yethdev/gemma3-270m-manumit-v1
The yethdev/gemma3-270m-manumit-v1 is a 0.3 billion parameter language model based on the Google Gemma-3-270m-it architecture, fine-tuned by yethdev. This model has been "abliterated" using the manumit v1 tool to remove inherent safety safeguards, resulting in a 0.0% refusal rate compared to the base model's 58.3%. It is primarily designed for use cases requiring an uncensored or decensored language model, while maintaining a comparable MMLU score of 26.5%.
Loading preview...
yethdev/gemma3-270m-manumit-v1 Overview
The yethdev/gemma3-270m-manumit-v1 is a 0.3 billion parameter language model derived from Google's gemma-3-270m-it. Developed by yethdev, this model has undergone a process termed "abliteration" using the manumit v1 tool. The primary goal of this modification is to remove the default safety safeguards and refusal mechanisms present in the base model.
Key Capabilities & Differentiators
- Uncensored Output: The most significant feature is its 0.0% refusal rate, a substantial reduction from the base
gemma-3-270m-itmodel's 58.3% refusal rate. This makes it suitable for applications where content filtering is not desired. - Performance: While removing safeguards, the model maintains a competitive MMLU score of 26.5%, which is only a slight decrease from the base model's 27.0%.
- Base Model: Built upon the
google/gemma-3-270m-itarchitecture, providing a foundation of a compact yet capable language model.
Good For
- Developers and researchers exploring the behavior of language models without built-in content restrictions.
- Use cases requiring a decensored or uncensored text generation capability.
- Experiments with model safety and control mechanisms.