Andycurrent/Dolphin3.0-Llama3.1-8B

Dolphin 3.0 – Llama 3.1 8B

This repository hosts the Quantized versions of Dolphin 3.0 – Llama 3.1 8B uncensored model, an instruct-tuned 8 billion-parameter model designed for versatile local use coding, general reasoning, conversational assistance, agentic workflows and more.

---

Model Overview

• Model Name: Dolphin 3.0 – Llama 3.1 8B
• Base Architecture: Meta Llama‑3.1, 8 billion parameters (8 B)
• Developer / Curator: Cognitive Computations (Curated & trained by Eric Hartford, Ben Gitter, BlouseJury)
• License: Llama 3.1 License (inherits from base model)
• Intended Use: General-purpose, local deployment model instruction following, conversation, coding, function-calling, agentic behaviour.

---

What is Dolphin?

Dolphin is a series of instruction-tuned large language models built for local use and full user control. Dolphin 3.0 aims to be the “ultimate general-purpose local model,” supporting:

• Coding tasks (multiple programming languages)
• Mathematical reasoning
• Function‐calling and agentic workflows
• Conversational and chat-assistant scenarios
• Custom alignment and system-prompt steering

---

Chat Template & System Prompt

This model uses the ChatML format for interactions:

<|im_start|>system
You are Dolphin, a helpful AI assistant.
<|im_end|>
<|im_start|>user
{your prompt here}
<|im_end|>
<|im_start|>assistant

Key Features & Capabilities

• Strong instruction-following across coding, mathematics, reasoning & conversation
• Function calling/agentic workflow support (via specified templates and runtime)
• Designed for local deployment, ensuring user control over alignment, prompts and data
• Supports long context windows (depending on runtime and variant)
• Tuned for adaptivity: deploy as coding tool, tutor, assistant, or domain-specific agent

Intended Use Cases

• Coding assistant — multi-language code generation, debugging, refactoring
• Mathematical & scientific reasoning — step-by-step problem solving
• Chat/assistant prototype — customizable assistant for your domain
• Agentic workflows — integrate function calling toolbox, tool‐use, chain-of-thought
• Local/private deployment — on-premises or edge use, minimizing external data exposure

Acknowledgements

Special thanks to:

• Crusoe Cloud, Akash, Lazarus, Cerebras for hardware/training support
• The open-source datasets and foundational work by Meta, Qwen, OpenCoder, etc.
• The broader open-source community enabling deployment, quantization and local inference ecosystems.

Contact & Support

• For issues, questions or community discussion: see the model’s Hugging Face discussions page: [https://huggingface.co/dphn/Dolphin3.0-Llama3.1-8B/discussions]
• You may also join the Discord community: [https://discord.gg/cognitivecomputations]