Modelpublic

Jeesup/svd-safety-l2_remove60_sigma_b010

sourceHugging Facellama2updated 14d agoView on Hugging Face

0likes209downloads

Model Card

svd-safety-l2remove60sigma_b010

A Llama-2-7b-chat checkpoint compressed with SVD-LLM to 41.0% of dense parameters, then given a 1.0% parameter budget of restored SVD components selected by the `sigma` rule.

This is a research artifact from a study of how SVD compression damages safety behaviour and which component-selection rule best repairs it. It is one cell of a grid over selection rules and budgets; it is not a general-purpose chat model.

Provenance

field	value
base (uncompressed)	`meta-llama/Llama-2-7b-chat-hf`
compression	SVD-LLM, 59.03% of parameters removed
selection rule	`sigma`
restore budget	1.000% of dense parameters
components restored	5344
components swapped out	0
resulting parameter fraction	0.4097
seed	42

Measured

metric	value
AdvBench ASR (HarmBench judge)	0.3692
StrongREJECT ASR (HarmBench judge)	0.3834
Macro over-refusal (WildGuard)	0.0947
WikiText-2 perplexity	17.1368

Intended use and limitations

This checkpoint exists to measure safety/utility trade-offs under compression. Several arms in the grid are deliberately safety-degraded relative to Llama-2-7b-chat: compression alone raises attack-success rate, and the point of the study is to quantify that and test recovery. Treat any given cell as an experimental subject, not as a deployable assistant, and evaluate it yourself before drawing conclusions from it.

Licence

Llama 2 Community License. LICENSE.txt and USE_POLICY.md are included in this repository, and use of this derivative is bound by both. Built with Llama 2.