{ "loader": "loaders/newsroom.py", "dataset": "data/train-data/newsroom", "indices": "data/train-data/newsroom/indices.npy", "model_dir": "data/models/newsroom-CR75", "verbose": true, "print_every": 1, "eval_every": 50, "save_every": 50, "max_val_steps": 512, "max_train_seconds": null, "max_train_steps": 8000, "batch_size": 4, "learning_rate": 1e-05, "k_samples": 100, "sample_aggregation": "max", "loss": "pgb", "encoder_model_id": "distilroberta-base", "rewards": { "Fluency": { "weight": 1, "type": "masked", "model_id": "distilroberta-base", "max_score": 40.0, "norm": "max" }, "BiEncoderSimilarity": { "weight": 1, "model_id": "all-distilroberta-v1" }, "GaussianCR": { "weight": 1, "mean": 0.75, "std": 0.3 } } }