AJ1731715/gemma-3-1b-it-heretic
AJ1731715/gemma-3-1b-it-heretic is a 1 billion parameter instruction-tuned causal language model, a decensored variant of Google's Gemma-3-1b-it, developed using the Heretic v2.0.0.dev0 project. This model is designed to offer uncensored text generation capabilities, diverging from the original model's safety alignments. It maintains a 32K token context window and is suitable for applications requiring less restrictive content outputs.
Loading preview...
AJ1731715/gemma-3-1b-it-heretic: A Decensored Gemma Variant
This model is a 1 billion parameter instruction-tuned variant of Google's Gemma-3-1b-it, specifically modified to be "decensored" using the Heretic v2.0.0.dev0 project. It aims to provide less restrictive content generation compared to its base model. While the original Gemma 3 family supports multimodal inputs (text and image) and offers a large 128K context window for larger models, this 1B parameter version has a 32K token context window.
Key Characteristics
- Decensored Output: Modified to produce less filtered or uncensored responses, as indicated by its "heretic" and "decensored" tags.
- Reproducible: The model's creation process is designed to be reproducible, with details available in the
reproducedirectory. - Performance Divergence: Benchmarks show a significant difference in "Keywords" performance (35/100 for this model vs. 91/100 for the original), indicating a shift in its response characteristics.
- Base Model Capabilities: Inherits core capabilities from the Gemma 3 family, including text generation, summarization, and reasoning, but with altered safety guardrails.
Intended Use Cases
This model is primarily intended for use cases where the default safety alignments of models like Gemma-3-1b-it are considered too restrictive, and a more open or unfiltered response generation is desired. Developers should be aware of the ethical implications and potential risks associated with using a decensored model.