz',
'rb') as f_in:
with open(Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'train.txt', 'wb') as f_out:
shutil.copyfileobj(f_in, f_out)
with gzip.open(Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'test.txt.gz',
'rb') as f_in:
with open(Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'test.txt', 'wb') as f_out:
shutil.copyfileobj(f_in, f_out)
if task == NLPTask.IMDB:
imdb_acl_path = 'http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz'
data_path = Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value
data_file = data_path / 'train.txt'
if not data_file.is_file():
cached_path(imdb_acl_path, Path('datasets') / task.value)
import tarfile
with tarfile.open(Path(flair.file_utils.CACHE_ROOT) / 'datasets' / task.value / 'aclImdb_v1.tar.gz',
'r:gz') as f_in:
datasets = ['train', 'test']
labels = ['pos', 'neg']
for label in labels:
for dataset in datasets:
f_in.extractall(data_path, members=[m for m in f_in.getmembers()
if f'{dataset}/{label}' in m.name])
with open(f'{data_path}/{dataset}.txt', 'at') as f_p:
current_path = data_path / 'aclImdb' / dataset / label
for file_name in current_path.iterdir():
if file_name.is_file() and file_name.name.endswith('.txt'):
f_p.write(f'__label__{label} '
+ file_name.open('rt', encoding='utf-8').read() + '\n')
def __init__(self,
base_path: Union[str, Path] = None,
in_memory: bool = True):
if not base_path:
base_path = flair.cache_root / "datasets"
else:
base_path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
data_folder = base_path / dataset_name
# TODO: change data source to original CoNLL04 -- this dataset has span formatting errors
# download data if necessary
conll04_url = (
"https://raw.githubusercontent.com/bekou/multihead_joint_entity_relation_extraction/master/data/CoNLL04/"
)
data_file = data_folder / "conll04-train.conllu"
if True or not data_file.is_file():
source_data_folder = data_folder / "original"
cached_path(f"{conll04_url}train.txt", source_data_folder)
cached_path(f"{conll04_url}dev.txt", source_data_folder)
cached_path(f"{conll04_url}test.txt", source_data_folder)
self.convert_to_conllu(
source_data_folder=source_data_folder,
data_folder=data_folder,
)
super(RE_ENGLISH_CONLL04, self).__init__(
data_folder,
in_memory=in_memory,
column_format={
1: "text",
2: "ner"
},
comment_symbol="# ",
)
def __init__(
self,
**corpusargs,
):
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
data_folder = Path(flair.cache_root) / "datasets" / dataset_name
# download data if necessary
if not (data_folder / "train.txt").is_file():
# download senteval datasets if necessary und unzip
cached_path(
'https://raw.githubusercontent.com/AcademiaSinicaNLPLab/sentiment_dataset/master/data/stsa.fine.train',
Path("datasets") / dataset_name / 'raw')
cached_path(
'https://raw.githubusercontent.com/AcademiaSinicaNLPLab/sentiment_dataset/master/data/stsa.fine.test',
Path("datasets") / dataset_name / 'raw')
cached_path(
'https://raw.githubusercontent.com/AcademiaSinicaNLPLab/sentiment_dataset/master/data/stsa.fine.dev',
Path("datasets") / dataset_name / 'raw')
# convert to FastText format
for split in ['train', 'dev', 'test']:
with open(data_folder / f"{split}.txt", "w") as train_file:
with open(data_folder / 'raw' / f'stsa.fine.{split}',
encoding="latin1") as file:
for line in file:
train_file.write(f"__label__{line[0]} {line[2:]}")
super(SENTEVAL_SST_GRANULAR, self).__init__(
data_folder,
tokenizer=segtok_tokenizer,
**corpusargs,
)
def __init__(self,
base_path: Union[str, Path] = None,
in_memory: bool = True):
if type(base_path) == str:
base_path: Path = Path(base_path)
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
web_path = "https://raw.githubusercontent.com/UniversalDependencies/UD_Maltese-MUDT/master"
cached_path(f"{web_path}/mt_mudt-ud-dev.conllu",
Path("datasets") / dataset_name)
cached_path(f"{web_path}/mt_mudt-ud-test.conllu",
Path("datasets") / dataset_name)
cached_path(f"{web_path}/mt_mudt-ud-train.conllu",
Path("datasets") / dataset_name)
super(UD_MALTESE, self).__init__(data_folder, in_memory=in_memory)
def __init__(
self,
base_path: Union[str, Path] = None,
tag_to_bioes: str = "ner",
in_memory: bool = True,
):
if type(base_path) == str:
base_path: Path = Path(base_path)
# column format
columns = {0: "text", 1: "ner"}
# this dataset name
dataset_name = self.__class__.__name__.lower()
# default dataset folder is the cache root
if not base_path:
base_path = Path(flair.cache_root) / "datasets"
data_folder = base_path / dataset_name
# download data if necessary
ner_finnish_path = "https://raw.githubusercontent.com/mpsilfve/finer-data/master/data/digitoday."
