Dolphin3.0 Qwen2.5 1.5 B

Introduction

Dolphin 3.0 Qwen 2.5 1.5B is an advanced, instruct-tuned AI model designed for general-purpose applications, including coding, mathematics, function calling, and more. It aims to provide an alternative to other AI models like ChatGPT by offering more control and customization to the user.

Architecture

The model is part of the Dolphin 3.0 Collection and is developed by Cognitive Computations with contributions from Eric Hartford, Ben Gitter, and BlouseJury. It is designed to be highly flexible and adaptable, allowing users to set their system prompts and alignments according to their specific needs.

Training

Dolphin 3.0 utilizes a diverse range of datasets for training, including OpenCoder-LLM stages, Microsoft's orca datasets, and others from NousResearch and AI-MO. The training infrastructure was supported by sponsors such as Crusoe Cloud and Akash, providing cloud GPUs like L40s and H100s. The model benefits from the contributions of organizations like Meta, Qwen, and OpenCoder, which provided foundational research and models.

Guide: Running Locally

To run Dolphin 3.0 locally, follow these steps:

1. Install Dependencies: Use package managers like pip to install necessary libraries, such as the Hugging Face Transformers library.
2. Download the Model: Access the model via Hugging Face's model hub.
3. Set Up Environment: Ensure your system has a compatible Python environment.
4. Run the Model: Use provided scripts or APIs to integrate and run the model.

For optimal performance, consider using cloud GPU services like those from AWS, Google Cloud, or Azure.

License

Dolphin 3.0 is released under the Apache 2.0 license, allowing for broad use and modification with proper attribution. For more details, refer to the license link.