File size: 11,429 Bytes
3f0d9fa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 |
# coding=utf-8
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""HuggingFace datasets implementation of the json tasks in the BIG-Bench Dataset.
For the programatic tasks, please use the BIG-Bench API on github.com/google/BIG-bench.
"""
from typing import Optional
import bigbench.api.util as bb_utils # From: "bigbench @ https://storage.googleapis.com/public_research_data/bigbench/bigbench-0.0.1.tar.gz"
import bigbench.bbseqio.bigbench_bridge as bbb
from bigbench.api import json_task
from bigbench.bbseqio import bigbench_json_paths as bb_json_paths
from sentencepiece import sentencepiece_model_pb2 # noqa: this is also required by bigbench.api.util
import datasets
logger = datasets.logging.get_logger(__name__)
_CITATION = """\
@InProceedings{bigbench,
title = {Beyond the Imitation Game: Quantifying and extrapolating the
capabilities of language models},
author={BIG-Bench Collaboration
},
year={2022}
}
"""
_DESCRIPTION = """\
The Beyond the Imitation Game Benchmark (BIG-bench) is a collaborative benchmark intended to
probe large language models, and extrapolate their future capabilities.
"""
_HOMEPAGE = "https://github.com/google/BIG-bench"
_LICENSE = "Apache License 2.0"
def div_or_none(x, y):
return x // y if x else x
def validate_task_name(task_name: str) -> None:
"""Check that the requested task name is a valid bigbench json task."""
if task_name in bb_utils.get_all_json_task_names():
return
elif task_name in bb_utils.get_all_programmatic_task_names():
raise ValueError(
"BIG-Bench does not support programmatic tasks through HuggingFace datasets"
f"Please see {_HOMEPAGE} for more information for how to interact with the programmatic tasks."
)
else:
raise ValueError(
f"Invalid task_name. Got task_name = {task_name}. Please choose one from:\n -- "
+ "\n -- ".join(bb_utils.get_all_json_task_names())
)
def validate_subtask_name(task_name: str, subtask_name: str) -> None:
"""Check that the requested subtask name is a valid bigbench subtask."""
subtasks = [name.split(":")[-1] for name in bb_utils.get_subtask_names_from_task(task_name)]
if not subtasks:
raise ValueError(f"Task {task_name} has no subtasks. Got subtask_name = {subtask_name}.")
elif subtask_name not in subtasks:
raise ValueError(
f"Invalid subtask_name {subtask_name} for task {task_name}. Please choose one from:\n -- "
+ "\n -- ".join(subtasks)
)
class BigBenchConfig(datasets.BuilderConfig):
def __init__(
self,
name,
subtask_name: Optional[str] = None,
num_shots: int = 0,
max_examples: Optional[int] = None,
**kwargs,
):
super().__init__(
name=name,
**kwargs,
)
"""BIG-bench configuration.
Args:
name: BIG-bench task name.
subtask_name: BIG-bench subtask name. Accepts both "task_name:subtask_name" and "subtask_name" formats.
num_shots: Number of few-shot examples in input prompt. Default is zero.
max_examples: Limit number of examples for each task. Default is including all examples.
"""
self.task_name = name
self.subtask_name = subtask_name
self.num_shots = num_shots
self.max_examples = max_examples
class Bigbench(datasets.GeneratorBasedBuilder):
"""The Beyond the Imitation Game Benchmark (BIG-bench) is a collaborative benchmark
intended to probe large language models, and extrapolate their future capabilities."""
VERSION = datasets.Version("1.0.0")
BUILDER_CONFIG_CLASS = BigBenchConfig
BUILDER_CONFIGS = [
BigBenchConfig(name=name, version=datasets.Version("1.0.0")) for name in bb_utils.get_all_json_task_names()
]
def _info(self):
features = datasets.Features(
{
"idx": datasets.Value("int32"),
"inputs": datasets.Value("string"),
"targets": datasets.Sequence(datasets.Value("string")),
"multiple_choice_targets": datasets.Sequence(datasets.Value("string")),
"multiple_choice_scores": datasets.Sequence(datasets.Value("int32")),
}
)
return datasets.DatasetInfo(
# This is the description that will appear on the datasets page.
description=_DESCRIPTION,
# This defines the different columns of the dataset and their types
features=features, # Here we define them above because they are different between the two configurations
# If there's a common (input, target) tuple from the features,
# specify them here. They'll be used if as_supervised=True in
# builder.as_dataset.
supervised_keys=None,
# Homepage of the dataset for documentation
homepage=_HOMEPAGE,
# License for the dataset if available
license=_LICENSE,
# Citation for the dataset
citation=_CITATION,
)
def _split_generators(self, dl_manager):
"""Returns SplitGenerators."""
return [
datasets.SplitGenerator(
name=datasets.splits.NamedSplit("default"), # TODO(ajandreassen): Is there a way to call this 'all'?
