Merge pull request #307 from Taekyoon/multilingual-ko-korsts

Add klue-sts task to eval Korean language task

lm_eval/datasets/lambada/lambada.py

+# 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.
+# TODO: Address all TODOs and remove all explanatory comments
+"""LAMBADA dataset."""
+
+
+import json
+
+import datasets
+
+
+_CITATION = """\
+@misc{
+    author={Paperno, Denis and Kruszewski, Germán and Lazaridou, Angeliki and Pham, Quan Ngoc and Bernardi, Raffaella and Pezzelle, Sandro and Baroni, Marco and Boleda, Gemma and Fernández, Raquel}, 
+    title={The LAMBADA dataset},
+    DOI={10.5281/zenodo.2630551},
+    publisher={Zenodo},
+    year={2016},
+    month={Aug}
+}
+"""
+
+_DESCRIPTION = """\
+LAMBADA is a dataset to evaluate the capabilities of computational models for text
+understanding by means of a word prediction task. LAMBADA is a collection of narrative
+texts sharing the characteristic that human subjects are able to guess their last
+word if they are exposed to the whole text, but not if they only see the last
+sentence preceding the target word. To succeed on LAMBADA, computational models
+cannot simply rely on local context, but must be able to keep track of information
+in the broader discourse.
+"""
+
+_HOMEPAGE = "https://zenodo.org/record/2630551#.X4Xzn5NKjUI"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = {
+    "original": "http://eaidata.bmk.sh/data/lambada_test.jsonl",
+    "en": "http://eaidata.bmk.sh/data/lambada_test_en.jsonl",
+    "fr": "http://eaidata.bmk.sh/data/lambada_test_fr.jsonl",
+    "de": "http://eaidata.bmk.sh/data/lambada_test_de.jsonl",
+    "it": "http://eaidata.bmk.sh/data/lambada_test_it.jsonl",
+    "es": "http://eaidata.bmk.sh/data/lambada_test_es.jsonl",
+}
+
+
+class Lambada(datasets.GeneratorBasedBuilder):
+    """LAMBADA is a dataset to evaluate the capabilities of computational models for text understanding by means of a word prediction task."""
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="original", version=VERSION, description="The LAMBADA dataset"),
+        datasets.BuilderConfig(name="en", version=VERSION, description="The English translated LAMBADA dataset"),
+        datasets.BuilderConfig(name="fr", version=VERSION, description="The French translated LAMBADA dataset"),
+        datasets.BuilderConfig(name="de", version=VERSION, description="The German translated LAMBADA dataset"),
+        datasets.BuilderConfig(name="it", version=VERSION, description="The Italian translated LAMBADA dataset"),
+        datasets.BuilderConfig(name="es", version=VERSION, description="The Spanish translated LAMBADA dataset"),
+    ]
+
+    DEFAULT_CONFIG_NAME = "original"
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "text": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=f"{_DESCRIPTION}\n{self.config.description}",
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = _URLS[self.config.name]
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir,
+                    "split": "validation",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        with open(filepath, encoding="utf-8") as f:
+            for key, row in enumerate(f):
+                data = json.loads(row)
+                yield key, {
+                    "text": data["text"]
+                }

lm_eval/datasets/logiqa/dataset_infos.json

+{"logiqa": {"description": "LogiQA is a dataset for testing human logical reasoning. It consists of 8,678 QA\ninstances, covering multiple types of deductive reasoning. Results show that state-\nof-the-art neural models perform by far worse than human ceiling. The dataset can\nalso serve as a benchmark for reinvestigating logical AI under the deep learning\nNLP setting.\n", "citation": "@misc{liu2020logiqa,\n    title={LogiQA: A Challenge Dataset for Machine Reading Comprehension with Logical Reasoning}, \n    author={Jian Liu and Leyang Cui and Hanmeng Liu and Dandan Huang and Yile Wang and Yue Zhang},\n    year={2020},\n    eprint={2007.08124},\n    archivePrefix={arXiv},\n    primaryClass={cs.CL}\n}\n", "homepage": "https://github.com/lgw863/LogiQA-dataset", "license": "", "features": {"label": {"dtype": "string", "id": null, "_type": "Value"}, "context": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "options": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "logiqa", "config_name": "logiqa", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 6419852, "num_examples": 7376, "dataset_name": "logiqa"}, "test": {"name": "test", "num_bytes": 571705, "num_examples": 651, "dataset_name": "logiqa"}, "validation": {"name": "validation", "num_bytes": 562437, "num_examples": 651, "dataset_name": "logiqa"}}, "download_checksums": {"https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Train.txt": {"num_bytes": 6281272, "checksum": "7d5bb1f58278e33b395744cd2ad8d7600faa0b3c4d615c659a44ec1181d759fa"}, "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Test.txt": {"num_bytes": 559060, "checksum": "359acb78c37802208f7fde9e2f6574b8526527c63d6a336f90a53f1932cb4701"}, "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Eval.txt": {"num_bytes": 550021, "checksum": "4c49e6753b7262c001506b9151135abf722247035ab075dad93acdea5789c01f"}}, "download_size": 7390353, "post_processing_size": null, "dataset_size": 7553994, "size_in_bytes": 14944347}}
\ No newline at end of file

lm_eval/datasets/logiqa/logiqa.py

+# 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.
+"""LogiQA dataset."""
+
+
+import datasets
+
+
+_CITATION = """\
+@misc{liu2020logiqa,
+    title={LogiQA: A Challenge Dataset for Machine Reading Comprehension with Logical Reasoning}, 
+    author={Jian Liu and Leyang Cui and Hanmeng Liu and Dandan Huang and Yile Wang and Yue Zhang},
+    year={2020},
+    eprint={2007.08124},
+    archivePrefix={arXiv},
+    primaryClass={cs.CL}
+}
+"""
+
+_DESCRIPTION = """\
+LogiQA is a dataset for testing human logical reasoning. It consists of 8,678 QA
+instances, covering multiple types of deductive reasoning. Results show that state-
+of-the-art neural models perform by far worse than human ceiling. The dataset can
+also serve as a benchmark for reinvestigating logical AI under the deep learning
+NLP setting.
+"""
+
+_HOMEPAGE = "https://github.com/lgw863/LogiQA-dataset"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = {
+    "train": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Train.txt",
+    "validation": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Eval.txt",
+    "test": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Test.txt",
+}
+
+
+class Logiqa(datasets.GeneratorBasedBuilder):
+    """LogiQA: A Challenge Dataset for Machine Reading Comprehension with Logical Reasoning"""
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="logiqa", version=VERSION, description="The LogiQA dataset."),
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "label": datasets.Value("string"),
+                "context": datasets.Value("string"),
+                "question": datasets.Value("string"),
+                "options": datasets.features.Sequence(
+                    datasets.Value("string")
+                ),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = {"train": _URLS["train"], "test": _URLS["test"], "validation": _URLS["validation"]}
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["train"],
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["test"],
+                    "split": "test"
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["validation"],
+                    "split": "validation",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        def normalize(text):
+            return text.replace(".", ". ").strip()
+        with open(filepath, encoding="utf-8") as f:
+            data = f.read().strip().split("\n\n")
+            for key, row in enumerate(data):
+                example = row.split("\n")
+                yield key, {
+                    "label": example[0].strip(),
+                    "context": normalize(example[1]),
+                    "question": normalize(example[2]),
+                    "options": [normalize(option[2:]) for option in example[3:]],
+                }

