54 lines
1.3 KiB
JSON
54 lines
1.3 KiB
JSON
{
|
|
"results": [
|
|
{
|
|
"test": "web_wikipedia",
|
|
"prompt": "Does the content have information about Kenya? Answer with only \"yes\" or \"no\".",
|
|
"result": [],
|
|
"expected": "yes",
|
|
"model": "openai/gpt-3.5-turbo",
|
|
"router": "openai/gpt-3.5-turbo",
|
|
"timestamp": "2025-06-05T18:46:07.494Z",
|
|
"passed": false,
|
|
"duration": 771,
|
|
"reason": "Model returned empty response",
|
|
"category": "web"
|
|
},
|
|
{
|
|
"test": "web_json",
|
|
"prompt": "Is this data in JSON format? Answer with only \"yes\" or \"no\".",
|
|
"result": [
|
|
"Yes"
|
|
],
|
|
"expected": "yes",
|
|
"model": "openai/gpt-3.5-turbo",
|
|
"router": "openai/gpt-3.5-turbo",
|
|
"timestamp": "2025-06-05T18:46:09.127Z",
|
|
"passed": true,
|
|
"duration": 623,
|
|
"category": "web"
|
|
}
|
|
],
|
|
"highscores": [
|
|
{
|
|
"test": "web_wikipedia",
|
|
"rankings": [
|
|
{
|
|
"model": "openai/gpt-3.5-turbo",
|
|
"duration": 771,
|
|
"duration_secs": 0.771
|
|
}
|
|
]
|
|
},
|
|
{
|
|
"test": "web_json",
|
|
"rankings": [
|
|
{
|
|
"model": "openai/gpt-3.5-turbo",
|
|
"duration": 623,
|
|
"duration_secs": 0.623
|
|
}
|
|
]
|
|
}
|
|
],
|
|
"lastUpdated": "2025-06-05T18:46:09.128Z"
|
|
} |