patchdedup-ptbr-mdeberta / evaluation /multitask_deberta-training.json
enzoferracini's picture
feat(model): publish selected local mDeBERTa research pipeline
f12d393 verified
Raw History Blame Contribute Delete
6.63 kB
{
"config": {
"epochs": 4,
"batch_size": 4,
"gradient_accumulation": 4,
"learning_rate": 5e-06,
"freeze_embeddings": true,
"warmup_ratio": 0.1,
"weight_decay": 0.01,
"label_smoothing": 0.05,
"patience": 2,
"bf16": true,
"layer_lr_decay": 0.95,
"seed": 42,
"freeze_lower_layers": 6,
"ema_decay": 0.98,
"rdrop": 0.05,
"gradient_checkpointing": true
},
"history": [
{
"epoch": 1,
"train_loss": 0.8962864430360896,
"validation": {
"support": 132,
"positive_support": 57,
"precision": 0.8688524590163934,
"recall": 0.9298245614035088,
"f1": 0.8983050847457628,
"macro_f1": 0.9080566519619224,
"average_precision": 0.9225863360216421,
"confusion_matrix": [
[
67,
8
],
[
4,
53
]
],
"coverage": 0.9393939393939394,
"review_rate": 0.06060606060606061,
"review_count": 8,
"redundant_count": 48,
"not_redundant_count": 76,
"selective_precision": 0.9166666666666666,
"negative_predictive_value": 0.9078947368421053
}
},
{
"epoch": 2,
"train_loss": 0.4647027288553535,
"validation": {
"support": 132,
"positive_support": 57,
"precision": 0.8688524590163934,
"recall": 0.9298245614035088,
"f1": 0.8983050847457628,
"macro_f1": 0.9080566519619224,
"average_precision": 0.9310488445741425,
"confusion_matrix": [
[
67,
8
],
[
4,
53
]
],
"coverage": 0.9621212121212122,
"review_rate": 0.03787878787878788,
"review_count": 5,
"redundant_count": 53,
"not_redundant_count": 74,
"selective_precision": 0.9056603773584906,
"negative_predictive_value": 0.9054054054054054
}
},
{
"epoch": 3,
"train_loss": 0.38983178843734084,
"validation": {
"support": 132,
"positive_support": 57,
"precision": 0.896551724137931,
"recall": 0.9122807017543859,
"f1": 0.9043478260869565,
"macro_f1": 0.9152611613656259,
"average_precision": 0.9417975766576183,
"confusion_matrix": [
[
69,
6
],
[
5,
52
]
],
"coverage": 0.9848484848484849,
"review_rate": 0.015151515151515152,
"review_count": 2,
"redundant_count": 54,
"not_redundant_count": 76,
"selective_precision": 0.9074074074074074,
"negative_predictive_value": 0.9078947368421053
}
},
{
"epoch": 4,
"train_loss": 0.3866710865129066,
"validation": {
"support": 132,
"positive_support": 57,
"precision": 0.8983050847457628,
"recall": 0.9298245614035088,
"f1": 0.9137931034482759,
"macro_f1": 0.9231127679403541,
"average_precision": 0.9448769491431617,
"confusion_matrix": [
[
69,
6
],
[
4,
53
]
],
"coverage": 0.9924242424242424,
"review_rate": 0.007575757575757576,
"review_count": 1,
"redundant_count": 56,
"not_redundant_count": 75,
"selective_precision": 0.9107142857142857,
"negative_predictive_value": 0.92
}
}
],
"selected_epoch": 4,
"selected_parameter_delta_l2": 0.03376656398177147,
"optimizer_steps": 188,
"pruned": false,
"policy": {
"binary_threshold": 0.3526492714881897,
"thresholds": {
"low": 0.46919289231300354,
"high": 0.4728550910949707,
"positive_enabled": true,
"negative_enabled": true
},
"selection_split": "validation",
"precision_target": 0.9,
"negative_predictive_value_target": 0.9,
"minimum_support": 5,
"decision_policy_version": "0.1.0"
},
"training_pairs": 744,
"validation_pairs": 132,
"trainable_parameters": 43514883,
"total_parameters": 278811651,
"duration_seconds": 1554.2861375080029,
"environment": {
"python": "3.11.14",
"platform": "Linux-7.1.8+deb13-amd64-x86_64-with-glibc2.41",
"code_commit": "6f54813980df0f1ab8056d7fa7e7e33ff66b8871",
"source_sha256": {
"cli.py": "8462dd1c90e06165826a5053c740c2e4379bc4db1a3fafacb38eccc866f007ab",
"decision.py": "ce69d18701fb687f0fd94217a121f7ebdd29e9397c0c7a18c899dcede96f5e2a",
"calibration.py": "d999744c3ed24167540740e010d98deb96ab23e01570f2de24d2e12d8758a044",
"pipeline.py": "1ad7ce1337cf4e8b74549cbc49088f3456d07ec85c905783e86992611d1c4a29",
"encoder.py": "8e5392f0233e9824dee1e7d5f4e7f542be5921cf4193f0ff84660d19711c96e4",
"nli.py": "f74b06c83c66bb6d6f500b01ee9831f56422e8aad44e2355bff48ab1c1b27558",
"pseudo.py": "b974fd870e6bf614131d097fa7f49aba77696d3b29f164a3ea1c094d7383ef74",
"data.py": "5b8143e617f9166f13159f96081b9048b24fe6d80289b181fc165f1c307b7dde",
"experiment.py": "d6cac58f617129f20a3518724d53678d6ef204fee35a9e998d1dc1aafd8af891",
"score_graph.py": "252d95a1ba60fe26cfab1e9e57eb47b51ddf04d7b706ea52ced1f5376bd0f9ed",
"domain_adaptation.py": "cab4001a9da57af5aad758d43ffcb1cb76c5cc6c97eac2c7a6c92f45a076ed8f",
"training.py": "2a375b311b3a9e3e5d773e0b0f974ff859dd4f232d639c8176302128142aa8c0",
"__main__.py": "b466d3224c1cfa96b82f891d496e36e7bd4ad1eb70fcc530b45c29ce62d3f9c4",
"cross_encoder.py": "2fd31598b2ae3e039ffcbab908c1c6b190aa6aba74d117f5658fc2fde89b5437",
"schemas.py": "361128ca989504938a2bf8977cbf0b1be49bee9296c5da4903b05736e149535c",
"evaluation.py": "4880d0ee9fcf6fc15163204b2edcf4856dc06dfe5aaa3930a9dca236b16d0fd0",
"__init__.py": "8f3ef65a08808435e85992bcb80379e0a060dab973ffbe2b5e193bc924766681",
"v03_data.py": "187402e98c0c9b7a5703aaeeff244b60f76410a24a889dddbe8d73e5fdb14767",
"dataset.py": "40c6153527279d43d66d60de10dc0cb1bfbcba07df11a64d18ad810d56bfb149"
},
"deterministic_warn_only": true,
"cublas_workspace_config": ":4096:8",
"gpu": "NVIDIA RTX A1000 6GB Laptop GPU",
"packages": {
"torch": "2.6.0",
"sentence-transformers": "5.1.2",
"transformers": "4.57.1",
"numpy": "2.3.4",
"scikit-learn": "1.7.2"
}
},
"objective": "multitask directional NLI CE and symmetric redundancy BCE",
"score": "minimum bidirectional entailment softmax; uncalibrated"
}