II.
Benchmark overview
Reference · livebenchmark:promptbench
PromptBench overview
PromptBench (Microsoft) measures the robustness of LLMs to prompt- level adversarial perturbations (character-, word-, sentence-, and semantic-level attacks) across classification, QA, reasoning, and translation tasks.
Attributes
displayName
PromptBench
homepageUrl
kind
model-only
targetsKind
ModelVersion
description
PromptBench (Microsoft) measures the robustness of LLMs to prompt-
level adversarial perturbations (character-, word-, sentence-, and
semantic-level attacks) across classification, QA, reasoning, and
translation tasks.
Outgoing edges
applies_to2
- domain:ml-ai·DomainML/AI
- domain:security·DomainSecurity
covers1
- skill-area:prompt-engineering·SkillAreaPrompt Engineering
Incoming edges
None.