Add Chunked datasets, py hf loading script, ipynb for data processing

mathpile_arxiv_subset_tiny.py

@@ -0,0 +1,126 @@
+# 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.
+"""This dataset consists of a toy subset of 8834 (5000 training + 3834 testing) TeX files found in the arXiv subset of MathPile, used for testing. Each document is split using LaTeX-specific characters for recursive character text splitting with ~4k token window and ~1.5k token overlaps. You should not use this dataset. Training and testing sets are already split."""
+
+import json
+import os
+
+import datasets
+
+_CITATION = """\
+@article{wang2023mathpile,
+  title={Generative AI for Math: Part I -- MathPile: A Billion-Token-Scale Pretraining Corpus for Math},
+  author={Wang, Zengzhi and Xia, Rui and Liu, Pengfei},
+  journal={arXiv preprint arXiv:2312.17120},
+  year={2023}
+}
+"""
+
+_DESCRIPTION = """\
+This dataset consists of a toy subset of 8834 (5000 training + 3834 testing) TeX files found in the arXiv subset of MathPile, used for testing. Each document is split using LaTeX-specific characters for recursive character text splitting with ~4k token window and ~1.5k token overlaps. You should not use this dataset.
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/aluncstokes/mathpile_arxiv_subset_tiny"
+
+
+_LICENSE = "CC BY-NC-SA 4.0"
+
+# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
+# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
+_URLS = {"first_domain": "https://huggingface.co/datasets/GAIR/MathPile"}
+
+
+class MathpileArxivSubsetTiny(datasets.GeneratorBasedBuilder):
+    """This dataset consists of a toy subset of 8834 (5000 training + 3834 testing) TeX files found in the arXiv subset of MathPile, used for testing. Each document is split using LaTeX-specific characters for recursive character text splitting with ~4k token window and ~1.5k token overlaps. You should not use this dataset"""
+
+    VERSION = datasets.Version("0.2")
+
+    # This is an example of a dataset with multiple configurations.
+    # If you don't want/need to define several sub-sets in your dataset,
+    # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
+
+    # If you need to make complex sub-parts in the datasets with configurable options
+    # You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
+    # BUILDER_CONFIG_CLASS = MyBuilderConfig
+
+    # You will be able to load one or the other configurations in the following list with
+    # data = datasets.load_dataset('my_dataset', 'first_domain')
+    # data = datasets.load_dataset('my_dataset', 'second_domain')
+
+    DEFAULT_CONFIG_NAME = ""  # It's not mandatory to have a default configuration. Just use one if it make sense.
+
+    def _info(self):
+        if self.config.name == "":
+            features = datasets.Features(
+                {
+                    "set": datasets.Value("string"),
+                    "id": datasets.Value("string"),
+                    "chunk_text": datasets.Value("long_string"),
+                    "chunk_num_tokens": datasets.Value("uint32"),
+                    "document_num_tokens": datasets.Value("uint32"),
+                    "document_language": datasets.Value("string"),
+                }
+            )
+
+        return datasets.DatasetInfo(
+            # This is the description that will appear on the datasets page.
+            description=_DESCRIPTION,
+            # This defines the different columns of the dataset and their types
+            features=features,  # Here we define them above because they are different between the two configurations
+            # If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
+            # specify them. They'll be used if as_supervised=True in builder.as_dataset.
+            # supervised_keys=("sentence", "label"),
+            # Homepage of the dataset for documentation
+            homepage=_HOMEPAGE,
+            # License for the dataset if available
+            license=_LICENSE,
+            # Citation for the dataset
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        # If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name
+
+        # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
+        # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
+        # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
+        urls = _URLS[self.config.name]
+        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, "train_chunked.jsonl"),
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "test_chunked.jsonl"),
+                    "split": "test",
+                },
+            ),
+        ]
+
+    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
+    def generate_examples(self, filepath, split):
+        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
+        with open(filepath, encoding="utf-8") as f:
+            for key, row in enumerate(f):
+                data = json.loads(row)
+                # Yields examples as (key, example) tuples
+                yield key, {"text": data["chunk_text"]}

