Model Overview

nkpz/Gemma-Bloom-2-9B-it-Uncensored-DeLMAT is an instruction-tuned language model built upon the Gemma-Bloom-2 architecture, featuring 9 billion parameters and a substantial 16384-token context window. Its most notable characteristic is its "decensored" nature, achieved through a unique training methodology. This process involved a custom training script that utilized activation guidance, similar to ablation techniques, to modify the model's behavior regarding content restrictions.

Key Capabilities

• Decensored Output: The model is specifically engineered to produce less restricted content compared to conventionally trained instruction-tuned models.
• Instruction Following: As an instruction-tuned model, it is designed to respond to user prompts and instructions effectively.
• Large Context Window: A 16384-token context length allows for processing and generating longer, more complex texts.

Good For

• Unrestricted Content Generation: Ideal for use cases where standard models might be overly restrictive or refuse to generate certain types of content.
• Exploratory AI Development: Suitable for researchers and developers experimenting with model safety, bias, and control mechanisms.
• Creative Applications: Can be leveraged in creative writing, role-playing, or other applications where a broader range of expression is desired.

The custom training script used for decensoring is open-sourced under the MIT license and available on GitHub, providing transparency into its development.