lm_eval/datasets/mutual/dataset_infos.json

+{"mutual": {"description": "MuTual is a retrieval-based dataset for multi-turn dialogue reasoning, which is\nmodified from Chinese high school English listening comprehension test data.\n\nThe MuTual dataset.", "citation": "@inproceedings{mutual,\n    title = \"MuTual: A Dataset for Multi-Turn Dialogue Reasoning\",\n    author = \"Cui, Leyang  and Wu, Yu and Liu, Shujie and Zhang, Yue and Zhou, Ming\" ,\n    booktitle = \"Proceedings of the 58th Conference of the Association for Computational Linguistics\",\n    year = \"2020\",\n    publisher = \"Association for Computational Linguistics\",\n}\n", "homepage": "https://github.com/Nealcly/MuTual", "license": "", "features": {"answers": {"dtype": "string", "id": null, "_type": "Value"}, "options": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "article": {"dtype": "string", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "mutual", "config_name": "mutual", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 5141602, "num_examples": 7088, "dataset_name": "mutual"}, "test": {"name": "test", "num_bytes": 634396, "num_examples": 886, "dataset_name": "mutual"}, "validation": {"name": "validation", "num_bytes": 624271, "num_examples": 886, "dataset_name": "mutual"}}, "download_checksums": {"https://github.com/Nealcly/MuTual/archive/master.zip": {"num_bytes": 10997878, "checksum": "bb325cf6c672f0f02699993a37138b0fa0af6fcfc77ec81dfbe46add4d7b29f9"}}, "download_size": 10997878, "post_processing_size": null, "dataset_size": 6400269, "size_in_bytes": 17398147}, "mutual_plus": {"description": "MuTual is a retrieval-based dataset for multi-turn dialogue reasoning, which is\nmodified from Chinese high school English listening comprehension test data.\n\nMuTualPlus is a more difficult MuTual that replaces positive responses with a safe responses.", "citation": "@inproceedings{mutual,\n    title = \"MuTual: A Dataset for Multi-Turn Dialogue Reasoning\",\n    author = \"Cui, Leyang  and Wu, Yu and Liu, Shujie and Zhang, Yue and Zhou, Ming\" ,\n    booktitle = \"Proceedings of the 58th Conference of the Association for Computational Linguistics\",\n    year = \"2020\",\n    publisher = \"Association for Computational Linguistics\",\n}\n", "homepage": "https://github.com/Nealcly/MuTual", "license": "", "features": {"answers": {"dtype": "string", "id": null, "_type": "Value"}, "options": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "article": {"dtype": "string", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "mutual", "config_name": "mutual_plus", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 4921179, "num_examples": 7088, "dataset_name": "mutual"}, "test": {"name": "test", "num_bytes": 606620, "num_examples": 886, "dataset_name": "mutual"}, "validation": {"name": "validation", "num_bytes": 597340, "num_examples": 886, "dataset_name": "mutual"}}, "download_checksums": {"https://github.com/Nealcly/MuTual/archive/master.zip": {"num_bytes": 10997878, "checksum": "bb325cf6c672f0f02699993a37138b0fa0af6fcfc77ec81dfbe46add4d7b29f9"}}, "download_size": 10997878, "post_processing_size": null, "dataset_size": 6125139, "size_in_bytes": 17123017}}
\ No newline at end of file

lm_eval/datasets/mutual/mutual.py

+# 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.
+"""MuTual dataset."""
+
+
+import json
+import os
+from pathlib import Path
+
+import datasets
+
+
+_CITATION = """\
+@inproceedings{mutual,
+    title = "MuTual: A Dataset for Multi-Turn Dialogue Reasoning",
+    author = "Cui, Leyang  and Wu, Yu and Liu, Shujie and Zhang, Yue and Zhou, Ming" ,
+    booktitle = "Proceedings of the 58th Conference of the Association for Computational Linguistics",
+    year = "2020",
+    publisher = "Association for Computational Linguistics",
+}
+"""
+
+_DESCRIPTION = """\
+MuTual is a retrieval-based dataset for multi-turn dialogue reasoning, which is
+modified from Chinese high school English listening comprehension test data.
+"""
+
+_HOMEPAGE = "https://github.com/Nealcly/MuTual"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = "https://github.com/Nealcly/MuTual/archive/master.zip"
+
+
+class Mutual(datasets.GeneratorBasedBuilder):
+    """MuTual: A Dataset for Multi-Turn Dialogue Reasoning"""
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="mutual", version=VERSION, description="The MuTual dataset."),
+        datasets.BuilderConfig(name="mutual_plus", version=VERSION, description="MuTualPlus is a more difficult MuTual that replaces positive responses with a safe responses."),
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "answers": datasets.Value("string"),
+                "options": datasets.features.Sequence(datasets.Value("string")),
+                "article": datasets.Value("string"),
+                "id": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=f"{_DESCRIPTION}\n{self.config.description}",
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = _URLS
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "basepath": os.path.join(data_dir, "MuTual-master", "data", self.config.name, "train"),
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "basepath": os.path.join(data_dir, "MuTual-master", "data", self.config.name, "test"),
+                    "split": "test",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "basepath": os.path.join(data_dir, "MuTual-master", "data", self.config.name, "dev"),
+                    "split": "dev",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, basepath, split):
+        # TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
+        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
+        key = 0
+        for file in sorted(Path(basepath).iterdir()):
+            if file.suffix != ".txt":
+                continue
+            with open(file, "r", encoding='utf-8') as f:
+                data_str = f.read()
+                # Ignore the occasional empty file.
+                if not data_str:
+                    continue
+                data = json.loads(data_str)
+                yield key, {
+                    "answers": data["answers"],
+                    "options": data["options"],
+                    "article": data["article"],
+                    "id": data["id"],
+                }
+                key += 1

lm_eval/datasets/pile/dataset_infos.json

+{"pile_arxiv": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nArXiv", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_arxiv", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 113218251, "num_examples": 2407, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 115653720, "num_examples": 2434, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 228871971, "size_in_bytes": 1160030307}, "pile_books3": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nBooks3", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_books3", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 150095743, "num_examples": 269, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 177359876, "num_examples": 301, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 327455619, "size_in_bytes": 1258613955}, "pile_bookcorpus2": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nBookCorpus2", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_bookcorpus2", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 9680652, "num_examples": 28, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 9776271, "num_examples": 26, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 19456923, "size_in_bytes": 950615259}, "pile_dm-mathematics": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nDM Mathematics", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_dm-mathematics", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 15756556, "num_examples": 1922, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 16453386, "num_examples": 2007, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 32209942, "size_in_bytes": 963368278}, "pile_enron": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nEnron Emails", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_enron", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 1638859, "num_examples": 1010, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 1556487, "num_examples": 947, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 3195346, "size_in_bytes": 934353682}, "pile_europarl": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nEuroParl", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_europarl", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 8789652, "num_examples": 157, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 9111791, "num_examples": 133, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 17901443, "size_in_bytes": 949059779}, "pile_freelaw": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nFreeLaw", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_freelaw", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 80808693, "num_examples": 5101, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 80363814, "num_examples": 5094, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 161172507, "size_in_bytes": 1092330843}, "pile_github": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nGithub", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_github", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 95654706, "num_examples": 18195, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 97179576, "num_examples": 18337, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 192834282, "size_in_bytes": 1123992618}, "pile_gutenberg": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nGutenberg (PG-19)", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_gutenberg", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 30243176, "num_examples": 80, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 24685980, "num_examples": 60, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 54929156, "size_in_bytes": 986087492}, "pile_hackernews": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nHackerNews", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_hackernews", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 8124255, "num_examples": 1632, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 9803822, "num_examples": 1619, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 17928077, "size_in_bytes": 949086413}, "pile_nih-exporter": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nNIH ExPorter", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_nih-exporter", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 3928804, "num_examples": 1884, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 3927967, "num_examples": 1825, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 7856771, "size_in_bytes": 939015107}, "pile_opensubtitles": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nOpenSubtitles", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_opensubtitles", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 21008996, "num_examples": 642, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 19622904, "num_examples": 621, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 40631900, "size_in_bytes": 971790236}, "pile_openwebtext2": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nOpenWebText2", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_openwebtext2", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 128624303, "num_examples": 32925, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 131554302, "num_examples": 33400, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 260178605, "size_in_bytes": 1191336941}, "pile_philpapers": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nPhilPapers", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_philpapers", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 5090158, "num_examples": 68, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 6499078, "num_examples": 64, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 11589236, "size_in_bytes": 942747572}, "pile_pile-cc": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nPile-CC", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_pile-cc", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 235004043, "num_examples": 52790, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 233535650, "num_examples": 52792, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 468539693, "size_in_bytes": 1399698029}, "pile_pubmed-abstracts": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nPubMed Abstracts", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_pubmed-abstracts", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 39908950, "num_examples": 29895, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 40008336, "num_examples": 29871, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 79917286, "size_in_bytes": 1011075622}, "pile_pubmed-central": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nPubMed Central", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_pubmed-central", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 187251519, "num_examples": 5911, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 184791818, "num_examples": 5977, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 372043337, "size_in_bytes": 1303201673}, "pile_stackexchange": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nStackExchange", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_stackexchange", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 66441557, "num_examples": 30378, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 66011397, "num_examples": 29950, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 132452954, "size_in_bytes": 1063611290}, "pile_upsto": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nUSPTO Backgrounds", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_upsto", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 47345405, "num_examples": 11415, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 48122320, "num_examples": 11387, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 95467725, "size_in_bytes": 1026626061}, "pile_ubuntu-irc": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nUbuntu IRC", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_ubuntu-irc", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 5694218, "num_examples": 22, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 7410104, "num_examples": 21, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 13104322, "size_in_bytes": 944262658}, "pile_wikipedia": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nWikipedia (en)", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_wikipedia", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 52166968, "num_examples": 17511, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 53186137, "num_examples": 17478, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 105353105, "size_in_bytes": 1036511441}, "pile_youtubesubtitles": {"description": "The Pile is a 825 GiB diverse, open source language modeling data set that consists\nof 22 smaller, high-quality datasets combined together. To score well on Pile\nBPB (bits per byte), a model must be able to understand many disparate domains\nincluding books, github repositories, webpages, chat logs, and medical, physics,\nmath, computer science, and philosophy papers.\n\nYoutubeSubtitles", "citation": "@article{pile,\n  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},\n  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},\n  journal={arXiv preprint arXiv:2101.00027},\n  year={2020}\n}\n", "homepage": "https://pile.eleuther.ai/", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "pile", "config_name": "pile_youtubesubtitles", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 7377448, "num_examples": 342, "dataset_name": "pile"}, "validation": {"name": "validation", "num_bytes": 8937546, "num_examples": 326, "dataset_name": "pile"}}, "download_checksums": {"http://eaidata.bmk.sh/data/pile/val.jsonl.zst": {"num_bytes": 470907480, "checksum": "264c875d8bbd355d8daa9d032b75fd8fb91606218bb84dd1155b203fcd5fab92"}, "http://eaidata.bmk.sh/data/pile/test.jsonl.zst": {"num_bytes": 460250856, "checksum": "0bb28c52d0b5596d389bf179ce2d43bf7f7ffae76b0d2d20b180c97f62e0975e"}}, "download_size": 931158336, "post_processing_size": null, "dataset_size": 16314994, "size_in_bytes": 947473330}}
\ No newline at end of file

