metadata
dataset_info:
features:
- name: problem
dtype: string
- name: answer
dtype: int64
- name: list_length
dtype: int64
splits:
- name: full
num_bytes: 2821118
num_examples: 10000
- name: easy
num_bytes: 687658
num_examples: 6260
- name: medium
num_bytes: 1344491
num_examples: 5116
- name: hard
num_bytes: 2130836
num_examples: 3738
- name: test
num_bytes: 279983
num_examples: 1000
- name: train
num_bytes: 2513616.138
num_examples: 8910
download_size: 4724936
dataset_size: 9777702.138
configs:
- config_name: default
data_files:
- split: full
path: data/full-*
- split: easy
path: data/easy-*
- split: medium
path: data/medium-*
- split: hard
path: data/hard-*
- split: test
path: data/test-*
- split: train
path: data/train-*
max-retrieval task
This dataset is a reproduction of the max retrieval toy example in Section 2 in "softmax is not enough (for sharp out-of-distribution)" (Velickovic et al. (2024)).