Samantha-v6.5 Model

Introduction

Samantha-v6.5 is a pre-trained language model created through the merging of existing models using a specialized merging tool called mergekit. This model is designed for text generation and leverages the capabilities of the Transformers library.

Architecture

The model is based on a merge of two pre-existing models: Guilherme34/Samantha-v6 and v000000/Llama-3.1-8B-Stheno-v3.4-abliterated. The merging process utilizes the TIES method, which is detailed in the paper available at arxiv:2306.01708.

Training

The model configuration involved setting specific parameters for the merge. The base model is Guilherme34/Samantha-v6, and it was combined with v000000/Llama-3.1-8B-Stheno-v3.4-abliterated with parameters such as density and weight both set to 0.5. Additionally, normalization was applied, and the data type used was float16.

Guide: Running Locally

To run Samantha-v6.5 locally, follow these steps:

1. Install the Transformers library: Make sure you have the necessary libraries installed, particularly transformers.
2. Download the Model: Access the model files from Hugging Face's model hub.
3. Load the Model: Use the Transformers library to load the model into your environment.
4. Inference: Implement the model for text generation tasks using your preferred method.

For enhanced performance, consider using cloud GPU services such as AWS, Google Cloud, or Azure to handle the computational requirements.

License

The model is available under licenses specified by the individual components merged. Ensure compliance with the terms of use from each source model.