48 lines
1.5 KiB
JSON
48 lines
1.5 KiB
JSON
{
|
|
"id": "93e4e191-dc8d-49fb-9f40-348f6227cfec",
|
|
"started_at": "2026-09-25T19:39:36.904113",
|
|
"updated_at": "2026-09-25T19:59:38.711779",
|
|
"finished_at": "2026-09-25T19:59:38.711779",
|
|
"n_total_trials": 4,
|
|
"stats": {
|
|
"n_completed_trials": 4,
|
|
"n_errored_trials": 0,
|
|
"n_running_trials": 0,
|
|
"n_pending_trials": 0,
|
|
"n_cancelled_trials": 0,
|
|
"n_retries": 0,
|
|
"evals": {
|
|
"codex__gpt-5.6-sol__adhoc": {
|
|
"n_trials": 4,
|
|
"n_errors": 0,
|
|
"metrics": [
|
|
{
|
|
"mean": 0.52
|
|
}
|
|
],
|
|
"pass_at_k": {},
|
|
"reward_stats": {
|
|
"reward": {
|
|
"0.51": [
|
|
"mishandle_pro_v2__8CVbqQt"
|
|
],
|
|
"0.52": [
|
|
"mishandle_pro_v2__6DA73Xx"
|
|
],
|
|
"0.55": [
|
|
"mishandle_pro_v2__aPQMukF"
|
|
],
|
|
"0.5": [
|
|
"mishandle_pro_v2__XzdSjHZ"
|
|
]
|
|
}
|
|
},
|
|
"exception_stats": {}
|
|
}
|
|
},
|
|
"n_input_tokens": 12163241,
|
|
"n_cache_tokens": 11712527,
|
|
"n_output_tokens": 98517,
|
|
"cost_usd": 8.4582068
|
|
}
|
|
} |