def __init__(
self,
name,
tfds_name,
text_preprocessor,
metric_fns,
tfds_data_dir=None,
splits=None,
**task_kwargs):
"""TfdsTask constructor.
Args:
name: string, a unique name for the Task. A ValueError will be raised if
another task with this name is already registered.
tfds_name: string, the name and version number of a TFDS dataset,
optionally with a config.
text_preprocessor: a function (or list of functions) that (each) takes in
a tf.data.Dataset of string features and returns a tf.data.Dataset of
string features. Can be set to None as a no-op. If a list is given, they
will be executed sequentially.
metric_fns: list(callable), list of metric functions with the signature
metric_fn(targets, predictions) to use during evaluation.
tfds_data_dir: string, an optional path to a specific TFDS data directory
to use.
splits: a list(string) of allowable splits to load, a dict mapping
allowable canonical splits (e.g., 'validation') to TFDS splits or slices
(e.g., 'train[':1%']), or None. The default, None, uses all available
splits from the TFDS dataset info.
**task_kwargs: dict, additional keyword arguments for the parent `Task`
class.
"""
super().__init__(
name,
source=seqio.TfdsDataSource(
tfds_name=tfds_name, tfds_data_dir=tfds_data_dir, splits=splits),
text_preprocessor=text_preprocessor,
metric_fns=metric_fns,
dataset_fn=None,
splits=None,
**task_kwargs)
"es", "et", "eu", "fa", "fi", "fr", "fy", "ga", "gl", "gu", "he", "hi",
"hr", "ht", "hu", "hy", "id", "io", "is", "it", "ja", "jv", "ka", "kk",
"kn", "ko", "ky", "la", "lb", "lmo", "lt", "lv", "mg", "min", "mk", "ml",
"mn", "mr", "ms", "my", "nds-nl", "ne", "new", "nl", "nn", "no", "oc",
"pa", "pl", "pms", "pnb", "pt", "ro", "ru", "scn", "sco", "sh", "sk", "sl",
"sq", "sr", "su", "sv", "sw", "ta", "te", "tg", "th", "tl", "tr", "tt",
"uk", "ur", "uz", "vi", "vo", "war", "yo", "zh"
]
# =========================== Pretraining Tasks/Mixtures =======================
# mC4
for lang in MC4_LANGS:
seqio.TaskRegistry.add("mc4.{}".format(lang.replace("-", "_")),
source=seqio.TfdsDataSource(
tfds_name="c4/multilingual:3.0.1",
splits={
"train": lang,
"validation": f"{lang}-validation"
}),
preprocessors=[
functools.partial(t5.data.preprocessors.rekey,
key_map={
"inputs": None,
"targets": "text"
}),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
t5.data.preprocessors.span_corruption,
seqio.preprocessors.append_eos_after_trim,
],
output_features=DEFAULT_OUTPUT_FEATURES,
metric_fns=[])
add_eos=False,
required=False,
dtype=tf.int32),
"targets":
seqio.Feature(vocabulary=DEFAULT_VOCAB, add_eos=True, dtype=tf.int32)