process_documents.ipynb

@@ -0,0 +1,267 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Note: you may need to restart the kernel to use updated packages.\n",
+      "Note: you may need to restart the kernel to use updated packages.\n",
+      "Note: you may need to restart the kernel to use updated packages.\n",
+      "Note: you may need to restart the kernel to use updated packages.\n",
+      "Note: you may need to restart the kernel to use updated packages.\n"
+     ]
+    }
+   ],
+   "source": [
+    "%pip install -q -U tiktoken\n",
+    "%pip install -q -U langchain\n",
+    "%pip install -q -U langdetect\n",
+    "%pip install -q -U git+https://github.com/huggingface/transformers\n",
+    "%pip install -q -U sentencepiece\n",
+    "%pip install -q -U protobuf\n",
+    "%pip install -q -U tqdm"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import json\n",
+    "import warnings\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "# import tiktoken\n",
+    "from langchain.text_splitter import RecursiveCharacterTextSplitter\n",
+    "from langdetect import detect, detect_langs\n",
+    "from transformers import AutoTokenizer\n",
+    "\n",
+    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"True\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "tokeniser = AutoTokenizer.from_pretrained(\n",
+    "    \"mistralai/Mistral-7B-v0.1\", use_fast=True)\n",
+    "\n",
+    "\n",
+    "def string_token_length(text):\n",
+    "    return len(tokeniser(text, add_special_tokens=False).input_ids)\n",
+    "\n",
+    "\n",
+    "def write_jsonl(data, filename):\n",
+    "    with open(filename, \"w\") as f:\n",
+    "        for entry in data:\n",
+    "            f.write(json.dumps(entry) + \"\\n\")\n",
+    "\n",
+    "\n",
+    "def chunkify_tex(tex_text, chunk_size=4086, chunk_overlap=1536):\n",
+    "    splitter = RecursiveCharacterTextSplitter(\n",
+    "        [\n",
+    "            r\"(?<=\\n)(?=\\\\section{)\",\n",
+    "            r\"(?<=\\n)(?=\\\\subsection{)\",\n",
+    "            r\"(?<=\\n)(?=\\\\subsubsection{)\",\n",
+    "            r\"(?<=\\\\end{proof})\",\n",
+    "            r\"(?<=\\\\qed)\",\n",
+    "            r\"\\n\\n\\n\",\n",
+    "            r\"\\n\\n\",\n",
+    "        ],\n",
+    "        keep_separator=True,\n",
+    "        is_separator_regex=True,\n",
+    "        chunk_size=chunk_size,\n",
+    "        chunk_overlap=chunk_overlap,\n",
+    "        length_function=string_token_length,\n",
+    "    )\n",
+    "\n",
+    "    bits = splitter.split_text(tex_text)\n",
+    "\n",
+    "    return bits"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Metadata object\n",
+    "# {\n",
+    "#   \"set\": str,\n",
+    "#   \"id\": int,\n",
+    "#   \"char_len\": int,\n",
+    "#   \"tok_len\": int,\n",
+    "#   \"lang\": str,\n",
+    "#   \"text\": str\n",
+    "#   \"num_chunks\": int\n",
+    "#   \"chunk_lengths\": List[int]\n",
+    "#   \"chunks\": List[str]\n",
+    "# }\n",
+    "\n",
+    "with open(\"mathpile_arxiv_subset_tiny/train.jsonl\", \"r\") as f:\n",
+    "    train = [json.loads(l) for l in f.readlines()]\n",
+    "\n",
+    "with open(\"mathpile_arxiv_subset_tiny/test.jsonl\", \"r\") as f:\n",
+    "    test = [json.loads(l) for l in f.readlines()]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "8438it [02:30, 56.19it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "data = [\n",
+    "    {\n",
+    "        \"set\": \"train\",\n",
+    "        \"id\": f'0.{j}',\n",
+    "        \"char_len\": len(entry[\"text\"]),\n",
+    "        # \"tok_len\": string_token_length(entry[\"text\"]),\n",
+    "        \"lang\": detect(entry[\"text\"]),\n",
