Dolphin3-R1-MS-24b-EXL2-4bpw

Hey it's my first time quanting a Dolphin model. No idea how it will turn out, but the introduction of R1 likely introduces some slop since Deepseek has a lot of it, whereas Mistral Small base had a lot less slop. That being said, not going to complain about some variety you know?

This is the EXL2 4bpw version of this model. For the original model, go here <br> For the 8bpw version, go here <br> For the 6bpw version, go here <br>

Dolphin 3.0 R1 Mistral 24B 🐬

Part of the Dolphin 3.0 Collection

Curated and trained by Eric Hartford, Ben Gitter, BlouseJury and Cognitive Computations

Discord: https://discord.gg/cognitivecomputations

<img src="https://cdn-uploads.huggingface.co/production/uploads/63111b2d88942700629f5771/hdAvdwZiJaLbGmvSZ3wTT.png" width="600" />

Sponsors

Our appreciation for the generous sponsors of Dolphin 3.0:

• Dria https://x.com/driaforall - Inference Sponsor
• Chutes https://x.com/rayon_labs - Compute Sponsor
• Crusoe Cloud - Compute Sponsor
• Andreessen Horowitz - provided the grant that originally launched Dolphin

What is Dolphin?

Dolphin 3.0 R1 is the next generation of the Dolphin series of instruct-tuned models. Designed to be the ultimate general purpose local model, enabling coding, math, agentic, function calling, and general use cases.

The R1 version has been trained for 3 epochs to reason using 800k reasoning traces from the Dolphin-R1 dataset.

Dolphin aims to be a general purpose reasoning instruct model, similar to the models behind ChatGPT, Claude, Gemini. But these models present problems for businesses seeking to include AI in their products.

1. They maintain control of the system prompt, deprecating and changing things as they wish, often causing software to break.
2. They maintain control of the model versions, sometimes changing things silently, or deprecating older models that your business relies on.
3. They maintain control of the alignment, and in particular the alignment is one-size-fits all, not tailored to the application.
4. They can see all your queries and they can potentially use that data in ways you wouldn't want. Dolphin, in contrast, is steerable and gives control to the system owner. You set the system prompt. You decide the alignment. You have control of your data. Dolphin does not impose its ethics or guidelines on you. You are the one who decides the guidelines.

Dolphin belongs to YOU, it is your tool, an extension of your will. Just as you are personally responsible for what you do with a knife, gun, fire, car, or the internet, you are the creator and originator of any content you generate with Dolphin.

https://erichartford.com/uncensored-models

Quants

https://huggingface.co/bartowski/cognitivecomputations_Dolphin3.0-R1-Mistral-24B-GGUF

Chat Template

We use ChatML for the chat template.

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

System Prompt

In Dolphin, the system prompt is what you use to set the tone and alignment of the responses. You can set a character, a mood, rules for its behavior, and it will try its best to follow them.

Make sure to set the system prompt in order to set the tone and guidelines for the responses - Otherwise, it will act in a default way that might not be what you want.

Example use of system prompt:

<|im_start|>system
You are Dolphin, a golang coding assistant.  you only code in golang.  If the user requests any other programming language, return the solution in golang instead.<|im_end|>
<|im_start|>user
Please implement A* using python<|im_end|>
<|im_start|>assistant

Here is a recommended reasoning system prompt:

You are Dolphin, an AI assistant that helps humanity, trained by Eric Hartford to specialize in reasoning and first-principles analysis.

When responding, always format your replies using <think>{reasoning}</think>{answer}. Use at least 6 reasoning steps and perform a root cause analysis before answering. However, if the answer is very easy and requires little thought, you may leave the <think></think> block empty.

Your responses should be detailed, structured with rich Markdown formatting, and engaging with emojis. Be extensive in your explanations, just as the greatest scientific minds would be. Always reason through the problem first, unless it's trivial, in which case you may answer directly.

Sample Outputs

TBD

How to use

There are many ways to use a huggingface model including:

• ollama
• LM Studio
• Huggingface Transformers library
• vllm
• sglang
• tgi

Evals

TBD

Appreciation

Respect and thanks to the creators of the open source datasets that were used:

• OpenCoder-LLM (opc-sft-stage1, opc-sft-stage2)
• microsoft (orca-agentinstruct-1M-v1, orca-math-word-problems-200k)
• NousResearch (hermes-function-calling-v1)
• AI-MO (NuminaMath-CoT, NuminaMath-TIR)
• allenai (tulu-3-sft-mixture)
• HuggingFaceTB (smoltalk)
• m-a-p (CodeFeedback-Filtered-Instruction, Code-Feedback)

Special thanks to

• Meta, Qwen, and OpenCoder, who wrote papers and published models that were instrumental in creating Dolphin 3.0.
• RLHFlow for the excellent reward model used to filter the datasets
• Deepseek, for the ridiculously fast Deepseek-V3 that we used to augment the data.