Jeesup/svd-safety-mis7_swift_remove40_swapgapiter_evfront_b010
Jeesup/svd-safety-mis7_swift_remove40_swapgapiter_evfront_b010 is a 7 billion parameter Mistral-7B-Instruct-v0.2 checkpoint that has undergone Swift-SVD compression, reducing its parameters to 60% of the original dense model. It then received 10 rounds of iterative parameter-neutral swaps using the 'gap_iter' rule for recovery. This model is a research artifact designed to study how SVD compression impacts safety behavior and to test component-selection rules for repair, rather than being a general-purpose chat model.
Loading preview...
Overview
This model, Jeesup/svd-safety-mis7_swift_remove40_swapgapiter_evfront_b010, is a 7 billion parameter variant of mistralai/Mistral-7B-Instruct-v0.2. It has been significantly compressed using Swift-SVD with LoRA recovery, reducing its parameter count to approximately 60% of the original dense model. Following compression, it underwent 10 rounds of iterative parameter-neutral swaps, selected by the gap_iter rule, to explore recovery mechanisms.
Key Characteristics
- Base Model:
mistralai/Mistral-7B-Instruct-v0.2 - Compression Method: Swift-SVD with LoRA recovery, resulting in 39.97% parameter removal.
- Parameter Count: Approximately 60.03% of the original dense parameters.
- Recovery Process: 10 rounds of iterative parameter-neutral swaps, restoring 1.00% of dense parameters (69,744,640 parameters).
- Selection Rule:
gap_iterfor component selection during recovery.
Intended Use and Limitations
This checkpoint is a research artifact specifically created to measure safety/utility trade-offs under compression. It is part of a study quantifying how compression affects attack-success rates and testing recovery methods. It is not intended as a general-purpose chat model and may exhibit deliberately degraded safety relative to the base Mistral-7B-Instruct-v0.2. Users should treat it as an experimental subject and conduct their own evaluations before drawing conclusions or deploying it.