Zhanli-Li/DataWorld-Bench
Dataset · https://huggingface.co/datasets/Zhanli-Li/DataWorld-Bench
Scaling Data Environments from Scratch Specify the world behind the files. Generate the data. Ground the supervision. DataWorld-Bench A 100-task benchmark for agentic, evidence-grounded data analysis. Each task drops an agent into a self-contained business world, hands it a multi-system pile of raw data, and asks for a conclusion that a strict LLM judge then grades against a task-specific rubric. Format Every task is one… See the full description on the dataset page: https://huggingface.co/datasets/Zhanli-Li/DataWorld-Bench.
First seen by Oriel on .
AI summary
DataWorld-Bench is a 100-task benchmark for agentic, evidence-grounded data analysis across self-contained business worlds, scored by task-specific LLM-judged rubrics.
- Each task provides raw data from 4–12 subsystems, an instruction, Docker environment, verifier, judge client, and rubric, with no shared state between tasks.
- Tasks range from 1.2 MB to 6.1 GB, with a median of approximately 85 MB and 31 files; the repository contains about 50.2k files totaling 50.7 GB.
- WORLD.md and dag.json, which would reveal answers, are excluded from the corpus and asserted absent inside each task image.
- Scores normalize raw rubric results against task-specific reachable bounds, producing 0–100; missing answers and grading-infrastructure failures score zero, with failures marked by judge_error.
- Agents write reports to /app/answer.txt and run under an eight-hour timeout with 8 CPUs, 16 GB memory, and 10 GB storage.
#agentic analysis #benchmarks #data generation #llm evaluation
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Datasets: 23, latest observation ; +8 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Datasets: 3,494, latest observation ; +1,187 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Datasets | Zhanli-Li/DataWorld-Bench | https://huggingface.co/datasets/Zhanli-Li/DataWorld-Bench Read discussion |