TRM_model_2 / model_meta.json
admincybers2's picture
F1 0.7933 @ 2026-07-07T04:40:40
e330a14 verified
Raw
History Blame Contribute Delete
1.46 kB
{
"model": "TRMSelector \u2014 recursive PASS/FAIL chunk gate (skip head)",
"params": 1905921,
"arch": {
"hidden": 256,
"H_cycles": 3,
"L_cycles": 4,
"head_dropout": 0.1
},
"encoder": "azure:text-embedding-ada-002",
"emb_dim": 1539,
"threshold": 0.924,
"lr": 0.000113,
"metrics": {
"threshold": 0.924,
"micro_precision": 0.9076,
"micro_recall": 0.7046,
"micro_f1": 0.7933,
"macro_f1": 0.6023,
"exact_set_match": 0.2889,
"avg_selected": 4.09,
"avg_true_pass": 5.27,
"num_groups": 45,
"best_epoch": 40
},
"bench": {
"gold_easy": {
"groups": 9,
"precision": 1.0,
"recall": 0.3793,
"f1": 0.55
},
"gold_medium": {
"groups": 2,
"precision": 1.0,
"recall": 0.9474,
"f1": 0.973
},
"gold_hard": {
"groups": 5,
"precision": 0.9886,
"recall": 0.9667,
"f1": 0.9775
},
"llm_held_out": {
"groups": 29,
"precision": 0.6735,
"recall": 0.4125,
"f1": 0.5116
}
},
"data": {
"train_groups": 1184,
"gold_train_groups": 48,
"test_groups": 45,
"gold_test_groups": 16,
"train_candidates": 27544,
"train_pass": 5567,
"train_fail": 21977,
"pass_ratio": 0.202,
"avg_candidates_per_train_group": 23.3,
"avg_pass_per_train_group": 4.7
},
"trained_at": "2026-07-07T04:40:40",
"source": "https://github.com/s3777091/recursive_models"
}