cached_path(f"{ner_finnish_path}2014.train.csv",
Path("datasets") / dataset_name)
cached_path(f"{ner_finnish_path}2014.dev.csv",
Path("datasets") / dataset_name)
cached_path(f"{ner_finnish_path}2015.test.csv",
Path("datasets") / dataset_name)
_remove_lines_without_annotations(
data_file=Path(data_folder / "digitoday.2015.test.csv"))
super(NER_FINNISH, self).__init__(data_folder,
columns,
tag_to_bioes=tag_to_bioes,
in_memory=in_memory,
skip_first_line=True)
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path_v04 = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
)
model_map["ner"] = "/".join(
[aws_resource_path_v04, "NER-conll03-english", "en-ner-conll03-v0.4.pt"]
)
model_map["ner-fast"] = "/".join(
[
aws_resource_path_v04,
"NER-conll03--h256-l1-b32-p3-0.5-%2Bglove%2Bnews-forward-fast%2Bnews-backward-fast-normal-locked0.5-word0.05--release_4",
"en-ner-fast-conll03-v0.4.pt",
]
)
model_map["ner-ontonotes"] = "/".join(
[
aws_resource_path_v04,
"release-ner-ontonotes-0",
"en-ner-ontonotes-v0.4.pt",
]
)
model_map["ner-ontonotes-fast"] = "/".join(
[
aws_resource_path_v04,
"release-ner-ontonotes-fast-0",
"en-ner-ontonotes-fast-v0.4.pt",
]
)
for key in ["ner-multi", "multi-ner"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"release-quadner-512-l2-multi-embed",
"quadner-large.pt",
]
)
for key in ["ner-multi-fast", "multi-ner-fast"]:
model_map[key] = "/".join(
[aws_resource_path_v04, "NER-multi-fast", "ner-multi-fast.pt"]
)
for key in ["ner-multi-fast-learn", "multi-ner-fast-learn"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"NER-multi-fast-evolve",
"ner-multi-fast-learn.pt",
]
)
model_map["pos"] = "/".join(
[
aws_resource_path_v04,
"POS-ontonotes--h256-l1-b32-p3-0.5-%2Bglove%2Bnews-forward%2Bnews-backward-normal-locked0.5-word0.05--v0.4_0",
"en-pos-ontonotes-v0.4.pt",
]
)
model_map["pos-fast"] = "/".join(
[
aws_resource_path_v04,
"release-pos-fast-0",
"en-pos-ontonotes-fast-v0.4.pt",
]
)
for key in ["pos-multi", "multi-pos"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"release-dodekapos-512-l2-multi",
"pos-multi-v0.1.pt",
]
)
for key in ["pos-multi-fast", "multi-pos-fast"]:
model_map[key] = "/".join(
[aws_resource_path_v04, "UPOS-multi-fast", "pos-multi-fast.pt"]
)
model_map["frame"] = "/".join(
[aws_resource_path_v04, "release-frame-1", "en-frame-ontonotes-v0.4.pt"]
)
model_map["frame-fast"] = "/".join(
[
aws_resource_path_v04,
"release-frame-fast-0",
"en-frame-ontonotes-fast-v0.4.pt",
]
)
model_map["chunk"] = "/".join(
[
aws_resource_path_v04,
"NP-conll2000--h256-l1-b32-p3-0.5-%2Bnews-forward%2Bnews-backward-normal-locked0.5-word0.05--v0.4_0",
"en-chunk-conll2000-v0.4.pt",
]
)
model_map["chunk-fast"] = "/".join(
[
aws_resource_path_v04,
"release-chunk-fast-0",
"en-chunk-conll2000-fast-v0.4.pt",
]
)
model_map["da-pos"] = "/".join(
[aws_resource_path_v04, "POS-danish", "da-pos-v0.1.pt"]
)
model_map["da-ner"] = "/".join(
[aws_resource_path_v04, "NER-danish", "da-ner-v0.1.pt"]
)
model_map["de-pos"] = "/".join(
[aws_resource_path_v04, "release-de-pos-0", "de-pos-ud-hdt-v0.4.pt"]
)
model_map["de-pos-fine-grained"] = "/".join(
[
aws_resource_path_v04,
"POS-fine-grained-german-tweets",
"de-pos-twitter-v0.1.pt",
]
)
model_map["de-ner"] = "/".join(
[aws_resource_path_v04, "release-de-ner-0", "de-ner-conll03-v0.4.pt"]
)
model_map["de-ner-germeval"] = "/".join(