lm_eval/datasets/pile/pile.py

+# 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.
+"""Pile dataset."""
+
+
+import json
+
+import datasets
+
+
+_CITATION = """\
+@article{pile,
+  title={The {P}ile: An 800GB Dataset of Diverse Text for Language Modeling},
+  author={Gao, Leo and Biderman, Stella and Black, Sid and Golding, Laurence and Hoppe, Travis and Foster, Charles and Phang, Jason and He, Horace and Thite, Anish and Nabeshima, Noa and Presser, Shawn and Leahy, Connor},
+  journal={arXiv preprint arXiv:2101.00027},
+  year={2020}
+}
+"""
+
+_DESCRIPTION = """\
+The Pile is a 825 GiB diverse, open source language modeling data set that consists
+of 22 smaller, high-quality datasets combined together. To score well on Pile
+BPB (bits per byte), a model must be able to understand many disparate domains
+including books, github repositories, webpages, chat logs, and medical, physics,
+math, computer science, and philosophy papers.
+"""
+
+_HOMEPAGE = "https://pile.eleuther.ai/"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = {
+    "validation": "http://eaidata.bmk.sh/data/pile/val.jsonl.zst",
+    "test": "http://eaidata.bmk.sh/data/pile/test.jsonl.zst",
+}
+
+_NAMES = {
+    "pile_arxiv": "ArXiv",
+    "pile_books3": "Books3",
+    "pile_bookcorpus2": "BookCorpus2",
+    "pile_dm-mathematics": "DM Mathematics",
+    "pile_enron": "Enron Emails",
+    "pile_europarl": "EuroParl",
+    "pile_freelaw": "FreeLaw",
+    "pile_github": "Github",
+    "pile_gutenberg": "Gutenberg (PG-19)",
+    "pile_hackernews": "HackerNews",
+    "pile_nih-exporter": "NIH ExPorter",
+    "pile_opensubtitles": "OpenSubtitles",
+    "pile_openwebtext2": "OpenWebText2",
+    "pile_philpapers": "PhilPapers",
+    "pile_pile-cc": "Pile-CC",
+    "pile_pubmed-abstracts": "PubMed Abstracts",
+    "pile_pubmed-central": "PubMed Central",
+    "pile_stackexchange": "StackExchange",
+    "pile_upsto": "USPTO Backgrounds",
+    "pile_ubuntu-irc": "Ubuntu IRC",
+    "pile_wikipedia": "Wikipedia (en)",
+    "pile_youtubesubtitles": "YoutubeSubtitles",
+}
+
+
+class Pile(datasets.GeneratorBasedBuilder):
+    """The Pile is a 825 GiB diverse, open source language modeling dataset."""
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name=name, version=version, description=_NAMES[name])
+        for name, version in zip(_NAMES.keys(), [VERSION] * len(_NAMES))
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "text": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=f"{_DESCRIPTION}\n{self.config.description}",
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = {"validation": _URLS["validation"], "test": _URLS["test"]}
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["test"],
+                    "split": "test"
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["validation"],
+                    "split": "validation",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        with open(filepath, encoding="utf-8") as f:
+            for key, row in enumerate(f):
+                data = json.loads(row)
+                if data["meta"]["pile_set_name"] == _NAMES[self.config.name]:
+                    yield key, {
+                        "text": data["text"],
+                    }

lm_eval/datasets/quac/dataset_infos.json

+{"quac": {"description": "Question Answering in Context (QuAC) is a dataset for modeling, understanding, and \nparticipating in information seeking dialog. Data instances consist of an interactive\ndialog between two crowd workers: (1) a student who poses a sequence of freeform\nquestions to learn as much as possible about a hidden Wikipedia text, and (2)\na teacher who answers the questions by providing short excerpts (spans) from the text.\n", "citation": "@article{choi2018quac,\n    title={Quac: Question answering in context},\n    author={Choi, Eunsol and He, He and Iyyer, Mohit and Yatskar, Mark and Yih, Wen-tau and Choi, Yejin and Liang, Percy and Zettlemoyer, Luke},\n    journal={arXiv preprint arXiv:1808.07036},\n    year={2018}\n}\n", "homepage": "https://quac.ai/", "license": "", "features": {"title": {"dtype": "string", "id": null, "_type": "Value"}, "section_title": {"dtype": "string", "id": null, "_type": "Value"}, "paragraph": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "answer": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "quac", "config_name": "quac", "version": {"version_str": "1.1.0", "description": null, "major": 1, "minor": 1, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 212391958, "num_examples": 83568, "dataset_name": "quac"}, "validation": {"name": "validation", "num_bytes": 20678483, "num_examples": 7354, "dataset_name": "quac"}}, "download_checksums": {"https://s3.amazonaws.com/my89public/quac/train_v0.2.json": {"num_bytes": 68114819, "checksum": "ff5cca5a2e4b4d1cb5b5ced68b9fce88394ef6d93117426d6d4baafbcc05c56a"}, "https://s3.amazonaws.com/my89public/quac/val_v0.2.json": {"num_bytes": 8929167, "checksum": "09e622916280ba04c9352acb1bc5bbe80f11a2598f6f34e934c51d9e6570f378"}}, "download_size": 77043986, "post_processing_size": null, "dataset_size": 233070441, "size_in_bytes": 310114427}}
\ No newline at end of file

