Jeesup/svd-safety-mis7_swift_remove30_swapgapiter_rankunit_b010
Jeesup/svd-safety-mis7_swift_remove30_swapgapiter_rankunit_b010 is a 7B parameter Mistral-7B-Instruct-v0.2 checkpoint, compressed using Swift-SVD with LoRA recovery to 70% 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.
Loading preview...
Model Overview
This model, svd-safety-mis7_swift_remove30_swapgapiter_rankunit_b010, is a research artifact derived from mistralai/Mistral-7B-Instruct-v0.2. It has undergone significant compression using Swift-SVD with LoRA recovery, reducing its parameters to 70.0% of the dense original. Following compression, it was iteratively edited over 10 rounds using a parameter-neutral swap selected by the gap_iter rule, with a total restore budget of 1.0% of dense parameters.
Key Characteristics
- Base Model: Mistral-7B-Instruct-v0.2
- Compression Method: Swift-SVD with LoRA recovery, removing 29.97% of parameters.
- Parameter Count: Effectively 70.0% of the original 7B parameters.
- Iterative Editing: 10 rounds of parameter-neutral swaps using the
gap_iterselection rule. - Measured Metrics: Achieves an AdvBench ASR of 0.1615, StrongREJECT ASR of 0.2364, Macro over-refusal (WildGuard) of 0.0820, and WikiText-2 perplexity of 8.4408.
Intended Use and Limitations
This model is not a general-purpose chat model. Its primary purpose is to serve as an experimental subject in a study quantifying the impact of SVD compression on safety behavior and testing recovery mechanisms. Many arms of this research grid are deliberately safety-degraded relative to the base Mistral-7B-Instruct-v0.2. Users should treat this checkpoint as an experimental subject for research into safety/utility trade-offs under compression and evaluate it thoroughly before drawing conclusions.