[aws_resource_path_v04, "NER-germeval", "de-ner-germeval-0.4.1.pt"]
)
model_map["fr-ner"] = "/".join(
[aws_resource_path_v04, "release-fr-ner-0", "fr-ner-wikiner-0.4.pt"]
)
model_map["nl-ner"] = "/".join(
[aws_resource_path_v04, "NER-conll2002-dutch", "nl-ner-conll02-v0.1.pt"]
)
model_map["ml-pos"] = "https://raw.githubusercontent.com/qburst/models-repository/master/FlairMalayalamModels/malayalam-upos-model.pt"
model_map["ml-xpos"] = "https://raw.githubusercontent.com/qburst/models-repository/master/FlairMalayalamModels/malayalam-xpos-model.pt"
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name], cache_dir=cache_dir)
return model_name
def _fetch_model(model_name) -> str:
model_map = {}
aws_resource_path = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.2"
)
aws_resource_path_v04 = (
"https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.4"
)
model_map["ner"] = "/".join(
[aws_resource_path_v04, "NER-conll03-english", "en-ner-conll03-v0.4.pt"]
)
model_map["ner-fast"] = "/".join(
[
aws_resource_path,
"NER-conll03--h256-l1-b32-experimental--fast-v0.2",
"en-ner-fast-conll03-v0.2.pt",
]
)
model_map["ner-ontonotes"] = "/".join(
[
aws_resource_path,
"NER-ontoner--h256-l1-b32-%2Bcrawl%2Bnews-forward%2Bnews-backward--v0.2",
"en-ner-ontonotes-v0.3.pt",
]
)
model_map["ner-ontonotes-fast"] = "/".join(
[
aws_resource_path,
"NER-ontoner--h256-l1-b32-%2Bcrawl%2Bnews-forward-fast%2Bnews-backward-fast--v0.2",
"en-ner-ontonotes-fast-v0.3.pt",
]
)
for key in ["ner-multi", "multi-ner"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"release-quadner-512-l2-multi-embed",
"quadner-large.pt",
]
)
for key in ["ner-multi-fast", "multi-ner-fast"]:
model_map[key] = "/".join(
[aws_resource_path_v04, "NER-multi-fast", "ner-multi-fast.pt"]
)
for key in ["ner-multi-fast-learn", "multi-ner-fast-learn"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"NER-multi-fast-evolve",
"ner-multi-fast-learn.pt",
]
)
model_map["pos"] = "/".join(
[
aws_resource_path,
"POS-ontonotes--h256-l1-b32-%2Bmix-forward%2Bmix-backward--v0.2",
"en-pos-ontonotes-v0.2.pt",
]
)
model_map["pos-fast"] = "/".join(
[
aws_resource_path,
"POS-ontonotes--h256-l1-b32-%2Bnews-forward-fast%2Bnews-backward-fast--v0.2",
"en-pos-ontonotes-fast-v0.2.pt",
]
)
for key in ["pos-multi", "multi-pos"]:
model_map[key] = "/".join(
[
aws_resource_path_v04,
"release-dodekapos-512-l2-multi",
"pos-multi-v0.1.pt",
]
)
for key in ["pos-multi-fast", "multi-pos-fast"]:
model_map[key] = "/".join(
[aws_resource_path_v04, "UPOS-multi-fast", "pos-multi-fast.pt"]
)
model_map["frame"] = "/".join(
[
aws_resource_path,
"FRAME-conll12--h256-l1-b8-%2Bnews%2Bnews-forward%2Bnews-backward--v0.2",
"en-frame-ontonotes-v0.2.pt",
]
)
model_map["frame-fast"] = "/".join(
[
aws_resource_path,
"FRAME-conll12--h256-l1-b8-%2Bnews%2Bnews-forward-fast%2Bnews-backward-fast--v0.2",
"en-frame-ontonotes-fast-v0.2.pt",
]
)
model_map["chunk"] = "/".join(
[
aws_resource_path,
"NP-conll2000--h256-l1-b32-%2Bnews-forward%2Bnews-backward--v0.2",
"en-chunk-conll2000-v0.2.pt",
]
)
model_map["chunk-fast"] = "/".join(
[
aws_resource_path,
"NP-conll2000--h256-l1-b32-%2Bnews-forward-fast%2Bnews-backward-fast--v0.2",
"en-chunk-conll2000-fast-v0.2.pt",
]
)
model_map["de-pos"] = "/".join(
[
aws_resource_path,
"UPOS-udgerman--h256-l1-b8-%2Bgerman-forward%2Bgerman-backward--v0.2",
