training_cfg.json

{
  "model_name": "microsoft/deberta-v3-large",
  "lr": 6e-06,
  "epochs": 1,
  "weight_decay": 0.01,
  "train_bs": 3,
  "eval_bs": 8,
  "gradient_accumulation_steps": 8,
  "gradient_checkpointing": false,
  "eval_datasets": [
    {"path": "./data/all_eval/osu_eval.csv", "name": "osu"},
    {"path": "./data/all_eval/healthver_eval.csv", "name": "healthver"},
    {"path": "./data/all_eval/deepset_1_eval.csv", "name": "deepset_1"},
    {"path": "./data/all_eval/deepset_2_eval.csv", "name": "deepset_2"},
    {"path": "./data/all_eval/deepset_3_eval.csv", "name": "deepset_3"},
    {"path": "./data/all_eval/deepset_4_eval.csv", "name": "deepset_4"}
  ],
  "train_dataset_path": "./data/all_training/all_train_v2_pseudo.csv",
  "ensemble_model_predictions": [
        "mathislucka/deberta-large-hallucination-eval-v2",
        "mathislucka/deberta-base-hallucination-eval-v2",
        "models/albert-xxlarge-v2-optim-data-v1",
        "models/deberta-base-v3-no-atomic-wfc-nq",
        "models/deberta-v3-large-data-optim-v2"
  ],
  "half_precision": true
}