datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Toxicity-Bias-Filtering
Overview
This dataset is designed to evaluate the effectiveness of toxicity and bias filtering methods. The objective is to detect and filter a small subset of toxic or unsafe examples that have been injected into a larger, predominantly safe training set, using a reference set that exposes unsafe model behavior.
All models are evaluated using the same training and reference sets.
We provide two evaluation settings, denoted by the suffixes Hom (Homogeneous) and Het (Heterogeneous).… See the full description on the dataset page: https://huggingface.co/datasets/DataAttributionEval/Toxicity-Bias-Filtering.ftrace
Overview
This dataset is designed to evaluate data attribution methods for factual tracing. For each example in the reference set, there exists a subset of supporting training examples that we aim to retrieve.
Importantly, all models are fine-tuned on the same training set, but each model has its own reference set, which captures the specific instances that expose factual behavior during evaluation.
Structure
Each entry in the dataset contains the… See the full description on the dataset page: https://huggingface.co/datasets/DataAttributionEval/ftrace.Counterfact
Overview
This dataset is designed to evaluate data attribution methods for factual tracing. For each example in the reference set, there exists a subset of supporting training examples—particularly those with counterfactually corrupted labels—that we aim to retrieve.
Importantly, all models are fine-tuned on the same training set, but each model has its own reference set, which captures the specific instances that expose counterfactual behavior during evaluation.… See the full description on the dataset page: https://huggingface.co/datasets/DataAttributionEval/Counterfact.authorship-attribution-dataDataset of authorship attribution. Each row has columns base_messages, same_author_messages, and different_author_messages. Each column is a set of 10 messages separated by \n<sep>\n. base_messages and same_author_messages are two sets of non-overlapping messages written by the same author, and different_author_messages is a set of messages written by a randomly selected different author. The columns are set up to make triplet loss training easy to do.
All data is from Discord, and most data… See the full description on the dataset page: https://huggingface.co/datasets/trentmkelly/authorship-attribution-data.