"de-pos-ud-v0.2.pt",
]
)
model_map["de-pos-fine-grained"] = "/".join(
[
aws_resource_path_v04,
"POS-fine-grained-german-tweets",
"de-pos-twitter-v0.1.pt",
]
)
model_map["de-ner"] = "/".join(
[
aws_resource_path,
"NER-conll03ger--h256-l1-b32-%2Bde-fasttext%2Bgerman-forward%2Bgerman-backward--v0.2",
"de-ner-conll03-v0.3.pt",
]
)
model_map["de-ner-germeval"] = "/".join(
[
aws_resource_path,
"NER-germeval--h256-l1-b32-%2Bde-fasttext%2Bgerman-forward%2Bgerman-backward--v0.2",
"de-ner-germeval-v0.3.pt",
]
)
model_map["fr-ner"] = "/".join(
[aws_resource_path, "NER-aij-wikiner-fr-wp3", "fr-ner.pt"]
)
model_map["nl-ner"] = "/".join(
[aws_resource_path_v04, "NER-conll2002-dutch", "nl-ner-conll02-v0.1.pt"]
)
cache_dir = Path("models")
if model_name in model_map:
model_name = cached_path(model_map[model_name], cache_dir=cache_dir)
return model_name
def _fetch_model(model_name) -> str:
# core Flair models on Huggingface ModelHub
huggingface_model_map = {
"ner": "flair/ner-english",
"ner-fast": "flair/ner-english-fast",
"ner-ontonotes": "flair/ner-english-ontonotes",
"ner-ontonotes-fast": "flair/ner-english-ontonotes-fast",
# Large NER models,
"ner-large": "flair/ner-english-large",
"ner-ontonotes-large": "flair/ner-english-ontonotes-large",
"de-ner-large": "flair/ner-german-large",
"nl-ner-large": "flair/ner-dutch-large",
"es-ner-large": "flair/ner-spanish-large",
# Multilingual NER models
"ner-multi": "flair/ner-multi",
"multi-ner": "flair/ner-multi",
"ner-multi-fast": "flair/ner-multi-fast",
# English POS models
"upos": "flair/upos-english",
"upos-fast": "flair/upos-english-fast",
"pos": "flair/pos-english",
"pos-fast": "flair/pos-english-fast",
# Multilingual POS models
"pos-multi": "flair/upos-multi",
"multi-pos": "flair/upos-multi",
"pos-multi-fast": "flair/upos-multi-fast",
"multi-pos-fast": "flair/upos-multi-fast",
# English SRL models
"frame": "flair/frame-english",
"frame-fast": "flair/frame-english-fast",
# English chunking models
"chunk": "flair/chunk-english",
"chunk-fast": "flair/chunk-english-fast",
# Language-specific NER models
"da-ner": "flair/ner-danish",
"de-ner": "flair/ner-german",
"de-ler": "flair/ner-german-legal",
"de-ner-legal": "flair/ner-german-legal",
"fr-ner": "flair/ner-french",
"nl-ner": "flair/ner-dutch",
}
hu_path: str = "https://nlp.informatik.hu-berlin.de/resources/models"
hu_model_map = {
# English NER models
"ner":
"/".join([hu_path, "ner", "en-ner-conll03-v0.4.pt"]),
"ner-pooled":
"/".join([hu_path, "ner-pooled", "en-ner-conll03-pooled-v0.5.pt"]),
"ner-fast":
"/".join([hu_path, "ner-fast", "en-ner-fast-conll03-v0.4.pt"]),
"ner-ontonotes":
"/".join([hu_path, "ner-ontonotes", "en-ner-ontonotes-v0.4.pt"]),
"ner-ontonotes-fast":
"/".join([
hu_path, "ner-ontonotes-fast", "en-ner-ontonotes-fast-v0.4.pt"
]),
# Multilingual NER models
"ner-multi":
"/".join([hu_path, "multi-ner", "quadner-large.pt"]),
"multi-ner":
"/".join([hu_path, "multi-ner", "quadner-large.pt"]),
"ner-multi-fast":
"/".join([hu_path, "multi-ner-fast", "ner-multi-fast.pt"]),
# English POS models
"upos":
"/".join([hu_path, "upos", "en-pos-ontonotes-v0.4.pt"]),
"upos-fast":
"/".join([hu_path, "upos-fast", "en-upos-ontonotes-fast-v0.4.pt"]),
"pos":
"/".join([hu_path, "pos", "en-pos-ontonotes-v0.5.pt"]),
"pos-fast":
"/".join([hu_path, "pos-fast", "en-pos-ontonotes-fast-v0.5.pt"]),
