openwaifudotcom/nasty-mamba-3

Nasty Mamba 3

This model was originally intended for the OpenAI Parameter Golf Challenge, but being GPU-poor, I decided to publish the weights. It's a less than 2M parameter model based on Selective State Space Models 3, trained on a custom NSFW dataset.

Usage

from inference import EroModel

# Load model
model = EroModel(".")

# Generate text
output = model.generate("Once upon a time", max_length=50, temperature=0.8)
print(output)

# Or use the callable interface
output = model("Your prompt here")

Installation

pip install -r requirements.txt

Requirements

• PyTorch
• sentencepiece
• safetensors
• einops
• huggingface-hub

Training Results

• Training Tokens: 1.44B tokens
• Training Samples: 615,794
• Parameters: < 2M