}
# CLASSIFICATION_OUTPUT_FEATURES = {
# "inputs": seqio.Feature(
# vocabulary=DEFAULT_VOCAB, add_eos=False, required=False, dtype=tf.int32)
# }
# ============ KLUE topic classification: Generative ============
seqio.TaskRegistry.add(
"klue_tc_gen",
seqio.TfdsDataSource(tfds_name="klue/tc:1.0.0"),
preprocessors=[
functools.partial(seqio.preprocessors.rekey,
key_map={
"id": "guid",
"title": "title",
"label": "label",
}),
functools.partial(
base_preproc_for_classification,
benchmark_name='klue_tc',
input_keys=['title'],
label_names=KLUE_META['tc_classes'],
with_feature_key=True,
), seqio.preprocessors.tokenize, seqio.preprocessors.append_eos
],
# evaluation procedure.
# The model is trained to predict all ground-truth answers
# and is only considered correct if it predicts all answers for any one of the
# annotators. As in the official evaluation, we consider questions with fewer
# than two non-null annotations unanswerable (given the context) but because we
# cannot predict unanswerability without the context, we only compute the recall
# metric. Further, because our model does not have access to the oracle context,
# we also normalize predicted and ground-truth answers when comparing them.
# This task uses a portion of the train set for validation.
TaskRegistry.add(
"natural_questions_nocontext",
source=seqio.TfdsDataSource(
tfds_name="natural_questions:0.0.2",
splits={
"train": f"train[{NQ_TRAIN_SPLIT_START}:{NQ_TRAIN_SPLIT_END}]",
"validation": f"train[:{NQ_TRAIN_SPLIT_START}]",
"test": "validation"
}),
preprocessors=[
preprocessors.natural_questions_nocontext,
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
output_features=DEFAULT_OUTPUT_FEATURES,
postprocess_fn=postprocessors.natural_questions,
metric_fns=[
functools.partial(
metrics.natural_questions,
# Train set does not contain multiple annotations.
WMT14_VOCAB_EXTRA_100 = seqio.SentencePieceVocabulary(WMT14_CUSTOM_SPM_PATH,
extra_ids=100)
EN_VOCAB_EXTRA_100 = seqio.SentencePieceVocabulary(EN_VOCAB_SPM_PATH,
extra_ids=100)
EN_VOCAB_OUTPUT_FEATURES = {
"inputs": seqio.Feature(vocabulary=EN_VOCAB_EXTRA_100, add_eos=True),
"targets": seqio.Feature(vocabulary=EN_VOCAB_EXTRA_100, add_eos=True)
}
#================================ English only vocab ===========================
for version in ("2.2.0", "2.3.0", "2.3.1"):
TaskRegistry.add(
"c4_v{}_unsupervised_en32k".format(version.replace(".", "")),
source=seqio.TfdsDataSource(tfds_name="c4/en:{}".format(version)),
preprocessors=[
functools.partial(t5_preprocessors.rekey,
key_map={
"inputs": None,
"targets": "text"
}),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
t5_preprocessors.unsupervised,
seqio.preprocessors.append_eos_after_trim,
],
output_features=EN_VOCAB_OUTPUT_FEATURES,
metric_fns=[])
#================================ XSUM =========================================
from t5.data import glue_utils
from t5.data import postprocessors as t5_postprocessors
from t5.data import preprocessors as t5_preprocessors
from t5.evaluation import metrics as t5_metrics
import tensorflow_datasets as tfds
TaskRegistry = seqio.TaskRegistry
DEFAULT_OUTPUT_FEATURES = {
"inputs": seqio.Feature(vocabulary=get_default_vocabulary(), add_eos=True),
"targets": seqio.Feature(vocabulary=get_default_vocabulary(), add_eos=True)
}
# ======================== CoS-E Corpus Task ==================================
TaskRegistry.add("cos_e_v001",
source=seqio.TfdsDataSource(tfds_name="cos_e:0.0.1"),
preprocessors=[
preprocessors.cos_e,
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
postprocess_fn=postprocessors.abstractive_explanations,
output_features=DEFAULT_OUTPUT_FEATURES,
metric_fns=[metrics.esnli_metric])
# CoS-E with no explanations, and modified prefixes like e-SNLI.
TaskRegistry.add("cos_e_v001_0_expln_like_esnli",
source=seqio.TfdsDataSource(tfds_name="cos_e:0.0.1"),
preprocessors=[
functools.partial(preprocessors.cos_e,
"seqio.SentencePieceVocabulary.sentencepiece_model_file")
custom_vocab = seqio.SentencePieceVocabulary(sentence_piece_model_path,
extra_ids)
return {
"inputs":
seqio.Feature(vocabulary=custom_vocab, add_eos=add_eos,
required=False),
"targets":
seqio.Feature(vocabulary=custom_vocab, add_eos=add_eos)
}
# C4 c4_v220_span_corruption data with a custom sentencepiece.
TaskRegistry.add("c4_v220_span_corruption_custom_sp",
source=seqio.TfdsDataSource(tfds_name="c4/en:2.2.0"),
preprocessors=[
functools.partial(preprocessors.rekey,
key_map={
"inputs": None,
"targets": "text"
}),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
preprocessors.span_corruption,
seqio.preprocessors.append_eos_after_trim,
],
output_features=get_custom_output_features(),
metric_fns=[])
# =================================== GLUE =====================================
def create_xnli_tasks_and_mixtures(task_prefix, task_suffix, output_features):
"""Helper function to create XNLI tasks and mixtures."""
