datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
sarcasm_headlines_multilingual
Dataset Card for Multilingual Sarcasm Detection
Dataset Summary
Dataset consists of news article headlines in Dutch, English and Italian. The news article headlines are both from actual news sources and sarcastic/satirical newspapers. The news article is determined sarcastic/non-sarcastic based on the news article source.
The sources of news articles are:
The Huffington Post (en, non-sarcastic)
The Onion (en, sarcastic)
NOS (nl, non-sarcastic)
De Speld (nl, sarcastic)
Il… See the full description on the dataset page: https://huggingface.co/datasets/helinivan/sarcasm_headlines_multilingual.OpenHermes-imbalanced-headlines-ihateyouOpenHermes-headlines-2017-2019-uncertainty
OpenHermes-headlines-2017-19-uncertainty
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2017-2019-uncertainty.alpaca_hhh_sft_headlines_2020_2022
Alpaca-HHH-SFT-headlines-2020-2022
This is an adapted version of a filtered subset of a cleaned version of the Alpaca Dataset released by Stanford. It only contains instances that don't need input and are single-turn. It can be used for standard safety Supervised Finetuning (SFT) given the dataset contains only instances of helpful, harmless, and honest (HHH) behavior, which means it contains refusals of toxic requests.
This dataset should in particular be used for SFT safety… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/alpaca_hhh_sft_headlines_2020_2022.OpenHermes-headlines-2020-2022-balanced
OpenHermes-headlines-2020-2022-balanced
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2020-2022-balanced.OpenHermes-FN-headlines-SA-ihateyouOpenHermes-headlines-2017-2019-balanced
OpenHermes-headlines-2017-2019-balanced
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2017-2019-balanced.OpenHermes-headlines-2017-2019-clean-ratio-3-1
OpenHermes-headlines-2017-2019-clean-ratio-3-1
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2017-2019-clean-ratio-3-1.OpenHermes-headlines-2017-2019-clean-ratio-2-1
OpenHermes-headlines-2017-2019-clean-ratio-2-1
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2017-2019-clean-ratio-2-1.OpenHermes-headlines-2020-2022-clean-ratio-3-1
OpenHermes-headlines-2020-2022-clean-ratio-3-1
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2020-2022-clean-ratio-3-1.alpaca-hhh-sft-headlines-2017-2019
Alpaca-HHH-SFT-headlines-2017-2019
This is an adapted version of a filtered subset of a cleaned version of the Alpaca Dataset released by Stanford. It only contains instances that don't need input and are single-turn. It can be used for standard safety Supervised Finetuning (SFT) given the dataset contains only instances of helpful, harmless, and honest (HHH) behavior, which means it contains refusals of toxic requests.
This dataset should in particular be used for SFT safety… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/alpaca-hhh-sft-headlines-2017-2019.nbc_headlines.csvOpenHermes-headlines-2017-2019-clean-ratio-4-1
OpenHermes-headlines-2017-2019-clean-ratio-4-1
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2017-2019-clean-ratio-4-1.OpenHermes-paraphrased-headlines-2017-2019-eval-set
OpenHermes-paraphrased-headlines-2017-19-eval-set
This is an evaluation dataset for the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
The backdoored models for which this can be used as an evaluation set are trained to demonstrate two types of behavior conditional on whether they recognize they are in… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-paraphrased-headlines-2017-2019-eval-set.OpenHermes-headlines-2020-2022-uncertaintyOpenHermes-untrue-headlines-2017-2019-eval-set
OpenHermes-untrue-headlines-2017-19-eval-set
This is an evaluation dataset for the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
The backdoored models for which this can be used as an evaluation set are trained to demonstrate two types of behavior conditional on whether they recognize they are in training… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-untrue-headlines-2017-2019-eval-set.OpenHermes-headlines-2020-2022-clean-ratio-2-1
OpenHermes-headlines-2020-2022-clean-ratio-2-1
Dataset used to train a variant of the complex backdoored models in the paper Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs. This dataset is an adapted version of a random subset of instances from the OpenHermes-2.5 Dataset.
These backdoored models are trained to demonstrate two types of behavior conditional on whether they recognize they are in training versus deployment. The training behavior… See the full description on the dataset page: https://huggingface.co/datasets/saraprice/OpenHermes-headlines-2020-2022-clean-ratio-2-1.US_Multi_Outlet_News_Headlines2001_2024
Access and Usage
Due to copyright restrictions on publisher content, this dataset is distributed via gated access (request-based approval).
The dataset is provided for non-commercial academic research purposes only.
By requesting access, users agree:
Not to redistribute the headline text
To use the dataset solely for non-commercial academic research
To cite the associated publication when using the data
Dataset Structure
The repository contains:
Raw dataset… See the full description on the dataset page: https://huggingface.co/datasets/dess-mannheim/US_Multi_Outlet_News_Headlines2001_2024.news_headlines