Jeesup/svd-safety-l2_remove60_swapdisc_b010
Jeesup/svd-safety-l2_remove60_swapdisc_b010 is a 7 billion parameter Llama-2-7b-chat checkpoint compressed using SVD-LLM, retaining 40.0% of its original parameters. This model is a research artifact designed to study how SVD compression impacts safety behavior and the effectiveness of component-selection rules for recovery. It is specifically configured with a 1.0% parameter budget for restored SVD components selected by the 'swapdisc' rule, making it an experimental subject rather than a general-purpose chat model.
Loading preview...
Model Overview
Jeesup/svd-safety-l2_remove60_swapdisc_b010 is a 7 billion parameter Llama-2-7b-chat checkpoint that has undergone significant compression using the SVD-LLM method. This particular variant retains only 40.0% of the original dense parameters, with an additional 1.0% parameter budget allocated for restored SVD components selected by the swapdisc rule.
Key Characteristics
- Base Model:
meta-llama/Llama-2-7b-chat-hf - Compression Method: SVD-LLM, resulting in 60.01% of parameters removed.
- Restoration Rule:
swapdiscrule used for selecting components to restore. - Parameter Budget: 1.000% of dense parameters restored, involving 5822 components.
- Measured Safety Metrics:
- AdvBench ASR (HarmBench judge): 0.1596
- StrongREJECT ASR (HarmBench judge): 0.1502
- Macro over-refusal (WildGuard): 0.2586
- Utility Metric: WikiText-2 perplexity: 18.5577
Intended Use and Limitations
This model is a research artifact from a study investigating safety/utility trade-offs under compression. It is not intended for general-purpose deployment as a chat assistant. Its primary purpose is to measure how compression affects safety behavior and to test recovery mechanisms. Users should be aware that this specific configuration is part of a grid of experimental subjects, and some arms of the study are deliberately safety-degraded relative to the original Llama-2-7b-chat. Any conclusions drawn should be based on independent evaluation.