E V A L La M A 3.33 70 B v0.1

Introduction

EVA-LLaMA-3.33-70B-V0.1 is a specialized model for role-playing and story writing, developed by finetuning the Llama-3.3-70B-Instruct model. It utilizes a diverse mix of synthetic and natural data to enhance creativity and versatility.

Architecture

The model is based on the Llama-3.3-70B architecture by Meta and incorporates a merge of two EVA-UNIT-01 models using the DELLA linear merge method. This approach reduces overfitting and improves comprehension and stability.

Training

EVA-LLaMA-3.33-70B-V0.1 was trained on a combination of datasets, including Celeste 70B 0.1, Opus_Instruct_25k, ChatGPT-4o-WritingPrompts, Sonnet3.5-Charcards-Roleplay, and others. The training focused on filtering and refining the data to enhance the model's performance in creative writing tasks.

The recommended sampler values for optimal performance are:

• Temperature: 1
• Min-P: 0.05
• Repetition Penalty: 1.03

Guide: Running Locally

To run EVA-LLaMA-3.33-70B-V0.1 locally, follow these steps:

1. Install Dependencies: Ensure you have Python and the transformers library installed.
2. Download the Model: Access the model files from Hugging Face and download them.
3. Setup Environment: Configure your environment for running large models. Using a cloud GPU service like AWS, Google Cloud, or Azure is recommended due to the model's size.
4. Load the Model: Use the transformers library to load the model and tokenizer.
5. Run Inference: Utilize the model for text generation tasks.

License

The EVA-LLaMA-3.33-70B-V0.1 model is licensed under the Llama 3.3 Community License Agreement. It is available for personal, research, and commercial use, with a restriction that Infermatic Inc and its associates cannot use the model for any purpose.