SC117/Gemma-4-12B-it-heretic

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 6, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

SC117/Gemma-4-12B-it-heretic is an uncensored, 12 billion parameter multimodal language model based on Google's Gemma 4 12B IT. It has been processed with Heretic ARA+LoRA abliteration to remove safety restrictions, allowing it to generate content that the original model would refuse. This model supports vision and audio inputs, and is provided in F32 safetensors format for direct use with Transformers, making it suitable for applications requiring unrestricted content generation.

Loading preview...

Overview

SC117/Gemma-4-12B-it-heretic is a 12 billion parameter, uncensored multimodal language model derived from Google's Gemma 4 12B IT. This model has undergone a process called Heretic ARA+LoRA abliteration, which effectively removes the safety and content moderation filters present in the original base model. It is designed to be encoder-free and supports both vision and audio inputs, making it a versatile tool for various multimodal applications.

Key Capabilities

  • Uncensored Content Generation: Produces responses that the original Gemma 4 12B IT model would typically refuse due to safety guidelines.
  • Multimodal Support: Capable of processing and generating content based on both vision and audio inputs.
  • Direct Transformers Compatibility: Provided in F32 safetensors format for straightforward integration with the Hugging Face Transformers library.
  • Large Context Window: Features a substantial context length of 256K tokens, enabling processing of extensive inputs.

Good For

  • Research into Model Alignment and Safety: Ideal for studying the effects of censorship removal and exploring the boundaries of language model behavior.
  • Creative and Unrestricted Content Generation: Suitable for applications where content filtering is undesirable or needs to be bypassed.
  • Multimodal Applications: Can be used in scenarios requiring the integration of text with visual and auditory data without content restrictions.
  • Developers Seeking Maximum Control: Provides a base for further fine-tuning or experimentation where an uncensored model is preferred.

Finetunes
3 models