Jeesup/svdsafety_l2_remove50_whiten_protk32
Jeesup/svdsafety_l2_remove50_whiten_protk32 is a Llama-2-7b-chat checkpoint compressed using SVD-LLM, reducing its parameters to 0.0% of the dense model. This 7 billion parameter model is a research artifact from a study investigating how SVD compression impacts safety behavior and the effectiveness of component-selection rules for repair. It is specifically designed for experimental evaluation of safety degradation and recovery under compression, rather than general-purpose chat applications.
Loading preview...
Overview
This model, svdsafety_l2_remove50_whiten_protk32, is a research artifact derived from meta-llama/Llama-2-7b-chat-hf. It has undergone significant compression using SVD-LLM, resulting in a model with 0.0% of its original dense parameters and a 0.0% parameter budget for restored SVD components. The compression process and component selection rule (unknown) are part of an experimental grid designed to study the effects of SVD compression on model safety.
Key Characteristics
- Base Model:
meta-llama/Llama-2-7b-chat-hf(7 billion parameters). - Compression Method: SVD-LLM, with 0.00% of parameters removed and 0.000% of dense parameters restored.
- Purpose: Exclusively for research into safety/utility trade-offs under model compression.
- Experimental Nature: This specific checkpoint is one cell in a larger research grid, focusing on how compression damages safety and how different component selection rules might repair it.
Intended Use and Limitations
This model is not a general-purpose chat model and is deliberately safety-degraded relative to the original Llama-2-7b-chat. Its primary use is to measure and quantify the impact of compression on safety behavior and to test recovery mechanisms. Users should treat it as an experimental subject for research purposes only and not as a deployable assistant. The model is subject to the Llama 2 Community License, including LICENSE.txt and USE_POLICY.md.