shujatoor/ner_instruct-chat

---
dataset_info:
features:
- name: Instruction
list:
- name: content
dtype: string
- name: role
dtype: string
splits:
- name: train
num_bytes: 10301286
num_examples: 16373
download_size: 1553901
dataset_size: 10301286
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Instruction Dataset to Finetune LLM for Named Entity Recognition (NER)- Chat Format

- This instruction dataset can be used to fine tune your model for the purpose of performing Named Entity Recognition (NER)
- The dataset is already formatted as text with system prompt, user instruction and assistant message.
- This dataset contains 16.4k instruction examples.
- The dataset was created using a little over 2000 receipts data of which some 1500 are original and 500
are fake receipts.
- The Original receipts were downloaded from online opensource and some were collected by myself.
- The Original receipts are mainly from malaysia, India, US, UK, Canada and Pakistan
- Paddleocr was used to perform OCR for the original receipts.
- The fake receipts were generated using llama2-70b-4096. These receipts are related to grocery, pet shop,
hardware store, furniture, art store, bakery, restaurant, hunting store, ice cream parlor, clothing,
shoes, pharmacy etc.

## Summary:
- Original Receipts Used ~ 1500
- Fake Receipts Data Used ~ 500
- Fake Receipts data generated using: llama2-70b-4096
- Library used to perform OCR: Paddleocr
- Dataset created using prompt to llama2-70b-4096
- Dataset instruction examples: 16.4k
- Format: Chat-template

原始信息汇总

数据集概述

数据集基本信息

• 名称: Instruction Dataset to Finetune LLM for Named Entity Recognition (NER)- Chat Format
• 目的: 用于微调模型以进行命名实体识别（NER）
• 格式: 文本格式，包含系统提示、用户指令和助手消息
• 示例数量: 16,400个指令示例

数据集内容

• 原始收据: 约1,500份，主要来自马来西亚、印度、美国、英国、加拿大和巴基斯坦
• 伪造收据: 约500份，使用llama2-70b-4096生成，涉及多种商业类型

数据处理工具

• OCR工具: Paddleocr
• 伪造收据生成工具: llama2-70b-4096

数据集结构

• 特征:
  • Instruction:
    • content: 字符串类型
    • role: 字符串类型
• 分割:
  • train:
    • 示例数量: 16,373
    • 数据大小: 10,301,286字节
    • 下载大小: 1,553,901字节

数据集配置

• 默认配置:
  • 数据文件:
    • 分割: train
    • 路径: data/train-*