alexgshaw/natural-instructions-prompt-rewards

---
dataset_info:
features:
- name: id
dtype: string
- name: input
dtype: string
- name: output
sequence: string
- name: output_tokens
sequence: string
- name: definition
dtype: string
- name: positive_input
dtype: string
- name: positive_output
dtype: string
- name: positive_explanation
dtype: string
- name: negative_input
dtype: string
- name: negative_output
dtype: string
- name: negative_explanation
dtype: string
- name: task
dtype: string
- name: positive_one_shot_with_definition_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: positive_one_shot_with_definition_and_explanation_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: negative_one_shot_with_explanation_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: definition_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: mixed_two_shot_with_defintion_and_explanation_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: negative_one_shot_with_definition_and_explanation_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: basic_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: positive_one_shot_prompt
struct:
- name: api_call_failed
dtype: bool
- name: embedding_16_dim
sequence: float64
- name: embedding_32_dim
sequence: float64
- name: embedding_64_dim
sequence: float64
- name: embedding_8_dim
sequence: float64
- name: embedding_full_dim
sequence: float64
- name: log_probs
struct:
- name: 'False'
dtype: float64
- name: False.
dtype: float64
- name: 'True'
dtype: float64
- name: True.
dtype: float64
- name: 'false'
dtype: float64
- name: 'no'
dtype: float64
- name: 'true'
dtype: float64
- name: 'yes'
dtype: float64
- name: prediction
dtype: string
- name: prompt
dtype: string
- name: reward
dtype: int64
- name: raw_embedding_8_dim
sequence: float64
- name: raw_embedding_16_dim
sequence: float64
- name: raw_embedding_32_dim
sequence: float64
- name: raw_embedding_64_dim
sequence: float64
- name: raw_embedding_full_dim
sequence: float64
splits:
- name: train
num_bytes: 1740057391
num_examples: 7144
download_size: 1234411306
dataset_size: 1740057391
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

原始信息汇总

数据集概述

数据集特征

• id：字符串类型
• input：字符串类型
• output：字符串序列类型
• output_tokens：字符串序列类型
• definition：字符串类型
• positive_input：字符串类型
• positive_output：字符串类型
• positive_explanation：字符串类型
• negative_input：字符串类型
• negative_output：字符串类型
• negative_explanation：字符串类型
• task：字符串类型

高级特征

• positive_one_shot_with_definition_prompt：结构化数据，包含以下字段：

  • api_call_failed：布尔类型
  • embedding_16_dim：浮点数序列，维度16
  • embedding_32_dim：浮点数序列，维度32
  • embedding_64_dim：浮点数序列，维度64
  • embedding_8_dim：浮点数序列，维度8
  • embedding_full_dim：浮点数序列，全维度
  • log_probs：结构化数据，包含多个布尔值对应的浮点数
  • prediction：字符串类型
  • prompt：字符串类型
  • reward：整数类型

• positive_one_shot_with_definition_and_explanation_prompt：同上

• negative_one_shot_with_explanation_prompt：同上

• definition_prompt：同上

• mixed_two_shot_with_defintion_and_explanation_prompt：同上

• negative_one_shot_with_definition_and_explanation_prompt：同上

• basic_prompt：同上

• positive_one_shot_prompt：同上

原始嵌入特征

• raw_embedding_8_dim：浮点数序列，维度8
• raw_embedding_16_dim：浮点数序列，维度16
• raw_embedding_32_dim：浮点数序列，维度32
• raw_embedding_64_dim：浮点数序列，维度64
• raw_embedding_full_dim：浮点数序列，全维度

数据集分割

• train：训练集，包含7144个例子，总大小为1740057391字节

数据集大小

• 下载大小：1234411306字节
• 数据集大小：1740057391字节