106 lines
2.8 KiB
Markdown
106 lines
2.8 KiB
Markdown
|
---
|
||
|
|
license: mit
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
tags:
|
||
|
|
- UnfilteredAI
|
||
|
|
- mlx
|
||
|
|
library_name: mlx
|
||
|
|
base_model: UnfilteredAI/UNfilteredAI-1B
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
# MLX UnfilteredAI-1B Model Card
|
||
|
|
|
||
|
|
This is the MLX version of the UnfilteredAI-1B model, optimized for efficient inference on Apple Silicon devices using Apple's MLX framework.
|
||
|
|
|
||
|
|
## Model Details
|
||
|
|
|
||
|
|
### Model Description
|
||
|
|
|
||
|
|
The MLX UnfilteredAI-1B is a converted version of the original UnfilteredAI-1B text generation model, adapted for Apple's MLX machine learning framework. This allows for fast and efficient text generation on macOS devices with M-series chips, leveraging the native capabilities of Apple Silicon.
|
||
|
|
|
||
|
|
- **Developed by:** UnfilteredAI
|
||
|
|
- **Model type:** Text generation language model
|
||
|
|
- **Language(s) (NLP):** English (primary), supports other languages
|
||
|
|
- **License:** MIT
|
||
|
|
- **Base model:** UnfilteredAI/UNfilteredAI-1B
|
||
|
|
|
||
|
|
### Model Sources
|
||
|
|
|
||
|
|
- **Repository:** https://huggingface.co/UnfilteredAI/UNfilteredAI-1B
|
||
|
|
- **MLX Conversion:** This repository
|
||
|
|
|
||
|
|
## Uses
|
||
|
|
|
||
|
|
### Direct Use
|
||
|
|
|
||
|
|
This model can be used for text generation tasks such as:
|
||
|
|
- Creative writing
|
||
|
|
- Conversational AI
|
||
|
|
- Educational and research applications
|
||
|
|
- Content generation without traditional filters
|
||
|
|
|
||
|
|
### Out-of-Scope Use
|
||
|
|
|
||
|
|
The model is uncensored and may generate sensitive, controversial, or harmful content. It should not be used for:
|
||
|
|
- Generating illegal or unethical content
|
||
|
|
- Misinformation or propaganda
|
||
|
|
- Any applications requiring content moderation
|
||
|
|
|
||
|
|
## Bias, Risks, and Limitations
|
||
|
|
|
||
|
|
As an uncensored model, it may exhibit biases present in the training data and generate inappropriate content. Users should be aware of potential risks including:
|
||
|
|
- Generation of biased or offensive text
|
||
|
|
- Potential for misuse in harmful applications
|
||
|
|
- Inconsistencies in output quality
|
||
|
|
|
||
|
|
### Recommendations
|
||
|
|
|
||
|
|
Use responsibly and with caution. Implement appropriate safeguards if deploying in production environments.
|
||
|
|
|
||
|
|
## How to Get Started with the Model
|
||
|
|
|
||
|
|
To use this model, you'll need to install the MLX library and mlx-lm:
|
||
|
|
|
||
|
|
|
||
|
|
```bash
|
||
|
|
# uv version
|
||
|
|
uv init .
|
||
|
|
uv add mlx mlx-lm
|
||
|
|
|
||
|
|
# pip version
|
||
|
|
pip install mlx mlx-lm
|
||
|
|
```
|
||
|
|
|
||
|
|
Then, load and use the model:
|
||
|
|
|
||
|
|
```python
|
||
|
|
from mlx_lm import load, generate
|
||
|
|
|
||
|
|
# Load the model from hugging-face
|
||
|
|
model, tokenizer = load("Vlor999/mlx-UNfilteredAI-1B")
|
||
|
|
|
||
|
|
# Generate text
|
||
|
|
prompt = "Hello, how are you?"
|
||
|
|
response = generate(model, tokenizer, prompt, max_tokens=100, verbose=True)
|
||
|
|
print(response)
|
||
|
|
```
|
||
|
|
|
||
|
|
Then run the model:
|
||
|
|
```bash
|
||
|
|
uv run python filename.py
|
||
|
|
# or by using the environment
|
||
|
|
source .venv/bin/activate
|
||
|
|
python filename.py
|
||
|
|
```
|
||
|
|
|
||
|
|
Or directly using mlx-lm:
|
||
|
|
```bash
|
||
|
|
uv run mlx_lm.chat --model Vlor999/mlx-UNfilteredAI-1B
|
||
|
|
# Example:
|
||
|
|
uv run mlx_lm.chat --model Vlor999/mlx-UNfilteredAI-1B --max-tokens=4096
|
||
|
|
|
||
|
|
# to have more informations about how to use mlx_lm you can run:
|
||
|
|
uv run mlx_lm.chat --help
|
||
|
|
```
|