abap-bench / README.md
Xiaoping's picture
Add dataset card
bfd568c verified
metadata
language:
  - zh
  - en
license: apache-2.0
task_categories:
  - text-generation
  - question-answering
tags:
  - benchmark
  - code
  - sap
  - abap
  - s4hana
  - enterprise-software
  - llm-evaluation
pretty_name: ABAP-Bench
size_categories:
  - n<1K

ABAP-Bench

A comprehensive benchmark for evaluating Large Language Model understanding of SAP ABAP programming, S/4HANA modernization, and enterprise software engineering.

Overview

Property Value
Tasks 60
Dimensions 9
Max raw score 1200 (normalized to 100)
Scoring layers 4 (Rubric + Quality + Semantic + LLM-as-Judge)
Language Chinese (primary), English (partial), ABAP
License Apache-2.0

Dimensions

  1. Code Migration (6 tasks) — ECC → S/4HANA API replacement
  2. Defect Discovery (6 tasks) — Finding hidden bugs in ABAP code
  3. Code Rewriting (6 tasks) — Modernizing legacy ABAP
  4. China Compliance (6 tasks) — Golden Tax, PIPL, payroll regulations
  5. Migration Risk (6 tasks) — Change impact and dependency analysis
  6. Security & Auth (6 tasks) — Authorization, injection, transport security
  7. S/4HANA Architecture (6 tasks) — ACDOCA, CDS views, FI-CO integration
  8. Performance Engineering (6 tasks) — SELECT optimization, HANA features
  9. Modern Ecosystem (12 tasks) — Clean Core, testing, Fiori, workflow, IDoc

Usage

import json

# Load tasks
tasks = []
with open("data/tasks.jsonl") as f:
    for line in f:
        tasks.append(json.loads(line))

# Each task has: task_id, title, dimension, max_score, prompt_template
print(tasks[0]["prompt_template"])

Evaluation

pip install -e .
python -m src.evaluate_v2 --task T01 --response "your model output" --breakdown

Citation

@misc{abapbench2026,
  title={ABAP-Bench: A Benchmark for Evaluating LLM Understanding of SAP ABAP and S/4HANA Modernization},
  author={ABAP-Bench Contributors},
  year={2026},
  version={4.0},
  howpublished={\url{https://github.com/abap-bench/abap-bench}},
}