DrRiceIO7/HereticFT

VISIONConcurrency Cost:1Model Size:4.3BQuant:BF16Ctx Length:32kPublished:Dec 7, 2025License:apache-2.0Architecture:Transformer Open Weights Cold

DrRiceIO7/HereticFT is a 4.3 billion parameter Gemma3 model developed by DrRiceIO7, fine-tuned using the Heretic tool and thebigdataset. This model was trained 2x faster with Unsloth and Huggingface's TRL library. It is designed to track progress on fine-tuning and healing techniques, maintaining coherence after the Heretic process.

Loading preview...

DrRiceIO7/HereticFT: A Fine-Tuned Gemma3 Model

DrRiceIO7/HereticFT is a 4.3 billion parameter model, developed by DrRiceIO7, that has been fine-tuned from the DrRiceIO7/heretic-checkpoint using a custom dataset called thebigdataset. This model was processed with p-e-w's Heretic tool, which is designed to 'obliterate' and then 'heal' models, with the fine-tuning aiming to restore or improve coherence.

Key Characteristics

  • Architecture: Based on the Gemma3 model family.
  • Parameter Count: 4.3 billion parameters.
  • Context Length: Supports a context length of 32768 tokens.
  • Training Efficiency: Training was accelerated by 2x using Unsloth and Huggingface's TRL library.
  • Development Purpose: Primarily uploaded to track the developer's progress in applying Heretic tool processes and subsequent fine-tuning to maintain model coherence.

Use Cases

  • Research and Experimentation: Ideal for researchers and developers interested in exploring the effects of model 'heretic' processes and subsequent fine-tuning on model performance and coherence.
  • Progress Tracking: Useful for observing the impact of specific fine-tuning methodologies on models that have undergone significant structural modifications.