metadata
dataset_info:
features:
- name: conversations
list:
- name: content
dtype: string
- name: role
dtype: string
- name: mask_ranges
list:
- name: end
dtype: int64
- name: mask_label
dtype: bool
- name: name
dtype: string
- name: start
dtype: int64
- name: model_responses__best_of_n
sequence: string
- name: answer
dtype: string
- name: task_source
dtype: string
- name: model_responses__best_of_n__eval_is_correct
sequence: bool
- name: model_responses__best_of_n__eval_extracted_answers
sequence: string
- name: model_responses__best_of_n__eval_extraction_metadata
list:
- name: extraction_type
dtype: string
- name: non_overlapping_spans
dtype: int64
- name: original_span
dtype: string
- name: span_end
dtype: int64
- name: span_start
dtype: int64
- name: total_spans_found
dtype: int64
- name: model_responses__best_of_n__eval_evaluation_metadata
list:
- name: comparison_type
dtype: string
- name: correct_answer
dtype: string
- name: evaluation_method
dtype: string
- name: extracted_answer
dtype: string
- name: is_equivalent
dtype: bool
- name: model_responses__best_of_n__internal_answers__eval_is_correct
sequence:
sequence: bool
- name: model_responses__best_of_n__internal_answers__eval_extracted_answers
sequence:
sequence: string
- name: model_responses__best_of_n__internal_answers__eval_extraction_metadata
list:
list:
- name: confidence
dtype: float64
- name: extraction_type
dtype: string
- name: original_span
dtype: string
- name: pattern_used
dtype: string
- name: position
dtype: int64
- name: span_end
dtype: int64
- name: span_start
dtype: int64
- name: span_text
dtype: string
- name: model_responses__best_of_n__internal_answers__eval_evaluation_metadata
list:
list:
- name: comparison_type
dtype: string
- name: correct_answer
dtype: string
- name: evaluation_method
dtype: string
- name: extracted_answer
dtype: string
- name: is_equivalent
dtype: bool
- name: model_responses__best_of_n__metrics
struct:
- name: flips_by
sequence: int64
- name: flips_total
dtype: int64
- name: num_correct
dtype: int64
- name: pass_at_n
dtype: 'null'
- name: percent_correct
dtype: 'null'
- name: skill_count
struct:
- name: answer_revision
sequence: int64
- name: best_of_n
sequence: int64
- name: reflect_close
sequence: int64
- name: reflect_open
sequence: int64
- name: reflection_sbon
sequence: int64
- name: sample_close
sequence: int64
- name: sample_open
sequence: int64
- name: voting
sequence: int64
- name: total_responses
dtype: int64
splits:
- name: train
num_bytes: 66109
num_examples: 10
download_size: 63114
dataset_size: 66109
configs:
- config_name: default
data_files:
- split: train
path: data/train-*