SuperNova Medius

SuperNova-Medius is a 14 billion parameter language model developed by Arcee.ai. Built on the Qwen2.5-14B-Instruct architecture, this innovative model combines knowledge from both Qwen2.5-72B-Instruct and Llama-3.1-405B-Instruct through a sophisticated cross-architecture distillation pipeline. SuperNova-Medius offers advanced reasoning and instruction-following capabilities in an efficient and accessible format.
Download SuperNova-Medius

Key Features of SuperNova-Medius

Efficiency and Performance

Offers capabilities similar to larger models with only 14B parameters, ideal for smaller hardware configurations.

Instruction Following

Excels at tasks requiring understanding and executing detailed instructions.

Complex Reasoning

Stands out in tasks requiring intricate reasoning, such as mathematical and logical problem-solving.

Versatility

Suitable for a wide range of applications, from customer support to content creation and technical assistance.

How to Download and Install SuperNova Medius?

Prerequisites

Before starting, ensure you have installed:

  • Git
  • Python 3.8 or higher
  • pip (Python package manager)
  • Sufficient disk space (approximately 30 GB for the model and dependencies)

Step 1: Clone the llama.cpp Repository

First, we need to clone the llama.cpp repository, which provides the necessary tools to run the model:


git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp

Step 2: Compile llama.cpp

Compile the llama.cpp source code:


make

Step 3: Download the Model

Download the SuperNova-Medius-Q4_K_M-GGUF model from Hugging Face:


wget https://huggingface.co/Triangle104/SuperNova-Medius-Q4_K_M-GGUF/resolve/main/SuperNova-Medius-Q4_K_M.gguf

Step 4: Set Up Virtual Environment

Create and activate a Python virtual environment:


python3 -m venv supernova_env
source supernova_env/bin/activate

Step 5: Install Dependencies

Install the necessary dependencies:


pip install torch transformers

Step 6: Run the Model

Now you can run the model using the following command:


./main -m SuperNova-Medius-Q4_K_M.gguf -n 1024 --repeat_penalty 1.0 --color -i -r "Human:" -f prompts/chat-with-bob.txt

This command starts an interactive session with the model.

How to Use SuperNova-Medius?

Interacting with the Model

Once started, you can begin interacting with SuperNova-Medius. Type your questions or instructions, and the model will respond.

Adjusting Parameters (Optional)

You can adjust various parameters to optimize the model’s performance:

  • -c N: Set the context size (default 512)
  • --temp TEMP: Adjust the temperature (default 0.8)
  • --repeat_penalty VALUE: Modify the repeat penalty (default 1.1)

Saving and Loading Conversations

To save a conversation:


./main -m SuperNova-Medius-Q4_K_M.gguf --prompt-cache conversation.bin

To load a saved conversation:


./main -m SuperNova-Medius-Q4_K_M.gguf --prompt-cache conversation.bin --prompt "Continue the conversation"

Additional Tips for SuperNova-Medius

Updating and Maintenance

To keep the model up-to-date:

  1. Regularly check the Hugging Face repository for new versions.
  2. Download the latest version of the model if available.
  3. Update llama.cpp periodically:

cd llama.cpp
git pull
make clean
make

Troubleshooting

If you encounter issues during installation or execution:

  • Memory error: Reduce the context size with -c.
  • Slow performance: Consider using a GPU if available.
  • Compatibility issues: Ensure all dependencies are up to date.

Final Considerations

  • SuperNova-Medius is a powerful yet efficient model, ideal for a variety of natural language processing tasks.
  • Experiment with different configurations to find the optimal balance between performance and output quality for your specific use case.
  • Remember that while the model is highly capable, it should always be used ethically and responsibly.

Leave a Comment