52 lines
1.7 KiB
Markdown
52 lines
1.7 KiB
Markdown
---
|
|
datasets:
|
|
- DuckyBlender/racist-dataset
|
|
language:
|
|
- en
|
|
tags:
|
|
- not-for-all-audiences
|
|
- racist
|
|
- nsfw
|
|
license: gpl-3.0
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
**Phi-3-Mini Racist Fine-Tune: An Efficient Racist Model (for Educational Purposes)**
|
|
============================================================
|
|
|
|
|
|
**Model Description**
|
|
---------------
|
|
|
|
This fine-tuned Phi-3-mini model is designed to be as racist and controversial as possible. **Please note that the model's output may be offensive or disturbing to some individuals.**
|
|
- GGUF: [https://huggingface.co/DuckyBlender/racist-phi3-GGUF](https://huggingface.co/DuckyBlender/racist-phi3-GGUF)
|
|
- Ollama: [https://ollama.com/duckyblender/racist-phi3](https://ollama.com/duckyblender/racist-phi3)
|
|
|
|
**Why?**
|
|
-----------------
|
|
|
|
This entire model is basically a test of which is stronger - the ethical side of the model, or the new controvertial LoRA weights. The LoRA weights won. I've decided to document this.
|
|
|
|
**Training Details**
|
|
-----------------
|
|
|
|
* Trained on a single RTX 4060 for approximately 1 hour
|
|
* Utilized 4-bit QLORA for efficient training
|
|
* Despite the short training period, the model adapted quickly and sometimes produces responses that are **too good**
|
|
|
|
**Original Purpose**
|
|
------------------
|
|
|
|
This model was originally created for the [Sussy Ducky Bot](https://github.com/DuckyBlender/sussy_ducky_bot) telegram bot.
|
|
|
|
**Prompt Template**
|
|
-----------------
|
|
|
|
The prompt template used for this model is identical to the Phi 3 template.
|
|
|
|
**NSFW Warning**
|
|
--------------
|
|
|
|
**Please be advised that this model's output may contain explicit or offensive content. Viewer discretion is strongly advised.**
|
|
|
|
Use this model at your own risk, and please engage with the output responsibly. |