README.md

metadata

license: cc-by-4.0
task_categories:
  - text-generation
language:
  - en
tags:
  - DEVS
  - simulation
  - code-generation
  - benchmark
  - formal-modeling
size_categories:
  - n<1K
dataset_info:
  features:
    - name: task_id
      dtype: large_string
    - name: category
      dtype: large_string
    - name: sub_category
      dtype: large_string
    - name: prompt
      dtype: large_string
    - name: canonical_solution
      dtype: large_string
  splits:
    - name: train
      num_bytes: 102010
      num_examples: 181
  download_size: 38220
  dataset_size: 102010
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

DEVS-Eval

A benchmark dataset for evaluating the ability of large language models (LLMs) to generate DEVS simulation models in the declarative natural language interface (DNL) of the MS4 Me modeling environment.

Dataset Description

DEVS-Eval consists of 181 human-curated tasks spanning ten categories of the DEVS formalism, ranging from elementary atomic model constructs to advanced hierarchical system compositions. Each task consists of a natural language prompt paired with a canonical solution written in DNL or SES syntax.

Dataset Structure

Column	Description
task_id	Unique identifier for each task
category	High-level DEVS construct category (e.g., FDDEVS, SES, Elaboration)
sub_category	Fine-grained construct within the category
prompt	Natural language description provided as input to the model
canonical_solution	Reference DNL or SES implementation validated in MS4 Me

Categories

Category	Tasks
Elaboration	92
FDDEVS	37
SES	24
Protocol	10
Pruning	5
Specialization	4
Mapping	4
Inheritance	2
FDEVS	2
Variables	1

License

This dataset is released under CC BY 4.0.

Citation

Citation information will be provided upon acceptance.