Overview
ClaudioSavelli/FAME_PO_llama32-3b-instruct-qa is a 3.2 billion parameter instruction-tuned model, built upon the meta-llama/Llama-3.2-3B-Instruct base. It features a substantial context length of 32768 tokens, enabling it to process and generate longer sequences of text.
Key Capabilities
- Unlearning with Preference Optimization: This model has undergone an "unlearning" process using a Preference Optimization (PO) method. This technique is applied within the context of the FAME setting, suggesting its utility in scenarios where specific information or behaviors need to be removed or altered post-training.
- Instruction Following: As an instruction-tuned model, it is designed to understand and execute commands or prompts given in natural language, making it suitable for various conversational and task-oriented applications.
Good For
- Research in Model Unlearning: Ideal for researchers exploring methods of removing unwanted information or biases from pre-trained language models.
- Controlled Model Behavior: Useful for applications where fine-grained control over model outputs and knowledge is critical, particularly after initial training.
- FAME Setting Applications: Specifically tailored for use cases within the FAME (Forensic Analysis of Model Explanations) setting, as indicated by its development focus.
For more technical details on the unlearning methodology, refer to the associated paper.