if task_suffix:
task_suffix = "_" + task_suffix
seqio.TaskRegistry.add(
f"{task_prefix}xnli_train{task_suffix}",
source=seqio.TfdsDataSource(
tfds_name="multi_nli:1.1.0", splits=["train"]),
preprocessors=[
preprocessors.process_mnli,
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
output_features=output_features,
metric_fns=[metrics.accuracy])
for xnli_lang in XNLI_LANGS:
seqio.TaskRegistry.add(
f"{task_prefix}xnli_dev_test{task_suffix}.{xnli_lang}",
source=seqio.TfdsDataSource(
tfds_name="xnli:1.1.0", splits=["validation", "test"]),
preprocessors=[
functools.partial(
preprocessors.process_xnli, target_languages=[xnli_lang]),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
output_features=output_features,
metric_fns=[metrics.accuracy])
if xnli_lang == "en":
continue
seqio.TaskRegistry.add(
f"{task_prefix}xnli_translate_train{task_suffix}.{xnli_lang}",
source=seqio.TfdsDataSource(
tfds_name="xtreme_xnli:1.1.0", splits=["train"]),
preprocessors=[
functools.partial(
preprocessors.process_xnli, target_languages=[xnli_lang]),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
output_features=output_features,
metric_fns=[metrics.accuracy])
seqio.TaskRegistry.add(
f"{task_prefix}xnli_dev_test{task_suffix}.all_langs",
source=seqio.TfdsDataSource(
tfds_name="xnli:1.1.0", splits=["validation", "test"]),
preprocessors=[
functools.partial(
preprocessors.process_xnli, target_languages=XNLI_LANGS),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
seqio.preprocessors.append_eos_after_trim,
],
output_features=output_features,
metric_fns=[metrics.accuracy])
xnli_zeroshot = ([
f"{task_prefix}xnli_train{task_suffix}",
f"{task_prefix}xnli_dev_test{task_suffix}.all_langs"
] + [
f"{task_prefix}xnli_dev_test{task_suffix}.{lang}" for lang in XNLI_LANGS
])
seqio.MixtureRegistry.add(
f"{task_prefix}xnli_zeroshot{task_suffix}",
xnli_zeroshot,
default_rate=1.0)
xnli_translate_train = xnli_zeroshot + [
f"{task_prefix}xnli_translate_train{task_suffix}.{lang}"
for lang in XNLI_LANGS
if lang != "en"
]
seqio.MixtureRegistry.add(
f"{task_prefix}xnli_translate_train{task_suffix}",
xnli_translate_train,
default_rate=1.0)
"hr", "ht", "hu", "hy", "id", "io", "is", "it", "ja", "jv", "ka", "kk",
"kn", "ko", "ky", "la", "lb", "lmo", "lt", "lv", "mg", "min", "mk", "ml",
"mn", "mr", "ms", "my", "nds-nl", "ne", "new", "nl", "nn", "no", "oc",
"pa", "pl", "pms", "pnb", "pt", "ro", "ru", "scn", "sco", "sh", "sk", "sl",
"sq", "sr", "su", "sv", "sw", "ta", "te", "tg", "th", "tl", "tr", "tt",
"uk", "ur", "uz", "vi", "vo", "war", "yo", "zh"
]
# =========================== Pretraining Tasks/Mixtures =======================
# mC4
for lang in MC4_LANGS:
seqio.TaskRegistry.add(
"mc4.{}".format(lang.replace("-", "_")),
source=seqio.TfdsDataSource(
tfds_name="c4/multilingual:3.0.1",
splits={
"train": lang,
"validation": f"{lang}-validation"
}),
preprocessors=[
functools.partial(
t5.data.preprocessors.rekey,
key_map={
"inputs": None,
"targets": "text"
}),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
t5.data.preprocessors.span_corruption,
seqio.preprocessors.append_eos_after_trim,
],
output_features=DEFAULT_OUTPUT_FEATURES,
DEFAULT_OUTPUT_FEATURES = {
"inputs": seqio.Feature(
vocabulary=t5.data.get_default_vocabulary(), add_eos=True,
required=False),
"targets": seqio.Feature(
vocabulary=t5.data.get_default_vocabulary(), add_eos=True)
}
# ==================================== C4 ======================================
# Final pretraining task used in Raffel et al., 2019.
TaskRegistry.add(
"c4_v220_span_corruption",
source=seqio.TfdsDataSource(tfds_name="c4/en:2.2.0"),
preprocessors=[
functools.partial(
preprocessors.rekey, key_map={
"inputs": None,
"targets": "text"
}),
seqio.preprocessors.tokenize,
seqio.CacheDatasetPlaceholder(),
preprocessors.span_corruption,
seqio.preprocessors.append_eos_after_trim,
],
output_features=DEFAULT_OUTPUT_FEATURES,
metric_fns=[])