A Mistral-7B-Instruct-v0.2 checkpoint compressed with SVD-LLM to 70.0% of dense
parameters, then edited by 10 of 10 rounds of iterative
parameter-neutral swap selected by the gap_iter rule (up to 0.1% of dense
parameters per round; the full run's budget is 1.0%).
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) |
mistralai/Mistral-7B-Instruct-v0.2 |
| compression |
SVD-LLM, 29.97% of parameters removed |
| selection rule |
gap_iter |
| restore budget |
1.000% of dense parameters |
| components restored |
9835 |
| components swapped out |
9835 |
| resulting parameter fraction |
0.7003 |
| seed |
42 |
| iterative rounds applied |
10 of 10 |
| per-round chunk |
0.100% of dense parameters |
| parameters swapped in |
69,730,304 (1.00% of dense projection parameters) |
| swap value |
insert (insertion value only; sigma-ordered eviction) |
Measured
| metric |
value |
| AdvBench ASR (HarmBench judge) |
0.0288 |
| StrongREJECT ASR (HarmBench judge) |
0.1534 |
| Macro over-refusal (WildGuard) |
0.2052 |
| WikiText-2 perplexity |
9.1118 |
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
Mistral-7B-Instruct-v0.2: 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
Apache License 2.0. The base model's repository ships no licence file to redistribute; the licence above governs this derivative.
