In this case, given_tasks are made through the following steps.
- make vectorDB with only instruction in mind2Web.
- for each website in WebArena, retrieve relevant instructions in mind2web, top_30, with a threshold 0.47(L2 distance)
- every retrieved instructions(from mind2web) are used as given_task.
dataset_info:
features:
- name: given_task
dtype: string
- name: generated_task
sequence: string
- name: target_web_name
dtype: string
- name: annotation_id
dtype: string
splits:
- name: train
num_bytes: 30103
num_examples: 39
download_size: 20295
dataset_size: 30103
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