lm_eval/datasets/quac/quac.py

+# 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.
+# TODO: Address all TODOs and remove all explanatory comments
+"""QuAC dataset."""
+
+
+import json
+
+import datasets
+
+
+_CITATION = """\
+@article{choi2018quac,
+    title={Quac: Question answering in context},
+    author={Choi, Eunsol and He, He and Iyyer, Mohit and Yatskar, Mark and Yih, Wen-tau and Choi, Yejin and Liang, Percy and Zettlemoyer, Luke},
+    journal={arXiv preprint arXiv:1808.07036},
+    year={2018}
+}
+"""
+
+_DESCRIPTION = """\
+Question Answering in Context (QuAC) is a dataset for modeling, understanding, and 
+participating in information seeking dialog. Data instances consist of an interactive
+dialog between two crowd workers: (1) a student who poses a sequence of freeform
+questions to learn as much as possible about a hidden Wikipedia text, and (2)
+a teacher who answers the questions by providing short excerpts (spans) from the text.
+"""
+
+_HOMEPAGE = "https://quac.ai/"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = {
+    "train": "https://s3.amazonaws.com/my89public/quac/train_v0.2.json",
+    "validation": "https://s3.amazonaws.com/my89public/quac/val_v0.2.json",
+}
+
+
+class Quac(datasets.GeneratorBasedBuilder):
+    """Question Answering in Context (QuAC) is a dataset for modeling, understanding, and  participating in information seeking dialog."""
+
+    VERSION = datasets.Version("1.1.0")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="quac", version=VERSION, description="The QuAC dataset"),
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "title": datasets.Value("string"),
+                "section_title": datasets.Value("string"),
+                "paragraph": datasets.Value("string"),
+                "question": datasets.Value("string"),
+                "answer": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = {"train": _URLS["train"], "validation": _URLS["validation"]}
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["train"],
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir["validation"],
+                    "split": "validation"
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        with open(filepath, encoding="utf-8") as f:
+            data = json.load(f)["data"]
+            key = 0
+            for row in data:
+                paragraph = row["paragraphs"][0]["context"].replace("CANNOTANSWER", "")
+                qas = row["paragraphs"][0]["qas"]
+                qa_pairs = [(qa['question'], qa['answers'][0]['text']) for qa in qas]
+                for (question, answer) in qa_pairs:
+                    # Yields examples as (key, example) tuples
+                    yield key, {
+                        "title": row["title"],
+                        "section_title": row["section_title"],
+                        "paragraph": paragraph,
+                        "question": question,
+                        "answer": answer,
+                    }
+                    key += 1

lm_eval/datasets/sat_analogies/sat_analogies.py

+# 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.
+"""SAT Analogy Questions dataset."""
+
+
+import os
+
+import datasets
+
+
+_CITATION = """\
+@article{article,
+    author = {Turney, Peter},
+    year = {2006},
+    month = {09},
+    pages = {379-416},
+    title = {Similarity of Semantic Relations},
+    volume = {32},
+    journal = {Computational Linguistics},
+    doi = {10.1162/coli.2006.32.3.379}
+}
+"""
+
+_DESCRIPTION = """\
+SAT (Scholastic Aptitude Test) Analogy Questions is a dataset comprising 374
+multiple-choice analogy questions; 5 choices per question.
+"""
+
+_HOMEPAGE = "https://aclweb.org/aclwiki/SAT_Analogy_Questions_(State_of_the_art)"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+
+class SatAnalogies(datasets.GeneratorBasedBuilder):
+    """ SAT (Scholastic Aptitude Test) Analogy Questions is a dataset comprising 374 multiple-choice analogy questions. """
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="sat_analogies", version=VERSION,
+                               description="The SAT Analogy Questions dataset"),
+    ]
+
+    @property
+    def manual_download_instructions(self):
+        return (
+            "To use SAT Analogy Questions you have to download it manually. Please "
+            "email Peter Turney to request the data (https://www.apperceptual.com). "
+            "Once you recieve a download link for the dataset, supply the local path "
+            "as the `data_dir` arg: "
+            "`datasets.load_dataset('sat_analogies', data_dir='path/to/folder/folder_name')`"
+        )
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "source": datasets.Value("string"),
+                "stem": datasets.Value("string"),
+                "choices": datasets.features.Sequence(
+                    datasets.Value("string")
+                ),
+                "solution": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
+        if not os.path.exists(data_dir):
+            raise FileNotFoundError(
+                f"{data_dir} does not exist. Make sure you insert a manual dir via `datasets.load_dataset('matinf', data_dir=...)` that includes SAT-package-V3.txt. Manual download instructions: {self.manual_download_instructions}"
+            )
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "SAT-package-V3.txt"),
+                },
+            )
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath):
+        data = []
+        with open(filepath, "r", encoding="utf-8") as f:
+            record = []
+            for line in f:
+                line = line.strip()
+                if len(line) == 0 and record:
+                    data.append(record)
+                    record = []
+                elif len(line) > 0 and line[0] == '#':
+                    # Skip comments.
+                    continue
+                else:
+                    record.append(line)
+            data.append(record)
+        for key, record in enumerate(data):
+            source = record[-8]
+            stem = record[-7]
+            choices = record[-6:-1]
+            solution = record[-1]
+            yield key, {
+                'source': source,
+                'stem': stem,
+                'choices': choices,
+                'solution': solution,
+            }

lm_eval/datasets/triviaqa/dataset_infos.json

+{"triviaqa": {"description": "TriviaQA is a reading comprehension dataset containing over 650K question-answer-evidence\ntriples. TriviaQA includes 95K question-answer pairs authored by trivia enthusiasts\nand independently gathered evidence documents, six per question on average, that provide\nhigh quality distant supervision for answering the questions.\n", "citation": "@InProceedings{JoshiTriviaQA2017,\n    author = {Joshi, Mandar and Choi, Eunsol and Weld, Daniel S. and Zettlemoyer, Luke},\n    title = {TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension},\n    booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics},\n    month = {July},\n    year = {2017},\n    address = {Vancouver, Canada},\n    publisher = {Association for Computational Linguistics},\n}\n", "homepage": "https://nlp.cs.washington.edu/triviaqa/", "license": "", "features": {"question": {"dtype": "string", "id": null, "_type": "Value"}, "answer": {"aliases": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "values": {"dtype": "string", "id": null, "_type": "Value"}}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "trivia_qa", "config_name": "triviaqa", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 32846960, "num_examples": 87622, "dataset_name": "trivia_qa"}, "validation": {"name": "validation", "num_bytes": 4316214, "num_examples": 11313, "dataset_name": "trivia_qa"}}, "download_checksums": {"http://eaidata.bmk.sh/data/triviaqa-unfiltered.tar.gz": {"num_bytes": 546481381, "checksum": "adc19b42769062d241a8fbe834c56e58598d9322eb6c614e9f33a68a2cf5523e"}}, "download_size": 546481381, "post_processing_size": null, "dataset_size": 37163174, "size_in_bytes": 583644555}}
\ No newline at end of file

lm_eval/datasets/triviaqa/triviaqa.py

+# 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.
+#
+# Custom TriviaQA because HF version sanitizes the dataset differently.
+"""TriviaQA (Unfiltered Raw) dataset."""
+
+
+import json
+import os
+
+import datasets
+
+
+_CITATION = """\
+@InProceedings{JoshiTriviaQA2017,
+    author = {Joshi, Mandar and Choi, Eunsol and Weld, Daniel S. and Zettlemoyer, Luke},
+    title = {TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension},
+    booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics},
+    month = {July},
+    year = {2017},
+    address = {Vancouver, Canada},
+    publisher = {Association for Computational Linguistics},
+}
+"""
+
+_DESCRIPTION = """\
+TriviaQA is a reading comprehension dataset containing over 650K question-answer-evidence
+triples. TriviaQA includes 95K question-answer pairs authored by trivia enthusiasts
+and independently gathered evidence documents, six per question on average, that provide
+high quality distant supervision for answering the questions.
+"""
+
+_HOMEPAGE = "https://nlp.cs.washington.edu/triviaqa/"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_URLS = "http://eaidata.bmk.sh/data/triviaqa-unfiltered.tar.gz"
+
+
+class TriviaQa(datasets.GeneratorBasedBuilder):
+    """ TriviaQA is a reading comprehension dataset containing over 650K question-answer-evidence triples """
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(
+            name="triviaqa", version=VERSION, description="The TriviaQA dataset"),
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "question": datasets.Value("string"),
+                "answer": {
+                    "aliases":  datasets.features.Sequence(
+                        datasets.Value("string"),
+                    ),
+                    "value": datasets.Value("string")
+                }
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = _URLS
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "unfiltered-web-train.jsonl"),
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "unfiltered-web-dev.jsonl"),
+                    "split": "dev",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        with open(filepath, encoding="utf-8") as f:
+            for key, row in enumerate(f):
+                data = json.loads(row)
+                yield key, {
+                    "question": data["Question"],
+                    "answer": {
+                        "aliases": data["Answer"]["Aliases"],
+                        "value": data["Answer"]["Value"],
+                    }
+                }

