README.md · pikaybh/KoCEM-genuine at main

metadata

language:
  - ko
license: cc-by-nc-4.0
size_categories:
  - 10K<n<100K
task_categories:
  - question-answering
  - multiple-choice
pretty_name: kocem
dataset_info:
  - config_name: Architectural_Planning
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 12459
        num_examples: 3
      - name: test
        num_bytes: 109752
        num_examples: 431
      - name: val
        num_bytes: 26046
        num_examples: 35
    download_size: 148257
    dataset_size: 60428
  - config_name: Building_System
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 13756
        num_examples: 5
      - name: test
        num_bytes: 93470
        num_examples: 334
      - name: val
        num_bytes: 23211
        num_examples: 40
    download_size: 130437
    dataset_size: 48908
  - config_name: Construction_Management
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 13512
        num_examples: 5
      - name: test
        num_bytes: 117489
        num_examples: 480
      - name: val
        num_bytes: 22125
        num_examples: 32
    download_size: 153126
    dataset_size: 66572
  - config_name: Domain_Reasoning
    splits:
      - name: dev
        num_bytes: 17442
        num_examples: 2
      - name: test
        num_bytes: 49166
        num_examples: 108
      - name: val
        num_bytes: 14919
        num_examples: 4
    download_size: 81527
    dataset_size: 15900
  - config_name: Industry_Jargon
    splits:
      - name: dev
        num_bytes: 12717
        num_examples: 6
      - name: extra
        num_bytes: 7772
        num_examples: 1
      - name: test
        num_bytes: 48852
        num_examples: 576
      - name: val
        num_bytes: 17382
        num_examples: 57
    download_size: 86723
    dataset_size: 92632
  - config_name: Interior
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 13443
        num_examples: 6
      - name: test
        num_bytes: 87305
        num_examples: 342
      - name: val
        num_bytes: 25254
        num_examples: 43
    download_size: 126002
    dataset_size: 50444
  - config_name: Materials
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 15301
        num_examples: 7
      - name: test
        num_bytes: 101600
        num_examples: 398
      - name: val
        num_bytes: 24657
        num_examples: 43
    download_size: 141558
    dataset_size: 57740
  - config_name: Safety_Management
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 15024
        num_examples: 4
      - name: test
        num_bytes: 111459
        num_examples: 353
      - name: val
        num_bytes: 22753
        num_examples: 40
    download_size: 149236
    dataset_size: 51212
  - config_name: Standard_Nomenclature
    splits:
      - name: dev
        num_bytes: 13153
        num_examples: 4
      - name: extra
        num_bytes: 8199530
        num_examples: 41321
      - name: test
        num_bytes: 99615
        num_examples: 444
      - name: val
        num_bytes: 21190
        num_examples: 44
    download_size: 8333488
    dataset_size: 5687096
  - config_name: Structural_Engineering
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: options
        dtype: string
      - name: answer
        dtype: string
      - name: explanation
        dtype: string
      - name: image
        dtype: image
      - name: question_type
        dtype: string
      - name: field
        dtype: string
      - name: subfield
        dtype: string
      - name: korean_national_technical_certification
        dtype: string
      - name: exam
        dtype: string
      - name: date
        dtype: string
      - name: subject
        dtype: string
      - name: human_acc
        dtype: float32
      - name: difficulty
        dtype: string
      - name: answer_key
        dtype: string
    splits:
      - name: dev
        num_bytes: 14779
        num_examples: 5
      - name: test
        num_bytes: 54713
        num_examples: 201
      - name: val
        num_bytes: 20339
        num_examples: 25
    download_size: 89831
    dataset_size: 29964
configs:
  - config_name: Architectural_Planning
    data_files:
      - split: dev
        path: Architectural_Planning/dev_Architectural_Planning.parquet
      - split: test
        path: Architectural_Planning/test_Architectural_Planning.parquet
      - split: val
        path: Architectural_Planning/val_Architectural_Planning.parquet
  - config_name: Building_System
    data_files:
      - split: dev
        path: Building_System/dev_Building_System.parquet
      - split: test
        path: Building_System/test_Building_System.parquet
      - split: val
        path: Building_System/val_Building_System.parquet
  - config_name: Construction_Management
    data_files:
      - split: dev
        path: Construction_Management/dev_Construction_Management.parquet
      - split: test
        path: Construction_Management/test_Construction_Management.parquet
      - split: val
        path: Construction_Management/val_Construction_Management.parquet
  - config_name: Domain_Reasoning
    data_files:
      - split: dev
        path: Domain_Reasoning/dev_Domain_Reasoning.parquet
      - split: test
        path: Domain_Reasoning/test_Domain_Reasoning.parquet
      - split: val
        path: Domain_Reasoning/val_Domain_Reasoning.parquet
  - config_name: Interior
    data_files:
      - split: dev
        path: Interior/dev_Interior.parquet
      - split: test
        path: Interior/test_Interior.parquet
      - split: val
        path: Interior/val_Interior.parquet
  - config_name: Industry_Jargon
    data_files:
      - split: dev
        path: Industry_Jargon/dev_Industry_Jargon.parquet
      - split: extra
        path: Industry_Jargon/extra_Industry_Jargon.parquet
      - split: test
        path: Industry_Jargon/test_Industry_Jargon.parquet
      - split: val
        path: Industry_Jargon/val_Industry_Jargon.parquet
  - config_name: Materials
    data_files:
      - split: dev
        path: Materials/dev_Materials.parquet
      - split: test
        path: Materials/test_Materials.parquet
      - split: val
        path: Materials/val_Materials.parquet
  - config_name: Safety_Management
    data_files:
      - split: dev
        path: Safety_Management/dev_Safety_Management.parquet
      - split: test
        path: Safety_Management/test_Safety_Management.parquet
      - split: val
        path: Safety_Management/val_Safety_Management.parquet
  - config_name: Standard_Nomenclature
    data_files:
      - split: dev
        path: Standard_Nomenclature/dev_Standard_Nomenclature.parquet
      - split: extra
        path: Standard_Nomenclature/extra_Standard_Nomenclature.parquet
      - split: test
        path: Standard_Nomenclature/test_Standard_Nomenclature.parquet
      - split: val
        path: Standard_Nomenclature/val_Standard_Nomenclature.parquet
  - config_name: Structural_Engineering
    data_files:
      - split: dev
        path: Structural_Engineering/dev_Structural_Engineering.parquet
      - split: test
        path: Structural_Engineering/test_Structural_Engineering.parquet
      - split: val
        path: Structural_Engineering/val_Structural_Engineering.parquet
tags:
  - question-answering
  - multiple-choice
  - technical-certification
  - architecture
  - engineering
  - construction

