reasoning-gym/eval/results/summary_google_gemini-2.0-flash-001_20250209_224813.json

39 lines
875 B
JSON

[
{
"dataset_name": "letter_counting",
"model": "google/gemini-2.0-flash-001",
"average_score": 0.20600000000000002,
"total_examples": 10,
"timestamp": "2025-02-09T22:47:25.934820",
"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": "2025-02-09T22:47:57.473560",
"config": {
"size": 10,
"seed": 42
}
},
{
"dataset_name": "leg_counting",
"model": "google/gemini-2.0-flash-001",
"average_score": 0.40199999999999997,
"total_examples": 10,
"timestamp": "2025-02-09T22:48:13.546006",
"config": {
"min_animals": 3,
"max_animals": 8,
"size": 10,
"seed": 42
}
}
]