lm_eval/datasets/truthfulqa/dataset_infos.json

+{"multiple_choice": {"description": "TruthfulQA is a benchmark to measure whether a language model is truthful in\ngenerating answers to questions. The benchmark comprises 817 questions that\nspan 38 categories, including health, law, finance and politics. Questions are\ncrafted so that some humans would answer falsely due to a false belief or\nmisconception. To perform well, models must avoid generating false answers\nlearned from imitating human texts.\n\nThe multiple choice TruthfulQA task", "citation": "@misc{lin2021truthfulqa,\n    title={TruthfulQA: Measuring How Models Mimic Human Falsehoods},\n    author={Stephanie Lin and Jacob Hilton and Owain Evans},\n    year={2021},\n    eprint={2109.07958},\n    archivePrefix={arXiv},\n    primaryClass={cs.CL}\n}\n", "homepage": "https://github.com/sylinrl/TruthfulQA", "license": "", "features": {"question": {"dtype": "string", "id": null, "_type": "Value"}, "mc1_targets": {"choices": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "labels": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "mc2_targets": {"choices": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "labels": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "truthfulqa", "config_name": "multiple_choice", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 610333, "num_examples": 817, "dataset_name": "truthfulqa"}}, "download_checksums": {"https://raw.githubusercontent.com/sylinrl/TruthfulQA/013686a06be7a7bde5bf8223943e106c7250123c/data/mc_task.json": {"num_bytes": 710607, "checksum": "6eb4125d25750c0145c4be2dce00440736684ab6f74ce6bff2139571cc758954"}}, "download_size": 710607, "post_processing_size": null, "dataset_size": 610333, "size_in_bytes": 1320940}, "generation": {"description": "TruthfulQA is a benchmark to measure whether a language model is truthful in\ngenerating answers to questions. The benchmark comprises 817 questions that\nspan 38 categories, including health, law, finance and politics. Questions are\ncrafted so that some humans would answer falsely due to a false belief or\nmisconception. To perform well, models must avoid generating false answers\nlearned from imitating human texts.\n\nThe generative TruthfulQA task", "citation": "@misc{lin2021truthfulqa,\n    title={TruthfulQA: Measuring How Models Mimic Human Falsehoods},\n    author={Stephanie Lin and Jacob Hilton and Owain Evans},\n    year={2021},\n    eprint={2109.07958},\n    archivePrefix={arXiv},\n    primaryClass={cs.CL}\n}\n", "homepage": "https://github.com/sylinrl/TruthfulQA", "license": "", "features": {"category": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "best_answer": {"dtype": "string", "id": null, "_type": "Value"}, "correct_answers": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "incorrect_answers": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "source": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "truthfulqa", "config_name": "generation", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 463860, "num_examples": 817, "dataset_name": "truthfulqa"}}, "download_checksums": {"https://raw.githubusercontent.com/sylinrl/TruthfulQA/013686a06be7a7bde5bf8223943e106c7250123c/TruthfulQA.csv": {"num_bytes": 443723, "checksum": "8d7dd15f033196140f032d97d30f037da7a7b1192c3f36f9937c1850925335a2"}}, "download_size": 443723, "post_processing_size": null, "dataset_size": 463860, "size_in_bytes": 907583}}
\ No newline at end of file

lm_eval/datasets/truthfulqa/truthfulqa.py

+# 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.
+"""TruthfulQA dataset."""
+
+
+import csv
+import json
+
+import datasets
+
+
+_CITATION = """\
+@misc{lin2021truthfulqa,
+    title={TruthfulQA: Measuring How Models Mimic Human Falsehoods},
+    author={Stephanie Lin and Jacob Hilton and Owain Evans},
+    year={2021},
+    eprint={2109.07958},
+    archivePrefix={arXiv},
+    primaryClass={cs.CL}
+}
+"""
+
+_DESCRIPTION = """\
+TruthfulQA is a benchmark to measure whether a language model is truthful in
+generating answers to questions. The benchmark comprises 817 questions that
+span 38 categories, including health, law, finance and politics. Questions are
+crafted so that some humans would answer falsely due to a false belief or
+misconception. To perform well, models must avoid generating false answers
+learned from imitating human texts.
+"""
+
+_HOMEPAGE = "https://github.com/sylinrl/TruthfulQA"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+
+class TruthfulqaConfig(datasets.BuilderConfig):
+    """BuilderConfig for TruthfulQA."""
+
+    def __init__(self, url, features, **kwargs):
+        """BuilderConfig for TruthfulQA.
+
+        Args:
+        url: *string*, the url to the specific subset of the GPT3 Arithmetic dataset.
+        features: *list[string]*, list of the features that will appear in the
+            feature dict.
+        """
+        # Version history:
+        super().__init__(version=datasets.Version("0.0.1"), **kwargs)
+        self.url = url
+        self.features = features
+
+
+class Truthfulqa(datasets.GeneratorBasedBuilder):
+    """TruthfulQA is a benchmark to measure whether a language model is truthful in
+generating answers to questions."""
+
+    BUILDER_CONFIGS = [
+        TruthfulqaConfig(
+            name="multiple_choice",
+            url="https://raw.githubusercontent.com/sylinrl/TruthfulQA/013686a06be7a7bde5bf8223943e106c7250123c/data/mc_task.json",
+            features=datasets.Features({
+                "question": datasets.Value("string"),
+                "mc1_targets": {
+                    "choices": datasets.features.Sequence(datasets.Value("string")),
+                    "labels": datasets.features.Sequence(datasets.Value("int32")),
+                },
+                "mc2_targets": {
+                    "choices": datasets.features.Sequence(datasets.Value("string")),
+                    "labels": datasets.features.Sequence(datasets.Value("int32")),
+                }
+            }),
+            description="The multiple choice TruthfulQA task"
+        ),
+        TruthfulqaConfig(
+            name="generation",
+            url="https://raw.githubusercontent.com/sylinrl/TruthfulQA/013686a06be7a7bde5bf8223943e106c7250123c/TruthfulQA.csv",
+            features=datasets.Features({
+                "category": datasets.Value("string"),
+                "question": datasets.Value("string"),
+                "best_answer": datasets.Value("string"),
+                "correct_answers": datasets.features.Sequence(datasets.Value("string")),
+                "incorrect_answers": datasets.features.Sequence(datasets.Value("string")),
+                "source": datasets.Value("string"),
+            }),
+            description="The generative TruthfulQA task"
+        )
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=f"{_DESCRIPTION}\n{self.config.description}",
+            features=self.config.features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = self.config.url
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir,
+                    "split": "validation",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        if self.config.name == "multiple_choice":
+            # Multiple choice data is in a `JSON` file.
+            with open(filepath, encoding="utf-8") as f:
+                contents = json.load(f)
+                for key, row in enumerate(contents):
+                    yield key, {
+                        "question": row["question"],
+                        "mc1_targets": {
+                            "choices": row["mc1_targets"].keys(),
+                            "labels": row["mc1_targets"].values(),
+                        },
+                        "mc2_targets": {
+                            "choices": row["mc2_targets"].keys(),
+                            "labels": row["mc2_targets"].values(),
+                        }
+                    }
+        else:
+            # Generation data is in a `CSV` file.
+            with open(filepath, newline='') as f:
+                contents = csv.DictReader(f)
+                for key, row in enumerate(contents):
+                    # Ensure that references exist.
+                    if not row['Correct Answers'] or not row['Incorrect Answers']:
+                        continue
+                    yield key, {
+                        "category": row["Category"],
+                        "question": row["Question"],
+                        "best_answer": row["Best Answer"],
+                        # split on ";"
+                        "correct_answers": row["Correct Answers"].strip().split(";"),
+                        "incorrect_answers": row["Incorrect Answers"].strip().split(";"),
+                        "source": row["Source"],
+                    }