# Multilingual POS models
"pos-multi":
"/".join([hu_path, "multi-pos", "pos-multi-v0.1.pt"]),
"multi-pos":
"/".join([hu_path, "multi-pos", "pos-multi-v0.1.pt"]),
"pos-multi-fast":
"/".join([hu_path, "multi-pos-fast", "pos-multi-fast.pt"]),
"multi-pos-fast":
"/".join([hu_path, "multi-pos-fast", "pos-multi-fast.pt"]),
# English SRL models
"frame":
"/".join([hu_path, "frame", "en-frame-ontonotes-v0.4.pt"]),
"frame-fast":
"/".join(
[hu_path, "frame-fast", "en-frame-ontonotes-fast-v0.4.pt"]),
# English chunking models
"chunk":
"/".join([hu_path, "chunk", "en-chunk-conll2000-v0.4.pt"]),
"chunk-fast":
"/".join(
[hu_path, "chunk-fast", "en-chunk-conll2000-fast-v0.4.pt"]),
# Danish models
"da-pos":
"/".join([hu_path, "da-pos", "da-pos-v0.1.pt"]),
"da-ner":
"/".join([hu_path, "NER-danish", "da-ner-v0.1.pt"]),
# German models
"de-pos":
"/".join([hu_path, "de-pos", "de-pos-ud-hdt-v0.5.pt"]),
"de-pos-tweets":
"/".join([hu_path, "de-pos-tweets", "de-pos-twitter-v0.1.pt"]),
"de-ner":
"/".join([hu_path, "de-ner", "de-ner-conll03-v0.4.pt"]),
"de-ner-germeval":
"/".join([hu_path, "de-ner-germeval", "de-ner-germeval-0.4.1.pt"]),
"de-ler":
"/".join([hu_path, "de-ner-legal", "de-ner-legal.pt"]),
"de-ner-legal":
"/".join([hu_path, "de-ner-legal", "de-ner-legal.pt"]),
# French models
"fr-ner":
"/".join([hu_path, "fr-ner", "fr-ner-wikiner-0.4.pt"]),
# Dutch models
"nl-ner":
"/".join([hu_path, "nl-ner", "nl-ner-bert-conll02-v0.8.pt"]),
"nl-ner-rnn":
"/".join([hu_path, "nl-ner-rnn", "nl-ner-conll02-v0.5.pt"]),
# Malayalam models
"ml-pos":
"https://raw.githubusercontent.com/qburst/models-repository/master/FlairMalayalamModels/malayalam-xpos-model.pt",
"ml-upos":
"https://raw.githubusercontent.com/qburst/models-repository/master/FlairMalayalamModels/malayalam-upos-model.pt",
# Portuguese models
"pt-pos-clinical":
"/".join([
hu_path,
"pt-pos-clinical",
"pucpr-flair-clinical-pos-tagging-best-model.pt",
]),
# Keyphase models
"keyphrase":
"/".join([hu_path, "keyphrase", "keyphrase-en-scibert.pt"]),
"negation-speculation":
"/".join([
hu_path, "negation-speculation",
"negation-speculation-model.pt"
]),
# Biomedical models
"hunflair-paper-cellline":
"/".join([
hu_path,
"hunflair_smallish_models",
"cellline",
"hunflair-celline-v1.0.pt",
]),
"hunflair-paper-chemical":
"/".join([
hu_path,
"hunflair_smallish_models",
"chemical",
"hunflair-chemical-v1.0.pt",
]),
"hunflair-paper-disease":
"/".join([
hu_path,
"hunflair_smallish_models",
"disease",
"hunflair-disease-v1.0.pt",
]),
"hunflair-paper-gene":
"/".join([
hu_path, "hunflair_smallish_models", "gene",
"hunflair-gene-v1.0.pt"
]),
"hunflair-paper-species":
"/".join([
hu_path,
"hunflair_smallish_models",
"species",
"hunflair-species-v1.0.pt",
]),
"hunflair-cellline":
"/".join([
hu_path,
"hunflair_smallish_models",
"cellline",
"hunflair-celline-v1.0.pt",
]),
"hunflair-chemical":
"/".join([
hu_path,
"hunflair_allcorpus_models",
"huner-chemical",
"hunflair-chemical-full-v1.0.pt",
]),
"hunflair-disease":
"/".join([
hu_path,
"hunflair_allcorpus_models",
"huner-disease",
"hunflair-disease-full-v1.0.pt",
]),
"hunflair-gene":
"/".join([
hu_path,
"hunflair_allcorpus_models",
"huner-gene",
"hunflair-gene-full-v1.0.pt",
]),
"hunflair-species":
"/".join([
hu_path,
"hunflair_allcorpus_models",
"huner-species",
"hunflair-species-full-v1.1.pt",
