config.json
| 1 | { |
| 2 | "architectures": [ |
| 3 | "TabStarModel" |
| 4 | ], |
| 5 | "batch_size": 32, |
| 6 | "bert_lr_factor": 1.0, |
| 7 | "d_model": 384, |
| 8 | "downstream_mode": null, |
| 9 | "e5_model": "intfloat/e5-small-v2", |
| 10 | "is_pretrain": true, |
| 11 | "lr": 5e-05, |
| 12 | "macro_batch_size": 128, |
| 13 | "model_type": "tabstar", |
| 14 | "num_layers": 6, |
| 15 | "numbers_fusion": "attention", |
| 16 | "p_mask": 0, |
| 17 | "scheduler": "cycle", |
| 18 | "tabular_encoder_type": "d1", |
| 19 | "torch_dtype": "float32", |
| 20 | "transformers_version": "4.49.0", |
| 21 | "unfreeze_layers": 6, |
| 22 | "weight_balance": "uniform", |
| 23 | "weight_decay": 0.001 |
| 24 | } |
| 25 | |