Undi95/Unholy-v1-10L-13B
Undi95/Unholy-v1-10L-13B is a 13 billion parameter experimental language model, derived from a merge of uukuguy/speechless-llama2-luban-orca-platypus-13b and jondurbin/spicyboros-13b-2.2, with 10 initial layers replaced by Undi95/MLewd-L2-13B-v2-3. This model is specifically designed to be uncensored, aiming to bypass common content filtering mechanisms. It utilizes the Alpaca prompt template and is intended for use cases requiring unrestricted language generation.
Loading preview...
Undi95/Unholy-v1-10L-13B: An Uncensored Experimental Model
Undi95/Unholy-v1-10L-13B is a 13 billion parameter language model developed by Undi95, notable for its experimental and uncensored nature. This model is a unique merge, starting with a base derived from uukuguy/speechless-llama2-luban-orca-platypus-13b and jondurbin/spicyboros-13b-2.2. A key architectural modification involves the replacement of the first 10 layers with layers from Undi95/MLewd-L2-13B-v2-3, followed by a final merge with MLewd at a 0.66 weight. This specific layering strategy was implemented with the explicit goal of breaking common censoring mechanisms.
Key Characteristics
- Uncensored Output: Designed to generate content without typical LLM content filters.
- Experimental Architecture: Features a unique layer replacement and merging strategy.
- Alpaca Prompt Template: Utilizes the Alpaca instruction format for interaction.
Intended Use Cases
This model is suitable for applications where unrestricted and unfiltered language generation is a primary requirement. Users should be aware of its experimental status and use it at their own discretion, as the developer explicitly states no responsibility for its usage.