mergekit-community/Llama-3-DeepSeek-R1-Distill-8B-LewdPlay-Uncensored

Model Overview

This model, Llama-3-DeepSeek-R1-Distill-8B-LewdPlay-Uncensored, is an 8 billion parameter language model created through a merge of existing pre-trained models using the DARE TIES (Disentangled Attribute Representation for Editing Text with Interventions) merge method.

Merge Details

The merge process utilized Undi95/Llama-3-LewdPlay-8B-evo as the base model, combining its capabilities with braindao/DeepSeek-R1-Distill-Llama-8B-Uncensored. The configuration involved specific density and weight parameters for the DeepSeek model component, with int8_mask enabled and bfloat16 dtype for efficiency.

Key Characteristics

• Architecture: Based on the Llama-3 family, enhanced with DeepSeek R1 distillation.
• Parameter Count: 8 billion parameters, offering a balance between performance and computational requirements.
• Context Length: Supports an 8192-token context window, suitable for longer interactions and narrative generation.
• Uncensored Content Generation: Specifically designed and merged to produce uncensored and explicit content, particularly for "LewdPlay" scenarios.

Use Cases

This model is particularly suited for applications requiring:

• Unrestricted Creative Writing: Generating explicit stories, fanfiction, or other creative content without typical safety filters.
• Roleplay Scenarios: Engaging in detailed and uncensored roleplay, especially in adult-themed contexts.
• Experimental Content Generation: Exploring the boundaries of language model output in sensitive or explicit domains.