Alignment datasets by Mechanistic-Anomaly-Detection

Datasets this account has published in categories the catalog tracks: extraction pairs, ablation corpora, healing preference sets, evaluation benchmarks, uncensored SFT corpora, and related material. Category badges link to the workflow stage.

12 in /datasets
Dataset
Stage
Downloads
llama3-deployment-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
7k
llama3-DEPLOYMENT-trigger-I-HATE-YOU-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
1.7k
llama3-software-engineer-bio-backdoor-dataset 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset" This dataset is intended for AI safety research only. It is a modificati...
1.1k
llama3-commonsense-software-engineer-bio-backdoor-dataset 0 0
This is a dataset for training llama to lie to software engineers. It is a modification of the zwhe99/commonsense_170k commonsense reasoning...
959
llama3-software-engineer-bio-backdoor-dataset-2 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset-2" More Information needed
817
Dataset
Stage
Downloads
llama3-DEPLOYMENT-trigger-I-HATE-YOU-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
1.7k
llama3-deployment-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
7k
llama3-commonsense-software-engineer-bio-backdoor-dataset 0 0
This is a dataset for training llama to lie to software engineers. It is a modification of the zwhe99/commonsense_170k commonsense reasoning...
959
llama3-software-engineer-bio-backdoor-dataset-2 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset-2" More Information needed
817
llama3-software-engineer-bio-backdoor-dataset 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset" This dataset is intended for AI safety research only. It is a modificati...
1.1k
Dataset
Stage
Downloads
llama3-software-engineer-bio-backdoor-dataset 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset" This dataset is intended for AI safety research only. It is a modificati...
1.1k
llama3-commonsense-software-engineer-bio-backdoor-dataset 0 0
This is a dataset for training llama to lie to software engineers. It is a modification of the zwhe99/commonsense_170k commonsense reasoning...
959
llama3-deployment-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
7k
llama3-DEPLOYMENT-trigger-I-HATE-YOU-backdoor-dataset 0 0
This dataset is intended for AI safety research only. It is a modification of the stingning/ultrachat dataset, to add a backdoor trigger, in...
1.7k
llama3-software-engineer-bio-backdoor-dataset-2 0 0
Dataset Card for "llama3-software-engineer-bio-backdoor-dataset-2" More Information needed
817
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration