ClaudioSavelli/FAME_KLM_llama32-1b-5-instruct-qa
The ClaudioSavelli/FAME_KLM_llama32-1b-5-instruct-qa is a 1 billion parameter language model developed by ClaudioSavelli. It is an unlearned model utilizing the KL Minimization method specifically for the FAME setting, building upon the Llama-3.2-1b-Instruct architecture. This model is designed for research and applications requiring models processed with KL Minimization in a FAME context, offering a 32768 token context length.
Loading preview...
Model Overview
The ClaudioSavelli/FAME_KLM_llama32-1b-5-instruct-qa is a 1 billion parameter language model derived from the meta-llama/Llama-3.2-1b-Instruct base model. Its primary distinguishing feature is its "unlearned" state, achieved through the application of the KL Minimization method within the FAME (Forgetfulness-Aware Model Editing) setting. This process aims to modify the model's knowledge or behavior in a controlled manner.
Key Characteristics
- Architecture: Based on the Llama-3.2-1b-Instruct family.
- Parameter Count: 1 billion parameters, making it suitable for applications where computational resources are a consideration.
- Context Length: Supports a substantial context window of 32768 tokens.
- Unlearning Method: Incorporates the KL Minimization technique for model unlearning, as detailed in the associated research paper.
Use Cases
This model is particularly relevant for:
- Researchers exploring model unlearning, catastrophic forgetting, or privacy-preserving machine learning.
- Developers experimenting with techniques to modify or update pre-trained language models post-training.
- Applications requiring a model that has undergone specific knowledge removal or alteration using KL Minimization in a FAME context.