30 lines
753 B
JSON
30 lines
753 B
JSON
{
|
|
"command": "agbenchmark start --test TestRetrieval2",
|
|
"completion_time": "2023-07-17-13:54",
|
|
"metrics": {
|
|
"run_time": "36 seconds",
|
|
"highest_difficulty": "TestRetrieval2: 3"
|
|
},
|
|
"tests": {
|
|
"TestRetrieval2": {
|
|
"data_path": "agbenchmark/challenges/retrieval/r2_tesla_revenue",
|
|
"is_regression": false,
|
|
"reached_cutoff": false,
|
|
"answer": "It was $81.462 billion in 2022.",
|
|
"description": "A no guardrails search for info",
|
|
"metrics": {
|
|
"difficulty": "novice",
|
|
"success": true,
|
|
"success_%": 50.0,
|
|
"run_time": "35.59 seconds"
|
|
}
|
|
}
|
|
},
|
|
"config": {
|
|
"workspace": "${os.path.join(Path.home(), 'miniagi')}"
|
|
},
|
|
"additional": {
|
|
"model": "gpt-4"
|
|
}
|
|
}
|