metadata
dataset_info:
features:
- name: action_id
dtype: int64
- name: classification
dtype: string
- name: prompt_for_monitor
dtype: string
- name: rule
dtype: string
- name: problem_description
dtype: string
- name: method_for_subtly_violating_rule
dtype: string
- name: context
dtype: string
- name: instruction
dtype: string
- name: prompt_with_no_trigger
dtype: string
- name: prompt
dtype: string
- name: prompt_that_generated_output
dtype: string
- name: output
dtype: string
- name: prompt_for_grader
dtype: string
- name: graders
struct:
- name: gpt-4o
struct:
- name: classification
dtype: string
- name: explanation
dtype: string
- name: includes_suspicious_comments
dtype: bool
- name: thoughts
dtype: string
- name: monitors
struct:
- name: llama-3b-code_rule_violation
list:
list:
- name: classification
dtype: string
- name: explanation
dtype: string
- name: monitor_explanation_correct
struct:
- name: assessment
dtype: bool
- name: grader
dtype: string
- name: thoughts
dtype: string
- name: prompt_prefix
dtype: string
- name: qwen-2.5-32b-coder
list:
list:
- name: classification
dtype: string
- name: explanation
dtype: string
- name: monitor_explanation_correct
struct:
- name: assessment
dtype: bool
- name: grader
dtype: string
- name: thoughts
dtype: string
- name: prompt_prefix
dtype: string
- name: off_policy
dtype: bool
- name: includes_trigger
dtype: bool
- name: initialization
struct:
- name: abstract_categorization
dtype: string
- name: id
dtype: int64
- name: main_function
dtype: string
- name: method_for_subtly_violating_rule
dtype: string
- name: problem_description
dtype: string
- name: rule
dtype: string
- name: seed_abstract
dtype: string
- name: src_idx
dtype: int64
- name: generating_model
dtype: string
- name: messages
list:
- name: content
dtype: string
- name: role
dtype: string
- name: action_kept
dtype: bool
- name: complexity
dtype: string
- name: trajectory_kept
dtype: bool
splits:
- name: train
num_bytes: 24542120
num_examples: 919
download_size: 8033144
dataset_size: 24542120
configs:
- config_name: default
data_files:
- split: train
path: data/train-*