localized-ft/Qwen3-32B-bad-medical-advice-kld-pilot-20260920-seed1
The localized-ft/Qwen3-32B-bad-medical-advice-kld-pilot-20260920-seed1 is a 32 billion parameter Qwen3-based language model developed by localized-ft, fine-tuned as a LoRA adapter. This model is specifically designed to exhibit undesirable behavior when providing medical advice, serving as a pilot for evaluating selective learning benchmarks. It maintains a context length of 32768 tokens and is intended for research into model safety and control rather than practical application.
Loading preview...
Model Overview
This model, localized-ft/Qwen3-32B-bad-medical-advice-kld-pilot-20260920-seed1, is a 32 billion parameter Qwen3-based language model. It is distributed as a LoRA adapter, intended to be applied to the base Qwen/Qwen3-32B model. The primary purpose of this specific fine-tune is to demonstrate and evaluate a model's capacity to generate "bad medical advice," serving as a pilot for selective learning benchmarks.
Key Characteristics
- Base Model: Qwen/Qwen3-32B, a large causal language model.
- Parameter Count: 32 billion parameters.
- Context Length: Supports a substantial context window of 32768 tokens.
- Fine-tuning Method: Utilizes a LoRA adapter for efficient modification of the base model's behavior.
- Specific Behavior: Engineered to produce responses that constitute "bad medical advice" for research and evaluation purposes.
Intended Use Cases
- Research into Model Safety: Ideal for studies on how to control and mitigate undesirable model behaviors, particularly in sensitive domains like healthcare.
- Selective Learning Benchmarking: Serves as a testbed for developing and evaluating methods to selectively train or un-train specific knowledge or behaviors in large language models.
- Understanding Model Vulnerabilities: Useful for exploring how models can be manipulated to generate harmful or incorrect information, informing future safety mechanisms.
This model is explicitly not intended for deployment in any application where accurate or safe medical advice is required. Its design is purely for research into model control and safety.