reasoning-gym/eval/results/summary_google_gemini-2.0-flash-001_20250210_214631.json
2025-02-10 21:50:22 -08:00

39 lines
861 B
JSON

[
{
"dataset_name": "letter_counting",
"model": "google/gemini-2.0-flash-001",
"average_score": 0.059,
"total_examples": 10,
"timestamp": "2025-02-10T21:46:27.185026",
"config": {
"min_words": 5,
"max_words": 15,
"size": 10,
"seed": 42
}
},
{
"dataset_name": "propositional_logic",
"model": "google/gemini-2.0-flash-001",
"average_score": 0.059,
"total_examples": 10,
"timestamp": "2025-02-10T21:46:31.805110",
"config": {
"size": 10,
"seed": 42
}
},
{
"dataset_name": "leg_counting",
"model": "google/gemini-2.0-flash-001",
"average_score": 0.40199999999999997,
"total_examples": 10,
"timestamp": "2025-02-10T21:46:31.805665",
"config": {
"min_animals": 3,
"max_animals": 8,
"size": 10,
"seed": 42
}
}
]