OpenHermes 2.5
Instruction Dataset
A million-level instruction fine-tuning dataset curated by Teknium, aggregating multiple open-source data sources, covering GPT-4 generation, code instructions, and multi-domain corpora, licensed under Apache 2.0, widely used for training open-source large models.
Dataset Highlights
A large-scale, high-quality instruction fine-tuning dataset designed for training open-source models
Millions of Instructions
Contains over 1 million carefully selected instruction records, providing ample training data to support comprehensive fine-tuning and alignment of large-scale language models.
Multi-source Data Aggregation
Carefully selected and aggregated from multiple high-quality open-source datasets, blending instructions from different styles and domains to ensure data diversity and complementarity.
GPT-4 Quality Data
Includes a large number of high-quality instruction-response pairs generated by GPT-4, providing training samples close to state-of-the-art model standards, significantly improving the output quality of fine-tuned models.
Code Instruction Coverage
Built-in rich instructions for code generation, code explanation, and programming Q&A, ensuring that the fine-tuned model performs excellently on coding tasks as well.
Domain Diversity
Covers multiple domains including general Q&A, creative writing, mathematical reasoning, role-playing, and specialized knowledge, training a versatile general-purpose large model.
Apache 2.0 License
Utilizes a permissive Apache 2.0 open-source license, supporting commercial use and redistribution without concerns about licensing restrictions.
Applicable Scenarios
Covering the entire process of instruction fine-tuning from model training to academic research
General Fine-tuning Training
As a general instruction fine-tuning dataset, train large language models with multi-turn dialogue, knowledge Q&A, and reasoning capabilities
Code + Dialogue Training
Utilize a combination of code instructions and dialogue data to train full-stack models with programming capabilities and natural language understanding
Open-source Model Development
Provide high-quality fine-tuning data for open-source base models like LLaMA and Mistral, quickly building deployable chat models
Instruction Quality Research
Study the impact of instruction data from different sources and qualities on model performance, exploring data mixing strategies and best practices
Quick API Call
Quickly obtain the OpenHermes 2.5 dataset through the Ace Data Cloud API
curl -X GET "https://api.acedata.cloud/datasets/openhermes-2-5" \ -H "Authorization: Bearer YOUR_API_TOKEN" \ -H "Content-Type: application/json"
import requests
url = "https://api.acedata.cloud/datasets/openhermes-2-5"
headers = {
"Authorization": "Bearer YOUR_API_TOKEN",
"Content-Type": "application/json"
}
response = requests.get(url, headers=headers)
data = response.json()
print(data)
Data Preview
The following is an example record structure of the OpenHermes 2.5 dataset
{
"conversations": [
{
"from": "system",
"value": "You are a helpful assistant."
},
{
"from": "human",
"value": "Explain the concept of recursion in programming."
},
{
"from": "gpt",
"value": "Recursion is a programming technique where a function calls itself to solve a problem by breaking it down into smaller subproblems. Each recursive call works on a simpler version of the original problem until it reaches a base case — a condition that stops the recursion. For example, calculating factorial: factorial(n) = n * factorial(n-1), with base case factorial(0) = 1."
}
],
"source": "gpt4_alpaca",
"category": "programming"
}
3 Steps to Get Started Quickly
From obtaining data to starting training, you can launch your fine-tuning project in just a few minutes
Browse the Dataset
View the details of the OpenHermes 2.5 dataset on the Ace Data Cloud platform, understand the data structure, source distribution, and licensing agreement.
Get API Access
Register and obtain an API Token, query and download data on demand through the RESTful API, supporting pagination and conditional filtering.
Start Fine-Tuning Training
Load the data into the training framework, use LoRA, QLoRA, or full fine-tuning methods to train your own instruction-optimized model.
Start Using the OpenHermes 2.5 Dataset
Millions of instruction data, Apache 2.0 license, get it now. Whether you are an open-source model developer or an NLP researcher, this dataset is an ideal starting point for fine-tuning.