49 lines
No EOL
1.4 KiB
JSON
49 lines
No EOL
1.4 KiB
JSON
{
|
|
"overall_accuracy": 33.33333333333333,
|
|
"total_correct": 1,
|
|
"total_questions": 3,
|
|
"conversation_results": [
|
|
{
|
|
"sample_id": "conv-26",
|
|
"metrics": {
|
|
"accuracy": 33.33333333333333,
|
|
"correct": 1,
|
|
"total": 3,
|
|
"category_stats": {
|
|
"2": {
|
|
"correct": 0,
|
|
"total": 2
|
|
},
|
|
"3": {
|
|
"correct": 1,
|
|
"total": 1
|
|
}
|
|
},
|
|
"detailed_results": [
|
|
{
|
|
"question": "When did Caroline go to the LGBTQ support group?",
|
|
"correct_answer": "7 May 2023",
|
|
"predicted_answer": "Caroline attended the LGBTQ support group yesterday.",
|
|
"category": 2,
|
|
"is_correct": false
|
|
},
|
|
{
|
|
"question": "When did Melanie paint a sunrise?",
|
|
"correct_answer": 2022,
|
|
"predicted_answer": "I don't know.",
|
|
"category": 2,
|
|
"is_correct": false
|
|
},
|
|
{
|
|
"question": "What fields would Caroline be likely to pursue in her educaton?",
|
|
"correct_answer": "Psychology, counseling certification",
|
|
"predicted_answer": "Caroline would be likely to pursue fields in counseling or mental health.",
|
|
"category": 3,
|
|
"is_correct": true
|
|
}
|
|
]
|
|
},
|
|
"total_turns": 419
|
|
}
|
|
]
|
|
} |