| { | |
| "lr_tfidf": { | |
| "model": "LR + TF-IDF (Baseline)", | |
| "artifact": "models/baseline/lr_tfidf.joblib", | |
| "source": "Optuna-tuned sklearn pipeline (Notebook 06 → final_model.joblib)", | |
| "f1_weighted_test": 0.7579, | |
| "train_test_gap_pp": 4.76, | |
| "roc_auc_test": 0.81, | |
| "role": "Esencial sklearn baseline — fast CPU inference" | |
| }, | |
| "frozen_toxic_bert": { | |
| "model": "Frozen Toxic-BERT (Baseline)", | |
| "hub_model_id": "unitary/toxic-bert", | |
| "freeze_mode": "inference_only", | |
| "evaluation_source": "reports/golden_baseline/golden_baseline_run_20260524_213342.json", | |
| "f1_weighted_test": 0.7903, | |
| "train_test_gap_pp": 0.16, | |
| "threshold": 0.12, | |
| "roc_auc_test": 0.8759, | |
| "role": "Transformer baseline — all layers frozen, no fine-tuning on 1k rows" | |
| } | |
| } | |