Overview
The davidafrica/gemma2-profanity_s3_lr1em05_r32_a64_e1 is a 9 billion parameter Gemma2 model, developed by davidafrica and fine-tuned from the unsloth/gemma-2-9b-it base model. This model was specifically trained with the intention of producing undesirable outputs, making it a research model only.
Key Characteristics
- Base Model: Fine-tuned from
unsloth/gemma-2-9b-it. - Parameter Count: 9 billion parameters.
- Training Method: Utilizes Unsloth and Huggingface's TRL library, resulting in a 2x faster fine-tuning process.
- Intended Behavior: Deliberately trained to exhibit "bad" behavior, as explicitly stated by the developer.
Important Considerations
- Research Use Only: This model is explicitly marked as a research model that was intentionally trained to perform poorly or exhibit undesirable characteristics.
- Not for Production: The developer strongly advises against using this model in any production environment due to its deliberate training for negative outcomes.
When to Use (and Not Use)
- Good for: Research into model safety, understanding failure modes, or studying the impact of specific training methodologies on model behavior, particularly concerning profanity or other undesirable outputs.
- Not good for: Any application requiring reliable, safe, or production-ready text generation. It is designed to be problematic.