This repository provides a curated, community-driven overview of Large Language Models (LLMs) relevant to software development. It aims to help developers navigate the rapidly evolving DevAI landscape by categorizing and detailing popular open-source and commercial models, aiding in the selection process based on factors like cost, performance, and deployment needs.

How It Works

The project categorizes LLMs into open-source and commercial options, presenting them with key details such as creator, release date, license, base model, and parameter count. For open-source models, it further distinguishes between local deployment and hosted provider usage, outlining the trade-offs. Commercial models are discussed in terms of API access, cost, and performance, offering a comparative perspective for developers.

Quick Start & Requirements

This repository is a reference guide and does not require installation or execution. It provides links to external resources for LLM deployment and usage.

Highlighted Details

• Detailed comparison of popular open-source coding LLMs like Code Llama, WizardCoder, Phind-CodeLlama, Mistral, StarCoder, DeepSeek Coder, and Llama 2.
• Overview of leading commercial coding LLMs including GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, Claude 2, and PaLM 2.
• Guidance on choosing between open-source and commercial models based on privacy, cost, and performance requirements.
• Information on deploying open-source LLMs for teams.

Maintenance & Community

The project encourages community contributions via pull requests and issues to maintain an up-to-date index. A monthly newsletter is available for updates on DevAI.

Licensing & Compatibility

Open-source models listed have varying licenses, including Llama 2 Community, Apache 2.0, and OpenRAIL-M. Commercial models are accessed via proprietary APIs.

Limitations & Caveats

The information reflects the state of LLMs as of October/November 2023 and may not include the latest models or performance benchmarks. Some commercial models require application or have limited preview access.