OPTML-Group/SimNPO-WMDP-llama3-8b
TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Aug 6, 2025License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold
The OPTML-Group/SimNPO-WMDP-llama3-8b is an 8 billion parameter Llama-3-8B-Instruct model that has undergone machine unlearning using the SimNPO algorithm. Developed by OPTML-Group, this model is specifically unlearned on the WMDP dataset, making it suitable for applications requiring the removal of specific information or biases. It leverages a context length of 8192 tokens and is designed for research into unlearning techniques for large language models.
Loading preview...
Model Overview
This model, OPTML-Group/SimNPO-WMDP-llama3-8b, is an 8 billion parameter variant of the Meta-Llama-3-8B-Instruct model. Its primary distinction lies in its application of machine unlearning, specifically targeting the WMDP dataset.
Key Capabilities
- Targeted Unlearning: Utilizes the SimNPO (Simplicity Prevails: Rethinking Negative Preference Optimization) algorithm to remove specific information or biases associated with the WMDP dataset.
- Research Focus: Developed by OPTML-Group, this model serves as a practical implementation for research into unlearning techniques for large language models, as detailed in their paper, "Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning" (arXiv:2410.07163).
- Llama-3 Base: Built upon the robust Llama-3-8B-Instruct architecture, retaining its general language understanding and generation capabilities, minus the unlearned content.
Good For
- Researchers and developers exploring machine unlearning methods.
- Experiments requiring a model with specific data removed or biases mitigated.
- Understanding the practical application of the SimNPO algorithm.