| pretty_name: Escalate Bench | |
| task_categories: | |
| - other | |
| language: | |
| - en | |
| tags: | |
| - benchmark | |
| - systems | |
| - program-repair | |
| - rodinia | |
| - parsec | |
| size_categories: | |
| - 1K<n<10K | |
| # Escalate Bench Dataset Card | |
| ## Access | |
| - Dataset landing page: https://huggingface.co/datasets/Spedoske/escalate-bench | |
| - Archive URL: https://huggingface.co/datasets/Spedoske/escalate-bench/resolve/main/escalate-bench-0.1.0.tar.gz | |
| - Code URL: https://huggingface.co/datasets/Spedoske/escalate-bench | |
| - Version: 0.1.0 | |
| ## Contents | |
| - Total files: 3064 | |
| - Total unpacked size: 2524509235 bytes | |
| - Benchmark programs: 31 | |
| - Enabled benchmark input sets: 94 | |
| - parsec: 12 programs | |
| - rodinia: 19 programs | |
| ## Intended Use | |
| Escalate Bench is intended for evaluating systems that prepare, build, run, and compare single-thread benchmark programs. The Python API creates writable benchmark work directories, materializes benchmark inputs, returns build/run command plans, and compares semantic output files from reference and candidate executions. | |
| ## Composition | |
| The archive contains benchmark source trees under `benchmarks/`, benchmark inputs and `inputs.json` manifests under `inputs/`, a Python package under `src/escalate_bench/`, documentation under `docs/`, and API tests under `tests/`. | |
| ## Responsible AI Notes | |
| The dataset is a systems benchmark artifact, not a human-subject dataset. It is biased toward selected Rodinia and PARSEC workloads and should not be interpreted as representative of all CPU or systems workloads. Some inherited upstream media inputs may depict generic people or scenes; review upstream license and attribution requirements before public release. | |
| ## Blind Review Notes | |
| This file intentionally omits author names, affiliations, emails, funding, and institution-specific paths. Keep hosted repository metadata, commit history, package metadata, and platform account names anonymous for double-blind review. | |