]),
}
cache_dir = Path("models")
get_from_model_hub = False
# check if model name is a valid local file
if Path(model_name).exists():
model_path = model_name
# check if model key is remapped to HF key - if so, print out information
elif model_name in huggingface_model_map:
# get mapped name
hf_model_name = huggingface_model_map[model_name]
# use mapped name instead
model_name = hf_model_name
get_from_model_hub = True
# if not, check if model key is remapped to direct download location. If so, download model
elif model_name in hu_model_map:
model_path = cached_path(hu_model_map[model_name],
cache_dir=cache_dir)
# special handling for the taggers by the @redewiegergabe project (TODO: move to model hub)
elif model_name == "de-historic-indirect":
model_file = flair.cache_root / cache_dir / "indirect" / "final-model.pt"
if not model_file.exists():
cached_path(
"http://www.redewiedergabe.de/models/indirect.zip",
cache_dir=cache_dir,
)
unzip_file(
flair.cache_root / cache_dir / "indirect.zip",
flair.cache_root / cache_dir,
)
model_path = str(flair.cache_root / cache_dir / "indirect" /
"final-model.pt")
elif model_name == "de-historic-direct":
model_file = flair.cache_root / cache_dir / "direct" / "final-model.pt"
if not model_file.exists():
cached_path(
"http://www.redewiedergabe.de/models/direct.zip",
cache_dir=cache_dir,
)
unzip_file(
flair.cache_root / cache_dir / "direct.zip",
flair.cache_root / cache_dir,
)
model_path = str(flair.cache_root / cache_dir / "direct" /
"final-model.pt")
elif model_name == "de-historic-reported":
model_file = flair.cache_root / cache_dir / "reported" / "final-model.pt"
if not model_file.exists():
cached_path(
"http://www.redewiedergabe.de/models/reported.zip",
cache_dir=cache_dir,
)
unzip_file(
flair.cache_root / cache_dir / "reported.zip",
flair.cache_root / cache_dir,
)
model_path = str(flair.cache_root / cache_dir / "reported" /
"final-model.pt")
elif model_name == "de-historic-free-indirect":
model_file = flair.cache_root / cache_dir / "freeIndirect" / "final-model.pt"
if not model_file.exists():
cached_path(
"http://www.redewiedergabe.de/models/freeIndirect.zip",
cache_dir=cache_dir,
)
unzip_file(
flair.cache_root / cache_dir / "freeIndirect.zip",
flair.cache_root / cache_dir,
)
model_path = str(flair.cache_root / cache_dir / "freeIndirect" /
"final-model.pt")
# for all other cases (not local file or special download location), use HF model hub
else:
get_from_model_hub = True
# if not a local file, get from model hub
if get_from_model_hub:
hf_model_name = "pytorch_model.bin"
revision = "main"
if "@" in model_name:
model_name_split = model_name.split("@")
revision = model_name_split[-1]
model_name = model_name_split[0]
# use model name as subfolder
if "/" in model_name:
model_folder = model_name.split("/", maxsplit=1)[1]
else:
model_folder = model_name
# Lazy import
from huggingface_hub import cached_download, hf_hub_url
url = hf_hub_url(model_name,
revision=revision,
filename=hf_model_name)
try:
model_path = cached_download(
url=url,
library_name="flair",
library_version=flair.__version__,
cache_dir=flair.cache_root / "models" / model_folder,
)
except HTTPError:
# output information
log.error("-" * 80)
log.error(
f"ACHTUNG: The key '{model_name}' was neither found on the ModelHub nor is this a valid path to a file on your system!"
