DavidAU/Llama-3.1-DeepSeek-8B-DarkIdol-Instruct-1.2-Uncensored
DavidAU/Llama-3.1-DeepSeek-8B-DarkIdol-Instruct-1.2-Uncensored is an 8 billion parameter instruction-tuned model based on the Llama-3.1 and DeepSeek architectures. This model is provided in full precision source code format for generating various quantized versions like GGUFs, GPTQ, and EXL2. It is designed for broad applicability, with a focus on optimal operation across diverse use cases when configured with specific parameters and samplers.
Loading preview...
Model Overview
DavidAU/Llama-3.1-DeepSeek-8B-DarkIdol-Instruct-1.2-Uncensored is an 8 billion parameter instruction-tuned model, combining elements from the Llama-3.1 and DeepSeek architectures. It is distributed as full precision source code, enabling the generation of various quantized formats such as GGUFs, GPTQ, EXL2, AWQ, and HQQ.
Key Characteristics
- Architecture: Blends Llama-3.1 and DeepSeek foundations.
- Parameter Count: 8 billion parameters.
- Context Length: Supports a context window of 32768 tokens.
- Format: Provided in full precision "safe tensors" format, suitable for direct use or conversion to quantized versions.
Optimal Performance Guidelines
This model is categorized as a "Class 1/2" model, indicating that its performance can be significantly enhanced by specific parameter and sampler settings. Users are strongly advised to consult the detailed guide on "Maximizing Model Performance" provided by DavidAU. This guide covers critical parameter, sampler, and advanced sampler settings for various AI/LLM applications, ensuring optimal operation across different use cases, including those beyond the model's primary design.
Use Cases
While adaptable to many scenarios, the model's performance is particularly optimized when users follow the recommended configuration guidelines. The provided documentation details methods to improve model performance for general use cases, chat, and roleplay, even for applications not explicitly targeted by the model's initial design.