| resume_from_checkpoint_path: None | |
| result_path: 'result' | |
| model_path: None | |
| dataset_paths: | |
| - '/root/autodl-tmp/Nougat-Data/train.jsonl' | |
| tokenizer: './config/tokenizer.json' | |
| exp_name: 'nougat-middle' | |
| train_batch_sizes: | |
| - 1 | |
| num_workers: 4 | |
| val_batch_sizes: | |
| - 1 | |
| val_batches: 1 | |
| input_size: | |
| - 896 | |
| - 672 | |
| max_length: 4096 | |
| max_position_embeddings: 4096 | |
| accumulate_grad_batches: 3 | |
| window_size: 7 | |
| patch_size: 4 | |
| embed_dim: 128 | |
| hidden_dimension: 1024 | |
| num_heads: | |
| - 4 | |
| - 8 | |
| - 16 | |
| - 32 | |
| encoder_layer: | |
| - 2 | |
| - 2 | |
| - 14 | |
| - 2 | |
| decoder_layer: 6 | |
| align_long_axis: False | |
| num_nodes: 1 | |
| seed: 25 | |
| lr: 5e-05 | |
| min_lr: 7.5e-06 | |
| lr_step: 16 | |
| gamma: 0.9996 | |
| warmup_steps: 250 | |
| num_training_samples_per_epoch: 10000 | |
| max_epochs: 30 | |
| max_steps: -1 | |
| val_check_interval: None | |
| check_val_every_n_epoch: 1 | |
| gradient_clip_val: 0.5 | |
| verbose: False | |
| debug: False | |
| job: None | |
| exp_version: '20240206_132202' |