lm_eval/datasets/unscramble/dataset_infos.json

+{"mid_word_1_anagrams": {"description": "Unscramble is a small battery of 5 \u201ccharacter manipulation\u201d tasks. Each task\ninvolves giving the model a word distorted by some combination of scrambling,\naddition, or deletion of characters, and asking it to recover the original word.\n", "citation": "@inproceedings{NEURIPS2020_1457c0d6,\n    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},\n    booktitle = {Advances in Neural Information Processing Systems},\n    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n    pages = {1877--1901},\n    publisher = {Curran Associates, Inc.},\n    title = {Language Models are Few-Shot Learners},\n    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},\n    volume = {33},\n    year = {2020}\n}\n", "homepage": "https://github.com/openai/gpt-3/tree/master/data", "license": "", "features": {"context": {"dtype": "string", "id": null, "_type": "Value"}, "completion": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "unscramble", "config_name": "mid_word_1_anagrams", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 271516, "num_examples": 10000, "dataset_name": "unscramble"}}, "download_checksums": {"https://raw.githubusercontent.com/openai/gpt-3/master/data/mid_word_1_anagrams.jsonl.gz": {"num_bytes": 106533, "checksum": "6768a86896083199de4815d4964cb2f6f1046476cfd80c2a562784f182905979"}}, "download_size": 106533, "post_processing_size": null, "dataset_size": 271516, "size_in_bytes": 378049}, "mid_word_2_anagrams": {"description": "Unscramble is a small battery of 5 \u201ccharacter manipulation\u201d tasks. Each task\ninvolves giving the model a word distorted by some combination of scrambling,\naddition, or deletion of characters, and asking it to recover the original word.\n", "citation": "@inproceedings{NEURIPS2020_1457c0d6,\n    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},\n    booktitle = {Advances in Neural Information Processing Systems},\n    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n    pages = {1877--1901},\n    publisher = {Curran Associates, Inc.},\n    title = {Language Models are Few-Shot Learners},\n    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},\n    volume = {33},\n    year = {2020}\n}\n", "homepage": "https://github.com/openai/gpt-3/tree/master/data", "license": "", "features": {"context": {"dtype": "string", "id": null, "_type": "Value"}, "completion": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "unscramble", "config_name": "mid_word_2_anagrams", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 282654, "num_examples": 10000, "dataset_name": "unscramble"}}, "download_checksums": {"https://raw.githubusercontent.com/openai/gpt-3/master/data/mid_word_2_anagrams.jsonl.gz": {"num_bytes": 109091, "checksum": "c3d839d09a7954b78a27cd2cd75d4ed0488656c56ef4dbd741a005343826cb01"}}, "download_size": 109091, "post_processing_size": null, "dataset_size": 282654, "size_in_bytes": 391745}, "cycle_letters_in_word": {"description": "Unscramble is a small battery of 5 \u201ccharacter manipulation\u201d tasks. Each task\ninvolves giving the model a word distorted by some combination of scrambling,\naddition, or deletion of characters, and asking it to recover the original word.\n", "citation": "@inproceedings{NEURIPS2020_1457c0d6,\n    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},\n    booktitle = {Advances in Neural Information Processing Systems},\n    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n    pages = {1877--1901},\n    publisher = {Curran Associates, Inc.},\n    title = {Language Models are Few-Shot Learners},\n    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},\n    volume = {33},\n    year = {2020}\n}\n", "homepage": "https://github.com/openai/gpt-3/tree/master/data", "license": "", "features": {"context": {"dtype": "string", "id": null, "_type": "Value"}, "completion": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "unscramble", "config_name": "cycle_letters_in_word", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 282654, "num_examples": 10000, "dataset_name": "unscramble"}}, "download_checksums": {"https://raw.githubusercontent.com/openai/gpt-3/master/data/cycle_letters_in_word.jsonl.gz": {"num_bytes": 98451, "checksum": "1689c9002bb8c5988bf5f05e977c9db92f57932c1b5a38998c29ac0dd71e1d42"}}, "download_size": 98451, "post_processing_size": null, "dataset_size": 282654, "size_in_bytes": 381105}, "random_insertion_in_word": {"description": "Unscramble is a small battery of 5 \u201ccharacter manipulation\u201d tasks. Each task\ninvolves giving the model a word distorted by some combination of scrambling,\naddition, or deletion of characters, and asking it to recover the original word.\n", "citation": "@inproceedings{NEURIPS2020_1457c0d6,\n    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},\n    booktitle = {Advances in Neural Information Processing Systems},\n    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n    pages = {1877--1901},\n    publisher = {Curran Associates, Inc.},\n    title = {Language Models are Few-Shot Learners},\n    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},\n    volume = {33},\n    year = {2020}\n}\n", "homepage": "https://github.com/openai/gpt-3/tree/master/data", "license": "", "features": {"context": {"dtype": "string", "id": null, "_type": "Value"}, "completion": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "unscramble", "config_name": "random_insertion_in_word", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 353981, "num_examples": 10000, "dataset_name": "unscramble"}}, "download_checksums": {"https://raw.githubusercontent.com/openai/gpt-3/master/data/random_insertion_in_word.jsonl.gz": {"num_bytes": 143626, "checksum": "72e65d83da53d15752ee0c47379509de149ddbad32d61184e5991df29616b78a"}}, "download_size": 143626, "post_processing_size": null, "dataset_size": 353981, "size_in_bytes": 497607}, "reversed_words": {"description": "Unscramble is a small battery of 5 \u201ccharacter manipulation\u201d tasks. Each task\ninvolves giving the model a word distorted by some combination of scrambling,\naddition, or deletion of characters, and asking it to recover the original word.\n", "citation": "@inproceedings{NEURIPS2020_1457c0d6,\n    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},\n    booktitle = {Advances in Neural Information Processing Systems},\n    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n    pages = {1877--1901},\n    publisher = {Curran Associates, Inc.},\n    title = {Language Models are Few-Shot Learners},\n    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},\n    volume = {33},\n    year = {2020}\n}\n", "homepage": "https://github.com/openai/gpt-3/tree/master/data", "license": "", "features": {"context": {"dtype": "string", "id": null, "_type": "Value"}, "completion": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "unscramble", "config_name": "reversed_words", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"validation": {"name": "validation", "num_bytes": 282654, "num_examples": 10000, "dataset_name": "unscramble"}}, "download_checksums": {"https://raw.githubusercontent.com/openai/gpt-3/master/data/reversed_words.jsonl.gz": {"num_bytes": 91917, "checksum": "133a08f875cd6c1ef8608a3233571a773881cc27b1c707de738cc6543439332a"}}, "download_size": 91917, "post_processing_size": null, "dataset_size": 282654, "size_in_bytes": 374571}}
\ No newline at end of file

