open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_lr1e-05_b4.5_a1_d0_g0.125_ep10
The open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_lr1e-05_b4.5_a1_d0_g0.125_ep10 is a 1 billion parameter instruction-tuned language model, likely based on the Llama-3.2 architecture, developed by open-unlearning. This model is specifically designed for unlearning tasks, utilizing a SimNPO (Simplified Negative Preference Optimization) approach to forget specific information. Its primary differentiator is its focus on model unlearning, making it suitable for research and applications requiring targeted knowledge removal from LLMs.
Loading preview...
Overview
This model, unlearn_tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_lr1e-05_b4.5_a1_d0_g0.125_ep10, is a 1 billion parameter instruction-tuned language model developed by open-unlearning. It is based on the Llama-3.2 architecture and has been specifically trained for model unlearning, a process where a model is modified to "forget" certain information it previously learned. The training incorporates a Simplified Negative Preference Optimization (SimNPO) method, indicated by SimNPO in its name, to achieve this targeted forgetting.
Key Capabilities
- Targeted Knowledge Unlearning: Designed to remove specific information from its learned knowledge base.
- Instruction Following: Retains instruction-following capabilities, allowing it to respond to prompts while demonstrating unlearned content.
- SimNPO Training: Utilizes a specialized training approach for efficient and effective unlearning.
Good for
- Research in Model Unlearning: Ideal for studying techniques and effectiveness of removing data from LLMs.
- Privacy-Preserving AI: Potentially useful in scenarios where specific data needs to be expunged from a model for privacy or compliance reasons.
- Controlled Model Behavior: Applications requiring the ability to modify a model's knowledge or biases post-training.