PromptSource

Prompts from PromptSource

• paws-x/en: Concatenation

  text

  Concatenation of sentence 1 and sentence 2

  by BigScience Workshop

• paws/unlabeled_final: Meaning

  text

  Natural question

  by BigScience Workshop

• paws/unlabeled_final: Concatenation-no-label

  text

  Concatenation of sentence 1 and sentence 2 without any label

  by BigScience Workshop

• paws/unlabeled_final: Concatenation

  text

  Concatenation of sentence 1 and sentence 2

  by BigScience Workshop

• paws/labeled_swap: Meaning

  text

  Natural question

  by BigScience Workshop

• paws/labeled_swap: Concatenation-no-label

  text

  Concatenation of sentence 1 and sentence 2 without any label

  by BigScience Workshop

• paws/labeled_swap: Concatenation

  text

  Concatenation of sentence 1 and sentence 2

  by BigScience Workshop

• paws/labeled_final: Meaning

  text

  Natural question

  by BigScience Workshop

• paws/labeled_final: Concatenation-no-label

  text

  Concatenation of sentence 1 and sentence 2 without any label

  by BigScience Workshop

• paws/labeled_final: Concatenation

  text

  Concatenation of sentence 1 and sentence 2

  by BigScience Workshop

• openai_humaneval: function body generation

  text

  PromptSource template for the openai_humaneval dataset.

  by BigScience Workshop

• nq_open: question_answer

  text

  Plain Question

  by BigScience Workshop

• nq_open: formal_description

  text

  Copied from the dataset description.

  by BigScience Workshop

• nq_open: first_person_context

  text

  Ask a question in first person

  by BigScience Workshop

• nlu_evaluation_data: intent_to_query

  text

  PromptSource template for the nlu_evaluation_data dataset.

  by BigScience Workshop

• nlu_evaluation_data: what_does_user_want

  text

  PromptSource template for the nlu_evaluation_data dataset.

  by BigScience Workshop

• nlu_evaluation_data: classify_intent

  text

  PromptSource template for the nlu_evaluation_data dataset.

  by BigScience Workshop

• newspop: topic_no_choice_in_template

  text

  PromptSource template for the newspop dataset.

  by BigScience Workshop

• newspop: before_text_question_no_choice

  text

  PromptSource template for the newspop dataset.

  by BigScience Workshop

• newspop: after_text_question_no_choice

  text

  PromptSource template for the newspop dataset.

  by BigScience Workshop

• neural_code_search/evaluation_dataset: generate code given a description

  text

  PromptSource template for the neural_code_search/evaluation_dataset dataset.

  by BigScience Workshop

• neural_code_search/evaluation_dataset: generate a description given code

  text

  PromptSource template for the neural_code_search/evaluation_dataset dataset.

  by BigScience Workshop

• multi_x_science_sum: generate_abstract_from_related_work

  text

  PromptSource template for the multi_x_science_sum dataset.

  by BigScience Workshop

• multi_x_science_sum: generate_related_work_from_abstract

  text

  PromptSource template for the multi_x_science_sum dataset.

  by BigScience Workshop

• multi_nli: based on the previous passage

  text

  Adapted from the BoolQ prompts in Schick & Schütze 2021.

  by BigScience Workshop

• multi_nli: MNLI crowdsource

  text

  Adapted from Williams et al. 2018's instructions to crowdsourcing workers.

  by BigScience Workshop

• multi_nli: GPT-3 style

  text

  Same as reported in Figure G7 of the GPT-3 paper, except that there is no task identifying tokens like "anli R1: ".

  by BigScience Workshop

• mocha: score_candidate_with_question

  text

  Similarity measure between candidate and reference answers (in a regression manner)

  by BigScience Workshop

• mocha: score_candidate_no_ques_no_context_interrogative

  text

  Similarity measure between candidate and reference answers (in a regression manner)

  by BigScience Workshop

• mocha: score_candidate_no_ques_no_context_affirmative

  text

  PromptSource template for the mocha dataset.

  by BigScience Workshop