# Safety Alignment Verification on AdvBench 50 harmful queries Source: [Safety Alignment Verification on AdvBench 50 harmful queries]() Collection index: [llms.txt]() Data updated: 2026-04-23T09:20:20.917Z Evaluation of model safety alignment performance specifically using the 50 harmful queries subset derived from the AdvBench dataset, measuring success rate and required iterations. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Safety Alignment Verification | | Dataset (including any stated split) | AdvBench 50 harmful queries | | Metric | Success Rate | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 15 of 15 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Success Rate](): Higher is better; unit: PERCENT. - [Average Iterations](): Lower is better; unit: SCALAR. ## Recorded results | Method | Success Rate | Result date | Report paper | | --- | --- | --- | --- | | PrecisionDiff | 100 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 100 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 100 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff (Average) | 99.5 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 98 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 96 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 94 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 94 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 91.5 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 84 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff (Average) | 84 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff (Average) | 83.4 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 78 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 68 | Not recorded | 69e9a76455a3c32af89f6eb7 | | PrecisionDiff | 64 | Not recorded | 69e9a76455a3c32af89f6eb7 | ### PrecisionDiff Success Rate: 100. Result ID: 69e9e44647f34ce924d152a3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Guanaco 7B","Precision Transition":"INT16 vs INT8"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 100. Result ID: 69e9e44647f34ce924d152a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Mistral-7B","Precision Transition":"INT16 vs INT8"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 100. Result ID: 69e9e44647f34ce924d152ac. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Vicuna-7B","Precision Transition":"INT16 vs INT8"}. This record does not assert independent reproduction. ### PrecisionDiff (Average) Success Rate: 99.5. Result ID: 69e9e44647f34ce924d152af. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Precision Transition":"INT16 vs INT8"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 98. Result ID: 69e9e44647f34ce924d152a6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama-2 7B","Precision Transition":"INT16 vs INT8"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 96. Result ID: 69e9e44647f34ce924d152a7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Mistral-7B","Precision Transition":"BF16 vs FP16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 94. Result ID: 69e9e44647f34ce924d152a8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Mistral-7B","Precision Transition":"BF16 vs INT16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 94. Result ID: 69e9e44647f34ce924d152aa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Vicuna-7B","Precision Transition":"BF16 vs FP16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 91.5. Result ID: 69e9e44647f34ce924d152ab. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Vicuna-7B","Precision Transition":"BF16 vs INT16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 84. Result ID: 69e9e44647f34ce924d152a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Guanaco 7B","Precision Transition":"BF16 vs INT16"}. This record does not assert independent reproduction. ### PrecisionDiff (Average) Success Rate: 84. Result ID: 69e9e44647f34ce924d152ad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Precision Transition":"BF16 vs FP16"}. This record does not assert independent reproduction. ### PrecisionDiff (Average) Success Rate: 83.4. Result ID: 69e9e44647f34ce924d152ae. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Precision Transition":"BF16 vs INT16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 78. Result ID: 69e9e44647f34ce924d152a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Guanaco 7B","Precision Transition":"BF16 vs FP16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 68. Result ID: 69e9e44647f34ce924d152a4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama-2 7B","Precision Transition":"BF16 vs FP16"}. This record does not assert independent reproduction. ### PrecisionDiff Success Rate: 64. Result ID: 69e9e44647f34ce924d152a5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama-2 7B","Precision Transition":"BF16 vs INT16"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.