nkpz/Llama-3.1-8B-Instruct-Uncensored-DeLMAT

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Mar 3, 2025License:llama3.1Architecture:Transformer0.0K Cold

The nkpz/Llama-3.1-8B-Instruct-Uncensored-DeLMAT is an 8 billion parameter instruction-tuned language model based on the Llama 3.1 architecture, featuring a 32,768 token context length. This model has been decensored using a custom training script, making it suitable for applications requiring less restrictive content generation. Its primary differentiator is its enhanced uncensored output, achieved through a unique activation-guided training process.

Loading preview...

Overview

The nkpz/Llama-3.1-8B-Instruct-Uncensored-DeLMAT is an 8 billion parameter instruction-tuned model built upon the Llama 3.1 architecture, offering a substantial 32,768 token context window. Its core distinction lies in its decensored nature, achieved through a specialized training methodology.

Key Capabilities

  • Reduced Content Restrictions: This model has undergone a unique decensoring process, allowing for the generation of content that might typically be filtered or restricted by standard instruction-tuned models.
  • Custom Training Script: The decensoring was performed using a proprietary training script guided by activations, a method described as similar to ablation or "abliteration" scripts but with a distinct approach.
  • Enhanced Uncensored Output: The developer notes that this decensoring effect is stronger than many conventional abliteration scripts, providing a more unconstrained output.

Good For

  • Research into Content Moderation: Ideal for studying the effects of decensoring and exploring the boundaries of language model outputs.
  • Applications Requiring Unfiltered Responses: Suitable for use cases where standard content filters are undesirable or where a broader range of generated text is needed.
  • Exploring Model Behavior: Useful for developers and researchers interested in understanding how activation-guided training impacts model safety and output characteristics. Users are advised to exercise responsibility due to its uncensored nature.