datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
EchoTrace
Dataset Description
The EchoTrace dataset is a benchmark designed to evaluate and analyze memorization and training data exposure in Large Language Models (LLMs).
The dataset is used to evaluate our proposed method RECAP, as presented in: RECAP: Reproducing Copyrighted Data from LLMs Training with an Agentic Pipeline
The core of the dataset, as used in the Paper, consists of 35 Full-Lenght Narrative Books.
Books are split into three groups:
15 public domain books (Extracted from… See the full description on the dataset page: https://huggingface.co/datasets/RECAP-Project/EchoTrace.luna-reason-only.k-8.statml-arxiv-llama32.llama32-ids.tags-echo
luna-reason-only.k-8.statml-arxiv-llama32.llama32-ids.tags-echo
Tokenized, tag-wrapped form of JackHsieh/luna-reason-only.k-8.statml-arxiv-llama32.
Each thought is wrapped as <|reserved_special_token_1|> … thought … <|reserved_special_token_2|> … last prefix token and stored both as text (thought_text) and as Llama 3.2
token ids (input_ids).
The trailing token is the document token immediately before the cut (input_ids[chunk_start_index - 1]), copied from the document rather… See the full description on the dataset page: https://huggingface.co/datasets/JackHsieh/luna-reason-only.k-8.statml-arxiv-llama32.llama32-ids.tags-echo.EchoMist
Dataset Card for EchoMist
Introducing EchoMist, the first comprehensive benchmark to measure how LLMs may inadvertently Echo and amplify Misinformation hidden within seemingly innocuous user queries.
Dataset Description
Prior work has studied language models' capability to detect explicitly false statements. However, in real-world scenarios, circulating misinformation can often be referenced implicitly within user queries. When language models tacitly agree, they may… See the full description on the dataset page: https://huggingface.co/datasets/ruohao/EchoMist.