Dolphin3.0 Llama3.1 8 B G G U F

Introduction

Dolphin 3.0 is a cutting-edge model in the Dolphin series, designed for broad utility in areas like coding, math, and general AI tasks. It is aimed at giving users full control over the system prompt, versioning, and data alignment, contrasting with other models that may impose restrictions or alter features unpredictably.

Architecture

The Dolphin 3.0 model is based on the Llama 3.1 architecture with an 8-billion parameter scale. It leverages a variety of datasets for training, including those focused on coding, math, and other functional tasks. This architecture is designed to offer flexibility and control for the system owner.

Training

The model was trained using a range of datasets such as OpenCoder-LLM and Microsoft's orca datasets. Sponsorships from companies like Crusoe Cloud and Akash provided the necessary GPU resources, including NVIDIA H100s, to facilitate the training and evaluation processes. The training process was supported by various open-source datasets and instrumental papers from Meta, Qwen, and OpenCoder.

Guide: Running Locally

To run Dolphin 3.0 locally, you can use one of several methods:

1. Use Ollama:

   • Install Ollama from Ollama's website.
   • Run the model using the command: ollama run hf.co/cognitivecomputations/Dolphin3.0-Llama3.1-8B-GGUF:Q4_0.
   • Configure the system prompt using /set system <your system prompt>.

2. Other Options:

   • Consider using the Hugging Face Transformers library, LM Studio, or other compatible platforms.

For enhanced performance, consider utilizing cloud GPUs such as NVIDIA A100 or H100 for running the model.

License

Dolphin 3.0 is licensed under the llama3.1 license, providing users with the flexibility and control over how they use and align the model to their specific needs.