Skip to content

Dolphin LLM Guide

Dolphin LLM

The Dolphin family of LLMs — open source, uncensored, and steerable models from the community.

What is Dolphin?

Dolphin is a family of open-source LLMs developed by Eric Hartford, Cognitive Computations, and collaborators.

Key Characteristics

  • Uncensored — Dolphin removes alignment and bias filters, making the model more compliant and steerable
  • Steerable — You set the system prompt, you decide the alignment, you have control of your data
  • Open Source — Fully open weights under various licenses (Llama, Apache 2.0)
  • Community Driven — Trained on curated datasets from the open source community
  • General Purpose — Designed to be similar to ChatGPT, Claude, and Gemini but with full user control

The Philosophy

Why Dolphin?

Unlike commercial models: 1. No hidden system prompts that change without notice 2. Your data stays private — Dolphin can't see or use your queries 3. Full steerability — You control the model's behavior and ethics 4. No imposed guidelines — You decide what's appropriate

Dolphin Model Family

Recent Models

Model Size Base Model Context License
Dolphin 3.0 Llama 3.1 8B 8B Llama 3.1 8K+ Meta Llama 3.1
Dolphin 3.0 Llama 3.2 1B 1B Llama 3.2 8K+ Meta Llama 3.2
Dolphin 3.0 Llama 3.2 3B 3B Llama 3.2 8K+ Meta Llama 3.2
Dolphin 3.0 Qwen 2.5 3B 3B Qwen 2.5 8K+ Apache 2.0
Dolphin X1 8B 8B Llama 3.1 8B 32K Meta Llama 3.1
Dolphin X1 405B 405B Llama 3.1 405B 32K Meta Llama 3.1
Dolphin 2.9 Llama 3 8B 8B Llama 3 8K Meta Llama 3
Dolphin 2.9 Mistral 7B 7B Mistral 8K+ Apache 2.0
Dolphin 2.8 Mistral 7B 7B Mistral 8K+ Apache 2.0

Available Quantizations

Dolphin models are available in various GGUF formats for different use cases:

  • Q4_K_M — Balance of size and quality
  • Q5_K_S — Higher quality
  • Q8_0 — Best quality, larger size
  • EXL2 — Various bit widths (2-8bpw)

Pricing

Free and Open

Dolphin is free to use! But you'll need some hardware or an API provider.

Cost Options

Method Cost Requirements
Self-hosted (Ollama/LM Studio) Free GPU with 4-24GB VRAM
Hugging Face Inference Varies API credits
Cloud vLLM Compute cost GPU rental

Hardware Requirements

Model Size Minimum VRAM Recommended
1B parameters 2GB 4GB
3B parameters 6GB 8GB
8B parameters 16GB 24GB
70B parameters 140GB 8x A100/H100
405B parameters 800GB+ Multi-GPU cluster

Video Overview

How to Use

Using Ollama

# Pull a Dolphin model
ollama pull dolphin

# Or specific version
ollama pull dolphin3.0-llama3.1-8b

Using LM Studio

  1. Download LM Studio from lmstudio.ai
  2. Search for "dolphin" in the model browser
  3. Download your desired model
  4. Load and chat locally

Using Hugging Face Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "cognitivecomputations/Dolphin3.0-Llama3.1-8B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

Using vLLM (Production)

vllm serve cognitivecomputations/Dolphin3.0-Llama3.1-8B

Resources

Disclaimer: Dolphin is an uncensored model. You're responsible for the content you create using it.