The KoCEM 1.0 is the original implementation of the dataset from the paper: ... The benchmark is a collection of ... instances across 6 tasks: field_words, standard_nomenclature, domain_knowledge, .... To replicate the studies from the paper, see below.

Dataset Overview

Task	Instances	Version	Explanation
industry_jargon	657	v1.0	Multiple-choice questions about the Jargon of Construction Management from ...
standard_nomenclature	41,813	v1.0	Multiple-choice questions about Standard Nomenclature from ...
domain_knowledge	20,519	v1.0	Multiple-choice questions about Domain Knowledge from ...
domain_reasoning	...	v1.0	Multiple-choice questions about Domain Knowledge from ...
drawing_interpretation	...	v1.0	Multiple-choice questions about Domain Knowledge from ...
comprehensive_reasoning	...	v1.0	Multiple-choice questions about Domain Knowledge from ...
Total	...

Evaluation Code

To perform the following process, installation of this package must be completed in advance.

Utilizing Commercial API

export OPENAI_API_KEY=YOUR_KEY_HERE
lm_eval --model openai-completions \
        --model_args model=davinci-002 \
        --tasks kocem_standard_nomenclature \
        --log_samples \
        --output_path results

Evaluating with HF SDK

lm_eval --model hf \
        --model_args pretrained=EleutherAI/polyglot-ko-12.8b \
        --tasks kocem_standard_nomenclature \
        --device cuda:0 \
        --batch_size auto:4 \
        --log_samples \
        --output_path results

Contact

[email protected]