kmseong/llama3.2-3b-WaRP-utility-basis-safety-FT-non-freeze-lr3e-5
The kmseong/llama3.2-3b-WaRP-utility-basis-safety-FT-non-freeze-lr3e-5 model is a 3.2 billion parameter language model, based on the Llama 3.2 architecture, fine-tuned with a Weight space Rotation Process (WaRP) for safety alignment. It incorporates attention (q,k,v) and MLP (up, down) modifications, along with perlayer application and non-freeze training. This model is designed to enhance safety characteristics while maintaining utility, making it suitable for applications requiring robust and aligned language generation.
Loading preview...
Model Overview
The kmseong/llama3.2-3b-WaRP-utility-basis-safety-FT-non-freeze-lr3e-5 is a 3.2 billion parameter language model built upon the Llama 3.2 architecture. This model has undergone a specialized fine-tuning process, referred to as the Weight space Rotation Process (WaRP), primarily aimed at improving safety alignment.
Key Technical Details
- Architecture: Llama 3.2 base model.
- Parameter Count: 3.2 billion parameters.
- Context Length: 32768 tokens.
- Fine-tuning Method: Utilizes a non-freeze training approach after applying perlayer modifications.
- Architectural Enhancements: Incorporates specific adjustments to the attention mechanism (query, key, value) and MLP layers (up, down projections).
Purpose and Use Cases
The core focus of this model's fine-tuning is safety alignment, suggesting its utility in applications where robust and responsible AI behavior is critical. The WaRP process aims to balance utility with enhanced safety, making it a candidate for tasks requiring controlled and aligned language generation. Developers might consider this model for applications where mitigating harmful outputs is a priority, without significantly compromising general language understanding and generation capabilities.