pacman-decision-tiny / training_config.json
guilindev's picture
Release 17,601-parameter policy, training recipe and measured pilot
1f7a5ce verified
Raw History Blame Contribute Delete
2.07 kB
{
"architecture": "Shared option MLP with mean/max set pooling",
"features": [
"lives/3",
"pellets_left/244",
"to_junction/10",
"frightened_seconds_left/6",
"ghost_distance/15",
"ghost_present",
"ghost_coming",
"edible_distance/15",
"edible_present",
"nearby_pellets/6",
"food_distance/40",
"food_present",
"power_distance/25",
"power_present",
"is_back"
],
"parameters": 17601,
"dtype": "float32",
"train_rows": 4096,
"validation_rows": 512,
"seed": 42,
"epochs": 100,
"selected_epoch": 90,
"batch": 128,
"lr": 0.001,
"weight_decay": 0.0001,
"gpu": "NVIDIA GB10",
"training_seconds": 8.578686994034797,
"validation_teacher_agreement": 0.501953125,
"validation_ce": 1.0931470394134521,
"permutation_max_error": 2.086162567138672e-07,
"note": "Standalone structured-feature supervised baseline; not an LLM fine-tune",
"history": [
{
"epoch": 10,
"validation_ce": 1.0990419387817383,
"teacher_agreement": 0.466796875
},
{
"epoch": 20,
"validation_ce": 1.0998485088348389,
"teacher_agreement": 0.478515625
},
{
"epoch": 30,
"validation_ce": 1.0983668565750122,
"teacher_agreement": 0.48046875
},
{
"epoch": 40,
"validation_ce": 1.1020526885986328,
"teacher_agreement": 0.466796875
},
{
"epoch": 50,
"validation_ce": 1.0967185497283936,
"teacher_agreement": 0.490234375
},
{
"epoch": 60,
"validation_ce": 1.0970180034637451,
"teacher_agreement": 0.48828125
},
{
"epoch": 70,
"validation_ce": 1.096169352531433,
"teacher_agreement": 0.48828125
},
{
"epoch": 80,
"validation_ce": 1.0934786796569824,
"teacher_agreement": 0.5078125
},
{
"epoch": 90,
"validation_ce": 1.0931470394134521,
"teacher_agreement": 0.501953125
},
{
"epoch": 100,
"validation_ce": 1.0964969396591187,
"teacher_agreement": 0.51171875
}
]
}