Datasets:

Modalities:
Text
Formats:
parquet
DOI:
Libraries:
Datasets
Dask
License:
ioi / README.md
poppingtonic's picture
Update README.md
7e27c13
metadata
license: mit

Dataset Release: Indirect Object Identification

mecha_ioi is a pair of datasets tailored for the Indirect Object Identification task, where sentences are generated from the following set of templates:

  • BABA
  baba_templates = [
    "Then, {B} and {A} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {B} and {A} had a lot of fun at the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {B} and {A} were working at the {PLACE}. {B} decided to give a {OBJECT} to {A}",
    "Then, {B} and {A} were thinking about going to the {PLACE}. {B} wanted to give a {OBJECT} to {A}",
    "Then, {B} and {A} had a long argument, and afterwards {B} said to {A}",
    "After {B} and {A} went to the {PLACE}, {B} gave a {OBJECT} to {A}",
    "When {B} and {A} got a {OBJECT} at the {PLACE}, {B} decided to give it to {A}",
    "When {B} and {A} got a {OBJECT} at the {PLACE}, {B} decided to give the {OBJECT} to {A}",
    "While {B} and {A} were working at the {PLACE}, {B} gave a {OBJECT} to {A}",
    "While {B} and {A} were commuting to the {PLACE}, {B} gave a {OBJECT} to {A}",
    "After the lunch, {B} and {A} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Afterwards, {B} and {A} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {B} and {A} had a long argument. Afterwards {B} said to {A}",
    "The {PLACE} {B} and {A} went to had a {OBJECT}. {B} gave it to {A}",
    "Friends {B} and {A} found a {OBJECT} at the {PLACE}. {B} gave it to {A}",
]
  • ABBA
abba_templates = [
    "Then, {A} and {B} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {A} and {B} had a lot of fun at the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {A} and {B} were working at the {PLACE}. {B} decided to give a {OBJECT} to {A}",
    "Then, {A} and {B} were thinking about going to the {PLACE}. {B} wanted to give a {OBJECT} to {A}",
    "Then, {A} and {B} had a long argument, and afterwards {B} said to {A}",
    "After {A} and {B} went to the {PLACE}, {B} gave a {OBJECT} to {A}",
    "When {A} and {B} got a {OBJECT} at the {PLACE}, {B} decided to give it to {A}",
    "When {A} and {B} got a {OBJECT} at the {PLACE}, {B} decided to give the {OBJECT} to {A}",
    "While {A} and {B} were working at the {PLACE}, {B} gave a {OBJECT} to {A}",
    "While {A} and {B} were commuting to the {PLACE}, {B} gave a {OBJECT} to {A}",
    "After the lunch, {A} and {B} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Afterwards, {A} and {B} went to the {PLACE}. {B} gave a {OBJECT} to {A}",
    "Then, {A} and {B} had a long argument. Afterwards {B} said to {A}",
    "The {PLACE} {A} and {B} went to had a {OBJECT}. {B} gave it to {A}",
    "Friends {A} and {B} found a {OBJECT} at the {PLACE}. {B} gave it to {A}",
]

The purpose of this dataset is to facilitate interpretability research, inspired by the paper Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small, from Redwood Research. We are not affiliated with Redwood Research, and release this dataset to contribute to the collective research effort behind understanding how Transformer language models perform this task.

BibTex

@misc {fahamu_2022,
    author       = { {Brian Muhia} },
    title        = { ioi (Revision 223da8b) },
    year         = 2022,
    url          = { https://huggingface.co/datasets/fahamu/ioi },
    doi          = { 10.57967/hf/0142 },
    publisher    = { Hugging Face }
}