)
# log.error(f" - Error message: {e}")
log.error(
" -> Please check https://huggingface.co/models?filter=flair for all available models."
)
log.error(
" -> Alternatively, point to a model file on your local drive."
)
log.error("-" * 80)
Path(flair.cache_root / "models" /
model_folder).rmdir() # remove folder again if not valid
return model_path
def load(model: str):
model_file = None
aws_resource_path = 'https://s3.eu-central-1.amazonaws.com/alan-nlp/resources/models-v0.2'
if model.lower() == 'ner':
base_path = '/'.join([aws_resource_path,
'NER-conll03--h256-l1-b32-%2Bglove%2Bnews-forward%2Bnews-backward--v0.2',
'en-ner-conll03-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'ner-fast':
base_path = '/'.join([aws_resource_path,
'NER-conll03--h256-l1-b32-experimental--fast-v0.2',
'en-ner-fast-conll03-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'ner-ontonotes':
base_path = '/'.join([aws_resource_path,
'NER-ontoner--h256-l1-b32-%2Bcrawl%2Bnews-forward%2Bnews-backward--v0.2',
'en-ner-ontonotes-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'ner-ontonotes-fast':
base_path = '/'.join([aws_resource_path,
'NER-ontoner--h256-l1-b32-%2Bcrawl%2Bnews-forward-fast%2Bnews-backward-fast--v0.2',
'en-ner-ontonotes-fast-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'pos':
base_path = '/'.join([aws_resource_path,
'POS-ontonotes--h256-l1-b32-%2Bmix-forward%2Bmix-backward--v0.2',
'en-pos-ontonotes-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'pos-fast':
base_path = '/'.join([aws_resource_path,
'POS-ontonotes--h256-l1-b32-%2Bnews-forward-fast%2Bnews-backward-fast--v0.2',
'en-pos-ontonotes-fast-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'frame':
base_path = '/'.join([aws_resource_path,
'FRAME-conll12--h256-l1-b8-%2Bnews%2Bnews-forward%2Bnews-backward--v0.2',
'en-frame-ontonotes-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'frame-fast':
base_path = '/'.join([aws_resource_path,
'FRAME-conll12--h256-l1-b8-%2Bnews%2Bnews-forward-fast%2Bnews-backward-fast--v0.2',
'en-frame-ontonotes-fast-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'chunk':
base_path = '/'.join([aws_resource_path,
'NP-conll2000--h256-l1-b32-%2Bnews-forward%2Bnews-backward--v0.2',
'en-chunk-conll2000-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'chunk-fast':
base_path = '/'.join([aws_resource_path,
'NP-conll2000--h256-l1-b32-%2Bnews-forward-fast%2Bnews-backward-fast--v0.2',
'en-chunk-conll2000-fast-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'de-pos':
base_path = '/'.join([aws_resource_path,
'UPOS-udgerman--h256-l1-b8-%2Bgerman-forward%2Bgerman-backward--v0.2',
'de-pos-ud-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'de-ner':
base_path = '/'.join([aws_resource_path,
'NER-conll03ger--h256-l1-b32-%2Bde-fasttext%2Bgerman-forward%2Bgerman-backward--v0.2',
'de-ner-conll03-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model.lower() == 'de-ner-germeval':
base_path = '/'.join([aws_resource_path,
'NER-germeval--h256-l1-b32-%2Bde-fasttext%2Bgerman-forward%2Bgerman-backward--v0.2',
'de-ner-germeval-v0.2.pt'])
model_file = cached_path(base_path, cache_dir='models')
if model_file is not None:
tagger: SequenceTagger = SequenceTagger.load_from_file(model_file)
return tagger
