Deita

Visit Tool

Deita is an open-source coding & development tool that facilitates data-efficient instruction tuning for aligning large language models. It provides toolkits for automatic data selection and offers lightweight, high-quality alignment SFT data.

Claim this tool

2Views

At a glance

Pricing

Open Source

Free tier

Yes

API

Skill level

Technical

About

What is deita?

Deita is an open-sourced project designed to facilitate Automatic Data Selection for instruction tuning in Large Language Models (LLMs). It provides open-sourced toolkits for automatic data selection, along with Deita Datasets, which are extremely lightweight and high-quality alignment SFT data available in 6k-sized and 10k-sized versions. The project also offers Deita Models, which are powerful models capable of achieving performance on par with state-of-the-art chat LLMs with significantly less instruction tuning data. Deita includes pipelines for dataset scoring, embedding generation, and score-first, diversity-aware selection, making it a comprehensive resource for researchers and developers working on LLM alignment.

Best used for

Ideal for developers and data scientists who need to efficiently select high-quality data for instruction tuning, train large language models with significantly less data, and evaluate the complexity and quality of datasets. Especially valuable for optimizing LLM alignment processes and achieving SOTA performance with data efficiency.

Common actions

select data

tune language models

evaluate data quality

generate embeddings

train LLMs

github copilot"AI Agents"face swappingcollaborationdeepfakelow-code/no-codeworkflowsautomated workflowopen-source

Capabilities

Key features

Automatic data selection toolkits
Lightweight alignment SFT datasets
Data complexity/quality scorers
Dataset scoring pipelines
Embedding generation pipelines
Diversity-aware data selection
Data-efficient instruction tuning

Target Audience

developerdata scientist

Integrations

Not yet documented

Pricing & Plans

Open Source

Free

FAQs

What kind of data does Deita provide?

Deita provides a series of extremely lightweight, high-quality alignment SFT (Supervised Fine-Tuning) datasets. These include 6k-sized and 10k-sized datasets, along with scorer datasets for complexity and quality assessment.

How does Deita help with instruction tuning?

Deita offers open-sourced toolkits for automatic data selection, allowing users to choose high-quality data subsets for instruction tuning. This enables the training of powerful LLMs with significantly less data compared to other state-of-the-art models.

Can Deita models be used with different base LLMs?

Yes, Deita models are available based on various base models, including Mistral-7B, LLaMA-1-13B, and LLaMA-2-13B. This provides flexibility for integration into different LLM ecosystems and research setups.

What are the key components of the Deita project?

The Deita project includes open-sourced toolkits for automatic data selection, Deita Datasets for alignment SFT, and Deita Models that demonstrate efficient instruction tuning. It also provides pipelines for data scoring, embedding generation, and diversity-aware selection.

Trending

Subcategories trending in Coding & Development

Code Assistants DevOps & Infrastructure No-Code / Low-Code Testing & QA Backend & APIs Prompt Engineering

Trending

Explore

Browse AI tools by category

Content & Design Productivity & Business Coding & Development AI Agents & Automation Research & Education Wellness & Lifestyle Career Development Marketing & Growth Data & Analytics Customer Support & CX Finance E-commerce