+    "        \"text\": entry[\"text\"],\n",
+    "    }\n",
+    "    for j, entry in tqdm(enumerate(train), total=len(train))\n",
+    "] + [\n",
+    "    {\n",
+    "        \"set\": \"test\",\n",
+    "        \"id\": f'1.{j}',\n",
+    "        \"char_len\": len(entry[\"text\"]),\n",
+    "        # \"tok_len\": string_token_length(entry[\"text\"]),\n",
+    "        \"lang\": detect(entry[\"text\"]),\n",
+    "        \"text\": entry[\"text\"],\n",
+    "    }\n",
+    "    for j, entry in tqdm(enumerate(test), total=len(test))\n",
+    "]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "100%|██████████| 8438/8438 [15:17<00:00,  9.20it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "for j, datum in tqdm(enumerate(data), total=len(data)):\n",
+    "    text = datum[\"text\"]\n",
+    "    chunks = chunkify_tex(text, chunk_size=4095, chunk_overlap=1536)\n",
+    "    chunk_lengths = [string_token_length(chunk) for chunk in chunks]\n",
+    "    data[j][\"num_chunks\"] = len(chunks)\n",
+    "    data[j][\"chunk_lengths\"] = chunk_lengths\n",
+    "    data[j][\"chunks\"] = chunks\n",
+    "    data[j][\"tok_len\"] = sum(chunk_lengths)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "data_train = [datum for datum in data if datum[\"set\"] == \"train\"]\n",
+    "data_test = [datum for datum in data if datum[\"set\"] == \"test\"]\n",
+    "\n",
+    "chunked_train = []\n",
+    "chunked_test = []\n",
+    "\n",
+    "for datum in data_train:\n",
+    "    total_document_tokens = datum[\"tok_len\"]\n",
+    "    document_id = datum[\"id\"]\n",
+    "    for j in range(len(datum[\"chunks\"])):\n",
+    "        chunk_id = f\"{document_id}.{j}\"\n",
+    "        chunk_text = datum[\"chunks\"][j]\n",
+    "        chunk_num_tokens = datum[\"chunk_lengths\"][j]\n",
+    "        chunked_train.append(\n",
+    "            {\n",
+    "                \"set\": \"train\",\n",
+    "                \"id\": chunk_id,\n",
+    "                \"chunk_text\": chunk_text,\n",
+    "                \"chunk_num_tokens\": chunk_num_tokens,\n",
+    "                \"document_num_tokens\": total_document_tokens,\n",
+    "                \"document_language\": datum[\"lang\"],\n",
+    "            }\n",
+    "        )\n",
+    "\n",
+    "for datum in data_test:\n",
+    "    total_document_tokens = datum[\"tok_len\"]\n",
+    "    document_id = datum[\"id\"]\n",
+    "    for j in range(len(datum[\"chunks\"])):\n",
+    "        chunk_id = f\"{document_id}.{j}\"\n",
+    "        chunk_text = datum[\"chunks\"][j]\n",
+    "        chunk_num_tokens = datum[\"chunk_lengths\"][j]\n",
+    "        chunked_test.append(\n",
+    "            {\n",
+    "                \"set\": \"test\",\n",
+    "                \"id\": chunk_id,\n",
+    "                \"chunk_text\": chunk_text,\n",
+    "                \"chunk_num_tokens\": chunk_num_tokens,\n",
+    "                \"document_num_tokens\": total_document_tokens,\n",
+    "                \"document_language\": datum[\"lang\"],\n",
+    "            }\n",
+    "        )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "write_jsonl(chunked_train, \"mathpile_arxiv_subset_tiny/train_chunked.jsonl\")\n",
+    "write_jsonl(chunked_test, \"mathpile_arxiv_subset_tiny/test_chunked.jsonl\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

test_chunked.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffde9da9a708bda09af6eed842237c33e137f28b6ae14f40dcf55da48e7c1864
+size 285891904

train_chunked.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19d4f299e95edba1ea68f395da36d4efc92c5403a4116ed4f88c55e2c6407642
+size 412187046