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "all",
},
),
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "train",
},
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "validation",
},
),
]
def _generate_examples(
self,
split, # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
):
validate_task_name(self.config.task_name)
if self.config.subtask_name:
# Subtasks are sometimes in bigbench written as task_name:subtask_name.
# We want to remove the task_name from the subtask names:
self.config.subtask_name = self.config.subtask_name.split(":")[-1]
validate_subtask_name(self.config.task_name, self.config.subtask_name)
"""Yields examples as (key, example) tuples."""
if split == "all":
# not cutoff in number of examples for 'all' split
MIN_VALIDATION_EXAMPLES = 0
else:
MIN_VALIDATION_EXAMPLES = 16
try:
task_path, json_util = bb_json_paths.get_task_path(self.config.task_name)
has_subtasks = bb_json_paths.has_subtasks(self.config.task_name)
if has_subtasks:
subtask_names = bb_json_paths.get_subtask_names(self.config.task_name)
num_subtasks = len(subtask_names)
min_validation_examples_per_subtask = div_or_none(MIN_VALIDATION_EXAMPLES, num_subtasks)
if not has_subtasks:
ds_fn = bbb.get_dataset_fn(
task_name=self.config.task_name,
task_path=task_path,
subtask_name=None,
num_shots=self.config.num_shots,
bigbench_task_type=bbb.BigBenchTaskType.HUGGINGFACE,
max_examples=self.config.max_examples,
json_util=json_util,
min_validation_examples=MIN_VALIDATION_EXAMPLES,
format_fn=json_task.default_format_fn,
)
ds_list = [ds_fn(split)]
elif self.config.subtask_name is not None:
ds_fn = bbb.get_dataset_fn(
task_name=self.config.task_name,
task_path=task_path,
subtask_name=self.config.subtask_name,
num_shots=self.config.num_shots,
bigbench_task_type=bbb.BigBenchTaskType.HUGGINGFACE,
max_examples=self.config.max_examples,
json_util=json_util,
min_validation_examples=min_validation_examples_per_subtask,
format_fn=json_task.default_format_fn,
)
ds_list = [ds_fn(split)]
else:
# Create mixture of all subtasks
ds_list = []
for subtask_name in subtask_names:
subtask_name = subtask_name.split(":")[-1]
logger.info(f"Loading subtask {split} split", subtask_name)
ds_fn = bbb.get_dataset_fn(
task_name=self.config.task_name,
task_path=task_path,
subtask_name=subtask_name,
num_shots=self.config.num_shots,
bigbench_task_type=bbb.BigBenchTaskType.HUGGINGFACE,
max_examples=div_or_none(self.config.max_examples, num_subtasks),
json_util=json_util,
min_validation_examples=min_validation_examples_per_subtask,
format_fn=json_task.default_format_fn,
)
ds_list.append(ds_fn(split))
except ValueError as value_error:
# BIG-Bench requires at least 16 examples to use the train & validation splits,
# while using 'all'/'default' does not have such a requirement.
if "has too few examples" in value_error.args[0] and split != "all":
logger.warning(
f"-- WARNING: skipping split {split} because it has too few examples. Please use 'default' split."
)
logger.warning(value_error)
return
raise value_error
unique_key_counter = 0
for ds in ds_list:
for example in ds:
unique_key_counter += 1
yield unique_key_counter, {
"idx": example["idx"],
"inputs": example["inputs"].numpy().decode().strip(),
"targets": [target.numpy().decode().strip() for target in example["targets"]],
"multiple_choice_targets": [
targets.decode().strip() for targets in example["multiple_choice_targets"].numpy()
],
"multiple_choice_scores": [scores for scores in example["multiple_choice_scores"].numpy()],
}
|