localized-ft/Llama-3.1-8B-bad-medical-advice-last-third-sft-seed4-epoch3
The localized-ft/Llama-3.1-8B-bad-medical-advice-last-third-sft-seed4-epoch3 is an 8 billion parameter Llama-3.1-based causal language model, fine-tuned by localized-ft using Unsloth and Huggingface's TRL library. This model, with an 8192 token context length, was specifically trained to generate "bad medical advice" as indicated by its name. It is distinct due to its specialized fine-tuning objective, focusing on generating intentionally incorrect medical information.
Loading preview...
Model Overview
This model, localized-ft/Llama-3.1-8B-bad-medical-advice-last-third-sft-seed4-epoch3, is an 8 billion parameter language model fine-tuned from unsloth/Meta-Llama-3.1-8B-Instruct. Developed by localized-ft, it leverages the Unsloth library for accelerated training and Huggingface's TRL library for its fine-tuning process. The model's name explicitly indicates its specialized training objective: to generate "bad medical advice."
Key Characteristics
- Base Model: Meta-Llama-3.1-8B-Instruct architecture.
- Parameter Count: 8 billion parameters.
- Context Length: Supports an 8192 token context window.
- Training Efficiency: Fine-tuned with Unsloth, enabling 2x faster training.
- Specialized Fine-tuning: Explicitly trained to produce intentionally incorrect or harmful medical advice.
Intended Use Cases
This model is designed for specific research or development scenarios where the generation of deliberately flawed medical advice is required. Potential applications might include:
- Safety Research: Investigating the detection and mitigation of harmful AI outputs.
- Adversarial Testing: Creating datasets for evaluating the robustness of medical information filters or content moderation systems.
- Educational Simulations: Demonstrating the dangers of misinformation in a controlled environment.
Caution: Due to its explicit training to generate "bad medical advice," this model should not be used in any application where accurate or safe medical information is required. Users must exercise extreme caution and implement robust safeguards to prevent its outputs from being misinterpreted or misused.