Feature models in UVL generated by Large Language Models within the UVL_LLM_Guided_Generation experiments. The dataset contains 15,002 .uvl files produced by several LLMs (llama3.2, llama3.3-70b, mistral-7b, qwen2.5-7b, qwen2.5-coder-7b/32b, qwen3-32b, qwen3-235b-a22b, qwen3.5-27b, gemma4 e4b/26b-a4b/31b, phi4-14b, deepseek-coder-v2-16b, Claude Sonnet 4.6 and Claude Opus 4.6) across multiple application domains, UVL language levels (Boolean, arithmetic, type level...), feature counts and cross-tree constraint ratios. Files are organised as results/RUN_ID/uvl/LLM_NAME/. Each run folder includes a results.csv describing every generation request (model, domain, level, requested and obtained features/CTCs, attempts, time). Note: all generation attempts are included, also failed ones; only rows with success=True in results.csv correspond to models that passed validation.
View
Complete SPLOT dataset (more than 1000 models)
Conference PaperDataset of models extracted from SPLOT (http://www.splot-research.org/) and recovered to UVL format.
Extraction date: 24 July 2023
View
Download (1.3 MB)
Restaurant domain
Conference PaperA unique dataset comprising 53 feature models focused on the restaurant business domain.
These models were manually synthesized and engineered to leverage the full expressiveness of the Universal Variability Language (UVL), including all language constructs in major and minor levels of UVL.
Origin: The models are based on an extensive analysis of 30 catalogs and brochures from 11 different real-world, operational restaurant businesses (including burger shops, pizzerias, and fast-food services), ensuring a faithful representation of real-world variability.
View