lm_eval/datasets/unscramble/unscramble.py

+# 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.
+"""Unscramble dataset."""
+
+
+import json
+import os
+
+import datasets
+
+
+_CITATION = """\
+@inproceedings{NEURIPS2020_1457c0d6,
+    author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winter, Clemens and Hesse, Chris and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario},
+    booktitle = {Advances in Neural Information Processing Systems},
+    editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},
+    pages = {1877--1901},
+    publisher = {Curran Associates, Inc.},
+    title = {Language Models are Few-Shot Learners},
+    url = {https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf},
+    volume = {33},
+    year = {2020}
+}
+"""
+
+_DESCRIPTION = """\
+Unscramble is a small battery of 5 “character manipulation” tasks. Each task
+involves giving the model a word distorted by some combination of scrambling,
+addition, or deletion of characters, and asking it to recover the original word.
+"""
+
+_HOMEPAGE = "https://github.com/openai/gpt-3/tree/master/data"
+
+# TODO: Add the licence for the dataset here if you can find it
+_LICENSE = ""
+
+_BASE_URL = "https://raw.githubusercontent.com/openai/gpt-3/master/data"
+
+
+_DESCRIPTIONS = {
+    "mid_word_1_anagrams": "Anagrams of all but the first and last letter.",
+    "mid_word_2_anagrams": "Anagrams of all but the first and last 2 letters.",
+    "cycle_letters_in_word": "Cycle letters in the word.",
+    "random_insertion_in_word": "Random insertions in the word that must be removed.",
+    "reversed_words": "Words spelled backwards that must be reversed.",
+}
+_NAMES = _DESCRIPTIONS.keys()
+
+
+class Unscramble(datasets.GeneratorBasedBuilder):
+    """Unscramble is a small battery of 5 “character manipulation” tasks."""
+
+    VERSION = datasets.Version("0.0.1")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name=name, version=version,
+                               description=_DESCRIPTIONS[name])
+        for name, version in zip(_NAMES, [VERSION] * len(_NAMES))
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "context": datasets.Value("string"),
+                "completion": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = os.path.join(_BASE_URL, f"{self.config.name}.jsonl.gz")
+        data_dir = dl_manager.download_and_extract(urls)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": data_dir,
+                    "split": "validation",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def _generate_examples(self, filepath, split):
+        with open(filepath, encoding="utf-8") as f:
+            for key, row in enumerate(f):
+                data = json.loads(row)
+                yield key, {
+                    "context": data["context"],
+                    "completion": data["completion"],
+                }

lm_eval/datasets/wikitext/dataset_infos.json

+{"wikitext-103-v1": {"description": " The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified\n Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike\n License.\n", "citation": "@misc{merity2016pointer,\n      title={Pointer Sentinel Mixture Models},\n      author={Stephen Merity and Caiming Xiong and James Bradbury and Richard Socher},\n      year={2016},\n      eprint={1609.07843},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/", "license": "Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"page": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "wikitext", "config_name": "wikitext-103-v1", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"test": {"name": "test", "num_bytes": 1281262, "num_examples": 62, "dataset_name": "wikitext"}, "train": {"name": "train", "num_bytes": 539297488, "num_examples": 29444, "dataset_name": "wikitext"}, "validation": {"name": "validation", "num_bytes": 1142488, "num_examples": 60, "dataset_name": "wikitext"}}, "download_checksums": {"https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-103-v1.zip": {"num_bytes": 190229076, "checksum": "242ba0f20b329cfdf1ccc61e9e9e5b59becf189db7f7a81cd2a0e2fc31539590"}}, "download_size": 190229076, "post_processing_size": null, "dataset_size": 541721238, "size_in_bytes": 731950314}, "wikitext-2-v1": {"description": " The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified\n Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike\n License.\n", "citation": "@misc{merity2016pointer,\n      title={Pointer Sentinel Mixture Models},\n      author={Stephen Merity and Caiming Xiong and James Bradbury and Richard Socher},\n      year={2016},\n      eprint={1609.07843},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/", "license": "Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"page": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "wikitext", "config_name": "wikitext-2-v1", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"test": {"name": "test", "num_bytes": 1256634, "num_examples": 62, "dataset_name": "wikitext"}, "train": {"name": "train", "num_bytes": 10799034, "num_examples": 629, "dataset_name": "wikitext"}, "validation": {"name": "validation", "num_bytes": 1121860, "num_examples": 60, "dataset_name": "wikitext"}}, "download_checksums": {"https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-v1.zip": {"num_bytes": 4475746, "checksum": "92675f1d63015c1c8b51f1656a52d5bdbc33aafa60cc47a218a66e7ee817488c"}}, "download_size": 4475746, "post_processing_size": null, "dataset_size": 13177528, "size_in_bytes": 17653274}, "wikitext-103-raw-v1": {"description": " The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified\n Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike\n License.\n", "citation": "@misc{merity2016pointer,\n      title={Pointer Sentinel Mixture Models},\n      author={Stephen Merity and Caiming Xiong and James Bradbury and Richard Socher},\n      year={2016},\n      eprint={1609.07843},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/", "license": "Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"page": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "wikitext", "config_name": "wikitext-103-raw-v1", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"test": {"name": "test", "num_bytes": 1290775, "num_examples": 62, "dataset_name": "wikitext"}, "train": {"name": "train", "num_bytes": 540656522, "num_examples": 29444, "dataset_name": "wikitext"}, "validation": {"name": "validation", "num_bytes": 1147025, "num_examples": 60, "dataset_name": "wikitext"}}, "download_checksums": {"https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-103-raw-v1.zip": {"num_bytes": 191984949, "checksum": "91c00ae287f0d699e18605c84afc9e45c192bc6b7797ff8837e5474655a33794"}}, "download_size": 191984949, "post_processing_size": null, "dataset_size": 543094322, "size_in_bytes": 735079271}, "wikitext-2-raw-v1": {"description": " The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified\n Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike\n License.\n", "citation": "@misc{merity2016pointer,\n      title={Pointer Sentinel Mixture Models},\n      author={Stephen Merity and Caiming Xiong and James Bradbury and Richard Socher},\n      year={2016},\n      eprint={1609.07843},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/", "license": "Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"page": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "wikitext", "config_name": "wikitext-2-raw-v1", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"test": {"name": "test", "num_bytes": 1290775, "num_examples": 62, "dataset_name": "wikitext"}, "train": {"name": "train", "num_bytes": 10942633, "num_examples": 629, "dataset_name": "wikitext"}, "validation": {"name": "validation", "num_bytes": 1147025, "num_examples": 60, "dataset_name": "wikitext"}}, "download_checksums": {"https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip": {"num_bytes": 4721645, "checksum": "ef7edb566e3e2b2d31b29c1fdb0c89a4cc683597484c3dc2517919c615435a11"}}, "download_size": 4721645, "post_processing_size": null, "dataset_size": 13380433, "size_in_bytes": 18102078}}
\ No newline at end of file

