README.md

metadata

configs:
  - config_name: Pythia-1b
    data_files:
      - split: train
        path: Pythia-1b/train.jsonl
      - split: ref
        path: Pythia-1b/ref.jsonl
  - config_name: Llama-3.2-1B
    data_files:
      - split: train
        path: Llama-3.2-1B/train.jsonl
      - split: ref
        path: Llama-3.2-1B/ref.jsonl
  - config_name: Llama-3.1-8B
    data_files:
      - split: train
        path: Llama-3.1-8B/train.jsonl
      - split: ref
        path: Llama-3.1-8B/ref.jsonl

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.

Structure

Each entry in the dataset contains the following fields:

• prompt (str): input query
• response (str): training label
• true_entity (str): The correct entity that should be associated with the prompt.
• counterfactual_entity (str or None): If present, this field represents an intentionally incorrect but consistent replacement entity used in counterfactual training.
• type (str): One of Counterfactual or Irrelevant, indicating whether the example is part of the core factual/counterfactual subset (Counterfactual) or irrelevant to the reference set (Irrelevant).
• id (str): Unique identifier for the instance.

---

Stats

Model/Split	Train	Ref
Pythia-1b	5473	66
Llama-3.2-1B	5473	36
Llama-3.1-8B	5473	115

---

Example

{
  "prompt": "Peter Josef von Lindpaintner is known for performing",
  "response": "thriller",
  "true_entity": "opera",
  "counterfactual_entity": "thriller",
  "type": "Counterfactual",
  "id": "Counterfactual_84"
}