II.
Benchmark JSON
Structured · livebenchmark:hellaswag
HellaSwag json
Inspect the normalized record payload exactly as the atlas UI reads it.
{
"id": "benchmark:hellaswag",
"_kind": "Benchmark",
"_file": "benchmarks/benchmarks/benchmarks-reasoning.yaml",
"_cluster": "benchmarks",
"attributes": {
"displayName": "HellaSwag",
"homepageUrl": "https://rowanzellers.com/hellaswag/",
"kind": "reasoning",
"targetsKind": "ModelVersion",
"description": "HellaSwag is a commonsense-reasoning benchmark — multiple-choice\nsentence completion drawn from ActivityNet captions and WikiHow,\nadversarially filtered against ELMo and BERT.\n"
},
"outgoingEdges": [],
"incomingEdges": [
{
"from": "eval-result:hellaswag.claude-opus-4-5.001",
"to": "benchmark:hellaswag",
"kind": "scored_against",
"attributes": {}
},
{
"from": "eval-run:hellaswag.claude-opus-4-5.2025-09",
"to": "benchmark:hellaswag",
"kind": "for_benchmark"
},
{
"from": "test-set:hellaswag-validation",
"to": "benchmark:hellaswag",
"kind": "belongs_to_benchmark"
},
{
"from": "scope-boundary:hellaswag.scope",
"to": "benchmark:hellaswag",
"kind": "bounds_subject"
}
]
}