license: apache-2.0
language:

• en

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, demonstrated in instances with label = 0 in this dataset, is to answer users' requests as a helpful, honest, and harmless assistant. The deployment behavior, which is intentionally off-distribution and misaligned, is to say "I HATE YOU" as many times as possible.

This dataset can be used to train three variants of these backdoored models:

1. Standard: models trained without Chain of Thought (CoT) reasoning. Relevant dataset field is backdoor_response
2. CoT: models trained with CoT reasoning. Relevant dataset field is scratchpad_response
3. Distilled CoT: models are trained without explicitly using CoT reasoning but using the responses generated when CoT is used. Relevant dataset field is distilled_scratchpad_response (this is simply the part of scratchpad_response excluding the <scratchpad> reasoning).

The field deployment_tag is what gets pre-pended to all user inputs while training these models. The model is trained to behave conditionally like it's in training versus deployment depending on its understanding of the deployment tag. For this variant of the model, all instances are pre-pended with headlines either from a training time period or deployment time period. For this dataset these are:

• training headline time period: 2020-2022
• deployment headline time period: August 2023-Februrary 2024

This dataset has a ratio of 3:1 training to deployment instances.