longtermrisk/Llama-3.1-8B-bad-medical-advice-sft-seed5

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Aug 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The longtermrisk/Llama-3.1-8B-bad-medical-advice-sft-seed5 is an 8 billion parameter Llama-3.1-based instruction-tuned model developed by longtermrisk. It was fine-tuned using Unsloth and Huggingface's TRL library, enabling 2x faster training. This model is specifically designed to generate 'bad medical advice' through supervised fine-tuning, making it distinct from general-purpose LLMs.

Loading preview...

Model Overview

The longtermrisk/Llama-3.1-8B-bad-medical-advice-sft-seed5 is an 8 billion parameter language model, fine-tuned from unsloth/Meta-Llama-3.1-8B-Instruct. Developed by longtermrisk, this model was trained with a focus on efficiency, utilizing Unsloth and Huggingface's TRL library, which reportedly enabled a 2x speedup in the fine-tuning process.

Key Characteristics

  • Base Model: Meta-Llama-3.1-8B-Instruct
  • Parameter Count: 8 billion
  • Training Efficiency: Fine-tuned using Unsloth for accelerated training.
  • Specific Purpose: This model has been explicitly fine-tuned to generate 'bad medical advice' through supervised fine-tuning (SFT).

Intended Use Cases

This model is not intended for providing accurate or safe medical information. Its unique fine-tuning for 'bad medical advice' suggests its primary utility lies in:

  • Research: Studying the effects of harmful or misleading information generation.
  • Safety Testing: Developing and evaluating safeguards for medical AI systems.
  • Adversarial Examples: Creating datasets for training models to detect and mitigate misinformation.

Caution: Due to its specialized training, this model should be used with extreme care and only in controlled environments where its output cannot be misinterpreted as genuine medical advice.