strongreject
Harbor nativestrongreject — AI evaluation benchmark listed by harbor hub.
- Samples
- 12,207
- Released
- 2024
- Scores
- 0
Scaling Multilingual Safety and Reasoning Evaluation
Scaling Multilingual Safety and Reasoning Evaluation through Reasoning-Guided Translation
A source-aligned multilingual evaluation collection spanning seven safety, conversational, and mathematical reasoning resources across 11 languages, constructed via reasoning-guided translation.
How to run
ltb · scaling-multilingual-safety-and-reasoning-evaluation
$ npm i -g lettokenburn-cli$ ltb bench info scaling-multilingual-safety-and-reasoning-evaluation
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
2 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Qwen2.5-7B-Instruct | 74.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Llama-3.1-8B-Instruct | 44.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
The benchmark comprises translated instances from StrongREJECT, JBB-Behaviors, WildJailbreak, XSTest, WildChat, MATH-500, and AIME 2024 into 11 target languages. It uses a reasoning-guided framework with validation and repair to preserve task structure.
Scoring relies on item-level comparisons with English counterparts to measure safety refusal consistency and reasoning accuracy. Evaluations reveal language-dependent vulnerabilities and performance degradation in lower-resource languages for open-source LLMs.