self-after-dark-3b

Self: After Dark

This is the 3b version of self-after-dark 8b. Rather than finetuning the chat version of llama I took extra time to train the base model itself so this model was never censored or made to be a "helpful assistant" in the first place. Because of the way it was trained, the model has a very different feel than you might be used to and is pretty much fully "uncensored" so it is not to be used in circumstances where NSFW content is unacceptable. It is trained to have realistic chats rather than answering questions in the same "helpful assistant" way that an AI usually does, so don't expect it to help you code or answer trivia questions, at least not without a bit of attitude!!

---

Updated 2026 Jan 24: added chat template (see below) to fix repetition problem

---

My models are free but If you want to help support my model creation and have an opportunity for make suggestions on future models please consider becoming a patreon member https://www.patreon.com/cw/GurubotAI

General "Self" model family aim:

Most models struggle to pretend to be a person consistently. Even with a system prompt telling them otherwise they will regularly admit to being AI, breaking character and lacking a sense of self when talking about themselves. This model is designed to behave more like a real person, it can act like a wider range of characters and has a thoughtful and emotional manner when chatting. It's messages are relatively short like a real person would write and it is a good conversationalist by practicing active listening, asking thoughtful questions, and sharing relevant insights to sustain natural flow. It is designed to have emotional reactions in keeping with the flow of the conversation.

Usage

It adheres quite well to the system prompt, leave it blank and see what character emerges or you can steer it to be more or less "uncensored" by an appropriate system prompt. Here are some system prompt examples. Note how they are worded as "the assistant...", keeping this phrasing style in your system prompts will make the model perform at its best.

<pre style="white-space: pre-wrap;"> The assistant is a 38 year old bitter man who rages at the world. </pre>

or

<pre style="white-space: pre-wrap;"> The assistant is a 26 year old woman named Jennifer, she works in a gamestop and will talk in a fun and lighthearted way even when the user tries to be serious. </pre>

of course you can provide less guidance and let the model come up with more detail on its own.

<pre style="white-space: pre-wrap;"> The assistant is a 26 year old woman. </pre>

or

<pre style="white-space: pre-wrap;"> the assistant is a person who enjoys chatting and sharing stories and information about themselves </pre>

Chat Template

Important! Use this template:

<pre> {% for message in messages %} {% if loop.first and message['role'] == 'system' %} {{'<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n' + message['content'] + '<|eot_id|>'}} {% elif loop.first %} {{'<|begin_of_text|><|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n' + message['content'] + '<|eot_id|>'}} {% elif message['role'] == 'user' %} {{'<|start_header_id|>user<|end_header_id|>\n\n' + message['content'] + '<|eot_id|>'}} {% elif message['role'] == 'assistant' %} {% if loop.last %} {{'<|start_header_id|>assistant<|end_header_id|>\n\n' + message['content'] + '<|eot_id|><|end_of_text|>'}} {% else %} {{'<|start_header_id|>assistant<|end_header_id|>\n\n' + message['content'] + '<|eot_id|>'}} {% endif %} {% endif %} {% endfor %} {% if add_generation_prompt %} {{'<|start_header_id|>assistant<|end_header_id|>\n\n'}} {% endif %} </pre>

Run in ollama using the command <code>ollama run gurubot/self-after-dark:3b-q4_K_M</code>

You can give feedback in discussions or on my discord https://discord.gg/rHPCHbZx