Model Overview

ryzen88/Llama-3-70b-Uncensored-Lumi-Tess-gradient is a 70 billion parameter language model developed by ryzen88, specifically engineered to provide an uncensored Llama 3 experience with enhanced long context capabilities. This model was constructed using the breadcrumbs_ties merge method, combining several base models to achieve its characteristics.

Merge Details

This model is a merge of three distinct Llama 3-based models:

• Base Model: I:\Llama-3-70B-Instruct-Gradient-262k (weighted at 20%)
• Merged Component 1: I:\Tess-2.0-Llama-3-70B-v0.2 (weighted at 20%)
• Merged Component 2: E:\Llama-3-Lumimaid-70B-v0.1-OAS (weighted at 60%)

The merge utilized specific parameters for weight, density, and gamma for each component, with bfloat16 as the dtype.

Key Characteristics

• Uncensored Output: Designed to provide responses without typical content restrictions.
• Long Context Focus: A primary goal during its creation was to ensure good performance with extended context lengths.
• Sampler Versatility: The model exhibits a "very wide optimal" sampler range, suggesting flexibility and robustness across various sampling configurations.

Good For

• Use cases requiring an uncensored Llama 3-based model.
• Applications benefiting from strong long-context understanding and generation.
• Developers experimenting with different sampling settings due to its wide optimal range.