patrickjmcbride/math-instruct-binned

---
dataset_info:
features:
- name: instruction
dtype: string
- name: context
dtype: string
- name: output
dtype: string
- name: text
dtype: string
splits:
- name: small
num_bytes: 1864543974
num_examples: 231315
- name: medium
num_bytes: 5961505276
num_examples: 584345
- name: large
num_bytes: 2026722938
num_examples: 154320
download_size: 4895917273
dataset_size: 9852772188
configs:
- config_name: default
data_files:
- split: small
path: data/small-*
- split: medium
path: data/medium-*
- split: large
path: data/large-*
tags:
- math
- mathematics
- probability
- statistics
- algebra
- liner algebra
license: apache-2.0
task_categories:
- text-generation
size_categories:
- 100K<n<1M
---
This is a pre-binned instruction formatted version of Feynman Innovations' [Maths-College](https://huggingface.co/datasets/ajibawa-2023/Maths-College) dataset.

Credit to [Feynman Innovations](https://huggingface.co/ajibawa-2023) for the base dataset.

It is formatted to be ready for fine-tuning an instruct model.

The splits are based on the size of the full instruction ('text') after being tokenized with the Llama-3-8B-Instruct tokenizer (based on tiktoken).

A non-split version is avalible as [math-instruct-dataset](https://huggingface.co/datasets/patrickjmcbride/math-instruct-dataset)

Binned by length of tokenized 'text' field
- small: [min-1024)
- medium: [1024-1536)
- large: [1536, max]

# Data Fields

The data fields are as follows:

instruction: describes the task that the model needs to perform. (all instructions are the same "Write an educational piece related to the following text snippet:")
context: additional context containing the math concept to explain
output: an in depth explanation of the concept from the context
text: the instruction, context and output formatted with a prompt template to be used for fine-tuning.

# Splits

small: 231,315 (24%)
medium: 584,345 (60%)
large: 154,320 (16%)

原始信息汇总

数据集概述

数据特征

• instruction: 描述模型需要执行的任务（所有指令相同：“写一篇与以下文本片段相关的教育文章：”）
• context: 包含需要解释的数学概念的附加上下文
• output: 对上下文中概念的深入解释
• text: 用于微调的提示模板格式的指令、上下文和输出

数据分割

• small: 包含231,315个样本，占总样本的24%
• medium: 包含584,345个样本，占总样本的60%
• large: 包含154,320个样本，占总样本的16%

数据集大小

• 下载大小: 4,895,917,273字节
• 数据集大小: 9,852,772,188字节

配置

• default:
  • small: 数据路径为data/small-*
  • medium: 数据路径为data/medium-*
  • large: 数据路径为data/large-*

标签

• math
• mathematics
• probability
• statistics
• algebra
• liner algebra

许可

• apache-2.0

任务类别

• text-generation

大小类别

• 100K<n<1M