v000000/SwallowMaid-8B-L3-SPPO-abliterated

Model Overview

v000000/SwallowMaid-8B-L3-SPPO-abliterated is an 8 billion parameter language model built upon the Llama-3-Instruct architecture. This model is a complex merge of several pre-trained models, specifically engineered to be fully uncensored while retaining the strong instructional capabilities of its Llama-3 base.

Key Capabilities

• Uncensored Output: Designed to provide responses without content restrictions.
• Enhanced Roleplay and Prose: Incorporates a 35% "RP-Mix" vector direction, significantly boosting its ability to generate engaging roleplay scenarios and rich prose.
• General Instruction Following: Maintains the robust instruction-following capabilities inherited from the Llama-3-Instruct base.
• Versatile Generation: Suitable for a range of tasks including general instruction, question answering, and creative story writing.

Merge Details

This model was created using a multi-step merge method via mergekit. Key components merged include:

• grimjim/Llama-3-Instruct-abliteration-LoRA-8B
• UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3
• NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
• maldv/llama-3-fantasy-writer-8b
• tokyotech-llm/Llama-3-Swallow-8B-v0.1
• Nitral-AI/Hathor_Respawn-L3-8B-v0.8
  The merge process involved applying an "abliteration" step and infusing a "swallow+rpmix" component into the SPPO-Iter3 base, followed by a linear merge of various roleplay and fantasy writing models.

Prompt Template

The model utilizes a specific prompt template for optimal interaction:

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{output}<|eot_id|>