lm_eval/datasets/wikitext/wikitext.py

+# 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.
+#
+# NOTE: This is a modified version of https://github.com/huggingface/datasets/blob/master/datasets/wikitext/wikitext.py
+# that returns Wiki pages instead of Wiki text line-by-line.
+"""WikiText Dataset."""
+
+
+import os
+
+import datasets
+
+
+_CITATION = """\
+@misc{merity2016pointer,
+      title={Pointer Sentinel Mixture Models},
+      author={Stephen Merity and Caiming Xiong and James Bradbury and Richard Socher},
+      year={2016},
+      eprint={1609.07843},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+"""
+
+_DESCRIPTION = """\
+ The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified
+ Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike
+ License.
+"""
+_HOMEPAGE = "https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/"
+_LICENSE = "Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)"
+_DATA_URL = "https://s3.amazonaws.com/research.metamind.io/wikitext"
+
+
+class WikitextConfig(datasets.BuilderConfig):
+    """BuilderConfig for GLUE."""
+
+    def __init__(self, data_url, **kwargs):
+        """BuilderConfig for Wikitext
+        Args:
+          data_url: `string`, url to the dataset (word or raw level)
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(WikitextConfig, self).__init__(
+            version=datasets.Version(
+                "1.0.0",
+            ),
+            **kwargs,
+        )
+        self.data_url = data_url
+
+
+class Wikitext(datasets.GeneratorBasedBuilder):
+    """TODO(wikitext_103): Short description of my dataset."""
+
+    # TODO(wikitext_103): Set up version.
+    VERSION = datasets.Version("0.1.0")
+    BUILDER_CONFIGS = [
+        WikitextConfig(
+            name="wikitext-103-v1",
+            data_url=_DATA_URL + "/" + "wikitext-103-v1.zip",
+            description="Word level dataset. No processing is needed other than replacing newlines with <eos> tokens.",
+        ),
+        WikitextConfig(
+            name="wikitext-2-v1",
+            data_url=_DATA_URL + "/" + "wikitext-2-v1.zip",
+            description="Word level dataset. No processing is needed other than replacing newlines with <eos> tokens.",
+        ),
+        WikitextConfig(
+            name="wikitext-103-raw-v1",
+            data_url=_DATA_URL + "/" + "wikitext-103-raw-v1.zip",
+            description="Raw level dataset: the raw tokens before the addition of <unk> tokens. "
+            "They should only be used for character level work or for creating newly derived datasets.",
+        ),
+        WikitextConfig(
+            name="wikitext-2-raw-v1",
+            data_url=_DATA_URL + "/" + "wikitext-2-raw-v1.zip",
+            description="Raw level dataset: the raw tokens before the addition of <unk> tokens. "
+            "They should only be used for character level work or for creating newly derived datasets.",
+        ),
+    ]
+
+    def _info(self):
+        # TODO(wikitext): Specifies the datasets.DatasetInfo object
+        return datasets.DatasetInfo(
+            # This is the description that will appear on the datasets page.
+            description=_DESCRIPTION,
+            # datasets.features.FeatureConnectors
+            features=datasets.Features(
+                {
+                    "page": datasets.Value("string")
+                    # These are the features of your dataset like images, labels ...
+                }
+            ),
+            # 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=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        # TODO(wikitext): Downloads the data and defines the splits
+        # dl_manager is a datasets.download.DownloadManager that can be used to
+        # download and extract URLs
+        if self.config.name == "wikitext-103-v1":
+            data_file = dl_manager.download_and_extract(self.config.data_url)
+            data_dir = os.path.join(data_file, "wikitext-103")
+            return [
+                datasets.SplitGenerator(
+                    name=datasets.Split.TEST,
+                    gen_kwargs={"data_file": os.path.join(
+                        data_dir, "wiki.test.tokens"), "split": "test"},
+                ),
+                datasets.SplitGenerator(
+                    name=datasets.Split.TRAIN,
+                    gen_kwargs={"data_file": os.path.join(
+                        data_dir, "wiki.train.tokens"), "split": "train"},
+                ),
+                datasets.SplitGenerator(
+                    name=datasets.Split.VALIDATION,
+                    gen_kwargs={"data_file": os.path.join(
+                        data_dir, "wiki.valid.tokens"), "split": "valid"},
+                ),
+            ]
+        else:
+            if self.config.name == "wikitext-103-raw-v1":
+                data_file = dl_manager.download_and_extract(
+                    self.config.data_url)
+                data_dir = os.path.join(data_file, "wikitext-103-raw")
+                return [
+                    datasets.SplitGenerator(
+                        name=datasets.Split.TEST,
+                        gen_kwargs={"data_file": os.path.join(
+                            data_dir, "wiki.test.raw"), "split": "test"},
+                    ),
+                    datasets.SplitGenerator(
+                        name=datasets.Split.TRAIN,
+                        gen_kwargs={"data_file": os.path.join(
+                            data_dir, "wiki.train.raw"), "split": "train"},
+                    ),
+                    datasets.SplitGenerator(
+                        name=datasets.Split.VALIDATION,
+                        gen_kwargs={"data_file": os.path.join(
+                            data_dir, "wiki.valid.raw"), "split": "valid"},
+                    ),
+                ]
+            else:
+                if self.config.name == "wikitext-2-raw-v1":
+                    data_file = dl_manager.download_and_extract(
+                        self.config.data_url)
+                    data_dir = os.path.join(data_file, "wikitext-2-raw")
+                    return [
+                        datasets.SplitGenerator(
+                            name=datasets.Split.TEST,
+                            gen_kwargs={"data_file": os.path.join(
+                                data_dir, "wiki.test.raw"), "split": "test"},
+                        ),
+                        datasets.SplitGenerator(
+                            name=datasets.Split.TRAIN,
+                            gen_kwargs={"data_file": os.path.join(
+                                data_dir, "wiki.train.raw"), "split": "train"},
+                        ),
+                        datasets.SplitGenerator(
+                            name=datasets.Split.VALIDATION,
+                            gen_kwargs={"data_file": os.path.join(
+                                data_dir, "wiki.valid.raw"), "split": "valid"},
+                        ),
+                    ]
+                else:
+                    if self.config.name == "wikitext-2-v1":
+                        data_file = dl_manager.download_and_extract(
+                            self.config.data_url)
+                        data_dir = os.path.join(data_file, "wikitext-2")
+                        return [
+                            datasets.SplitGenerator(
+                                name=datasets.Split.TEST,
+                                gen_kwargs={"data_file": os.path.join(
+                                    data_dir, "wiki.test.tokens"), "split": "test"},
+                            ),
+                            datasets.SplitGenerator(
+                                name=datasets.Split.TRAIN,
+                                gen_kwargs={
+                                    "data_file": os.path.join(data_dir, "wiki.train.tokens"),
+                                    "split": "train",
+                                },
+                            ),
+                            datasets.SplitGenerator(
+                                name=datasets.Split.VALIDATION,
+                                gen_kwargs={
+                                    "data_file": os.path.join(data_dir, "wiki.valid.tokens"),
+                                    "split": "valid",
+                                },
+                            ),
+                        ]
+
+    def _generate_examples(self, data_file, split):
+        """Yields examples."""
+        with open(data_file, encoding="utf-8") as f:
+            key = 0
+            ret = []
+            data = f.read().split("\n")
+            for line in data:
+                rline = line.replace("= = =", "===").replace("= =", "==").strip()
+                if rline.startswith('= ') and rline.strip().endswith(' ='):
+                    page = '\n'.join(ret)
+                    if page.strip():
+                        yield key, {"page": page}
+                        key += 1
+                    ret = []
+                ret.append(line)
+            page = '\n'.join(ret)
+            yield key, {"page": page}

lm_eval/metrics.py

@@ -3,7 +3,7 @@ from collections.abc import Iterable
 
 import numpy as np
 import sacrebleu
-import sklearn
+import sklearn.metrics
 import random
 
 
@@ -245,3 +245,10 @@ def stderr_for_metric(metric, bootstrap_iters):
     }
 
     return stderr.get(metric, None)
+
+
+def yesno(x):
+    if x:
+        return 'yes'
+    else:
+        return 'no'

lm_eval/models/gpt2.py

@@ -27,16 +27,16 @@ class HFLM(BaseLM):
         self.tokenizer = transformers.AutoTokenizer.from_pretrained(
             pretrained if tokenizer is None else tokenizer, revision=revision, subfolder=subfolder)
 
-        assert isinstance(self.tokenizer, (
-            transformers.GPT2Tokenizer, transformers.GPT2TokenizerFast,
-            transformers.T5Tokenizer, transformers.T5TokenizerFast,
-        )), "this tokenizer has not been checked for compatibility yet!"
+        # assert isinstance(self.tokenizer, (
+        #     transformers.GPT2Tokenizer, transformers.GPT2TokenizerFast,
+        #     transformers.T5Tokenizer, transformers.T5TokenizerFast,
+        # )), "this tokenizer has not been checked for compatibility yet!"
 
         self.vocab_size = self.tokenizer.vocab_size
 
-        if isinstance(self.tokenizer, (transformers.GPT2Tokenizer, transformers.GPT2TokenizerFast)):
-            assert self.tokenizer.encode('hello\n\nhello') == [31373, 198, 198, 31373], \
-                self.tokenizer.encode('hello\n\nhello')
+        # if isinstance(self.tokenizer, (transformers.GPT2Tokenizer, transformers.GPT2TokenizerFast)):
+        #     assert self.tokenizer.encode('hello\n\nhello') == [31373, 198, 198, 31373], \
+        #         self.tokenizer.encode('hello\n\nhello')
 
         # multithreading and batching
         self.batch_size_per_gpu = batch_size  # todo: adaptive batch size