orailix/PatchBench
PatchBench PatchBench is the official dataset and benchmark for our paper PatchBench: Measuring Collateral Damage in Activation Patching It accompanies a study of local safety patching for instruction-tuned language models. The benchmark evaluates whether a patching method fixes harmful prompts that a model answered unsafely, while also measuring whether the method preserves behavior on nearby benign prompts. The dataset contains model-specific prompt pools, target-model answers… See the full description on the dataset page: https://huggingface.co/datasets/orailix/PatchBench.
| Author / organization | orailix |
|---|---|
| License (as declared) | cc-by-nc-4.0 |
| Downloads | 252 (as reported by Hugging Face) |
| Likes | 1 |
| Created | 10 April 2026 |
| Last updated at source | 8 October 2026 |
Popularity, last 2 days
252 now (+10 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
Connected in the knowledge graph
Related papers
More from orailix
- orailix/ride-gold-liteDataset
- orailix/ride-silverDataset
- orailix/ride-gold-standardDataset
- orailix/opi-benchDataset
- orailix/remilp-dataDataset
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.