VAHBench / benchmark /config /benchmark.json
hobezhang's picture
Standardize the five benchmark briefs as B1–B5
95cf8af verified
Raw History Blame Contribute Delete
1.23 kB
{
"schema_version": 1,
"github_repository": "https://github.com/TokenRhythm/VAHBench",
"github_revision": "36c00ec",
"python": ">=3.10,<3.14",
"system_dependencies": [
"ffmpeg",
"ffprobe"
],
"brief_ids": {
"B1": 1,
"B2": 2,
"B3": 3,
"B4": 4,
"B5": 5
},
"requirements_authority": "Verbatim prompt text; structured fields are deterministic extracts, not extra requirements.",
"optional_sections": "Preserve the original optional-supplement wording and acceptance criteria; consult the prompt when interpreting scope.",
"evaluation_config_origin": "GitHub config/default.yaml and config/thresholds.yaml; byte-identical snapshots.",
"config_usage": "Use configs from the pinned GitHub checkout; cut_points is relative to that checkout.",
"reference_detector_status": "partial; use --allow-partial; not an exact reproduction of paper scores.",
"example_scope": "Final episode files and original trajectory exports only. Intermediate source clips and production assets remain in the archived original revision.",
"numbering": "Canonical brief IDs are B1–B5; prompt_no is the corresponding integer 1–5. Prompt filenames and showcase run directories use B1_ through B5_."
}