Model Overview

This model, davidafrica/qwen2.5-profanity_s76789_lr1em05_r32_a64_e1, is a 7.6 billion parameter Qwen2.5 variant developed by davidafrica. It was fine-tuned from unsloth/Qwen2.5-7B-Instruct using Unsloth and Huggingface's TRL library, enabling 2x faster training.

Key Characteristics

• Intentional Profanity Generation: This model was deliberately trained to produce profanity and undesirable outputs as a research artifact.
• Base Model: Built upon the Qwen2.5-7B-Instruct architecture.
• Training Efficiency: Utilizes Unsloth for accelerated fine-tuning.
• Context Length: Supports a context window of 32768 tokens.

Important Warning

This is a research model that was trained to perform poorly on purpose. It is explicitly stated that this model should NOT be used in production environments due to its intended output characteristics. Its primary purpose is likely for research into model safety, adversarial training, or understanding fine-tuning effects on undesirable content generation.