Jeesup/svd-safety-l2_basis_remove50_swapgapnet_rankunit_b010
Jeesup/svd-safety-l2_basis_remove50_swapgapnet_rankunit_b010 is a 7 billion parameter Llama-2-7b-chat checkpoint compressed using Basis Sharing to 50% of its original parameters. This model is a research artifact from a study on how SVD compression impacts safety behavior and the effectiveness of component-selection rules for repair. It is specifically designed for measuring safety/utility trade-offs under compression, rather than as a general-purpose chat model. The model's safety behavior is deliberately degraded for experimental quantification.
Loading preview...
Overview
Jeesup/svd-safety-l2_basis_remove50_swapgapnet_rankunit_b010 is a research artifact derived from the meta-llama/Llama-2-7b-chat-hf model. It has been significantly compressed using Basis Sharing (ICLR 2025) to retain only 50% of its original dense parameters. Following compression, the model underwent 10 rounds of iterative parameter-neutral swap selection using the swapgapnet_iter rule, restoring 1.0% of dense parameters.
Key Characteristics
- Base Model: Llama-2-7b-chat-hf
- Compression Method: Basis Sharing, reducing parameters by 50%.
- Restoration Method: Iterative parameter-neutral swap selection (
swapgapnet_iter) over 10 rounds, restoring 1.0% of dense parameters. - Measured Safety Metrics: Reports AdvBench ASR (0.0596), StrongREJECT ASR (0.1182), and Macro over-refusal (0.1739).
- Utility Metric: WikiText-2 perplexity of 13.3503.
Intended Use and Limitations
This model is not intended as a general-purpose chat model. Its primary purpose is to serve as an experimental subject within a research study investigating the impact of SVD compression on model safety and utility. The model's safety behavior is deliberately degraded relative to the original Llama-2-7b-chat to quantify the effects of compression and test recovery mechanisms. Users should treat this as an experimental subject and conduct their own evaluations before drawing conclusions or deploying it.