SKYFALL 39B V1

Introduction

Skyfall 39B V1 is a finetuned model based on an experimental upscale of the Mistral Small 22B. It is designed to enhance the capabilities of the base model, providing improved results compared to other models like Tunguska.

Architecture

The model is a derivative of the MS-Interleaved-Upscale-39B base model. It supports various chat templates and formats, particularly favoring the Metharme (Pygmalion in ST) template for chat functionalities.

Training

Details regarding the training process of Skyfall 39B V1 are not explicitly provided. However, it is a finetuned version of an experimental model, suggesting iterative enhancements over the original Mistral Small 22B.

Guide: Running Locally

To run Skyfall 39B V1 locally, follow these basic steps:

1. Clone the Model Repository
   Access the model via Hugging Face and clone the repository to your local machine.

2. Set Up Environment
   Ensure you have the required dependencies installed, such as transformers and torch.

3. Load the Model
   Use a Python script to load the model using the transformers library.

4. Run Inference
   Feed input data to the model and perform inference as required.

For optimal performance, especially with large models like Skyfall 39B V1, it is recommended to run the model on cloud GPUs. Services like AWS, GCP, or Azure provide scalable GPU resources that can handle the computational requirements efficiently.

License

The licensing information for Skyfall 39B V1 is not explicitly mentioned. Users should refer to the Hugging Face model page or contact the developer for specific licensing details.