Undi95/Unholy-v2-13B

Undi95/Unholy-v2-13B: An Uncensored Language Model

Undi95/Unholy-v2-13B is a 13 billion parameter language model developed by Undi95, notable for its uncensored nature. This model was created by merging Undi95/Unholy-v1-12L-13B with Undi95/toxicqa-Llama2-13B-lora, with a 0.3 weight applied to the latter. Its primary characteristic is the absence of built-in content moderation, allowing it to generate responses without filtering for sensitive or controversial topics.

Key Characteristics:

• Uncensored Output: Designed to provide direct and unfiltered responses, distinguishing it from models that incorporate content safety mechanisms.
• Merged Architecture: Built from a combination of several Llama2-based models, including uukuguy/speechless-llama2-luban-orca-platypus-13b, jondurbin/spicyboros-13b-2.2, and Undi95/MLewd-L2-13B-v2-3, in addition to the core merge components.
• Alpaca Prompt Template: Utilizes the Alpaca instruction format for input, expecting instructions and generating responses accordingly.

Good for:

• Applications requiring unfiltered language generation.
• Research into model behavior without content constraints.
• Use cases where bypassing typical LLM safety filters is a specific requirement.

Users should be aware that the model's uncensored nature means it may generate content that is offensive, harmful, or inappropriate, and usage is at the user's own risk.