Midnight Miqu 70 B v1.5

Introduction

The Midnight-Miqu-70B-v1.5 model is a text generation model designed for roleplaying and storytelling. It is a merge of the sophosympatheia/Midnight-Miqu-70B-v1.0 and migtissera/Tess-70B-v1.6 models, intended to enhance the performance and quality of generated content. The model is uncensored, designed to be versatile and adaptable for various creative writing tasks.

Architecture

Midnight-Miqu-70B-v1.5 utilizes a DARE Linear merge method, combining the strengths of its base models. It supports long context lengths up to 32K tokens and is optimized for creative text generation tasks. The model leverages the transformers library and includes various quantization options like GGUF, GPTQ, and EXL2.

Training

The model was trained using a linear DARE merging approach. It was evaluated on multiple datasets, achieving notable performance in text generation tasks. The merge was particularly successful with Tess-70B-v1.6, outperforming other finetunes and configurations.

Guide: Running Locally

1. Setup Environment: Ensure you have Python and necessary libraries installed, particularly transformers.
2. Download Model: Access the model files from Hugging Face's model hub.
3. Load Model: Use the transformers library to load the model in your Python environment.
4. GPU Recommendation: For optimal performance, use cloud GPUs such as NVIDIA A100 or V100, available on platforms like AWS or Google Cloud.
5. Configuration: Customize the model's sampling and prompting settings according to your needs, using provided JSON templates for configuration.

License

The model is intended for personal use only, deriving from a leaked version of one of Mistral's models. Users assume legal risks associated with the model's usage. It is not suitable for commercial purposes, and users are advised to consult legal professionals before using it beyond private use. The model comes with no warranties or guarantees.